diff --git "a/Encoder.mlmodelc/model.mil" "b/Encoder.mlmodelc/model.mil"
--- "a/Encoder.mlmodelc/model.mil"
+++ "b/Encoder.mlmodelc/model.mil"
@@ -1,7 +1,7 @@
 program(1.3)
 [buildInfo = dict<string, string>({{"coremlc-component-MIL", "3600.16.1"}, {"coremlc-version", "3600.25.2"}})]
 {
-    func main<ios18>(tensor<fp32, [1, 128, 1501]> mel, tensor<int32, [1]> mel_length) {
+    func enc_10160ms<ios18>(tensor<fp32, [1, 128, 1017]> mel, tensor<int32, [1]> mel_length) {
             int32 var_30 = const()[name = string("op_30"), val = int32(-1)];
             tensor<int32, [3]> x_1_perm_0 = const()[name = string("x_1_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string mel_to_fp16_dtype_0 = const()[name = string("mel_to_fp16_dtype_0"), val = string("fp16")];
@@ -29,9 +29,9 @@ program(1.3)
             tensor<fp16, [1]> lengths_13_cast_fp16 = add(x = _inversed_106_cast_fp16, y = var_107_to_fp16)[name = string("lengths_13_cast_fp16")];
             tensor<fp16, [1]> lengths_cast_fp16 = floor(x = lengths_13_cast_fp16)[name = string("lengths_cast_fp16")];
             tensor<int32, [1]> input_1_axes_0 = const()[name = string("input_1_axes_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1, 128, 1501]> mel_to_fp16 = cast(dtype = mel_to_fp16_dtype_0, x = mel)[name = string("cast_2")];
-            tensor<fp16, [1, 1501, 128]> x_1_cast_fp16 = transpose(perm = x_1_perm_0, x = mel_to_fp16)[name = string("transpose_315")];
-            tensor<fp16, [1, 1, 1501, 128]> input_1_cast_fp16 = expand_dims(axes = input_1_axes_0, x = x_1_cast_fp16)[name = string("input_1_cast_fp16")];
+            tensor<fp16, [1, 128, 1017]> mel_to_fp16 = cast(dtype = mel_to_fp16_dtype_0, x = mel)[name = string("cast_2")];
+            tensor<fp16, [1, 1017, 128]> x_1_cast_fp16 = transpose(perm = x_1_perm_0, x = mel_to_fp16)[name = string("transpose_315")];
+            tensor<fp16, [1, 1, 1017, 128]> input_1_cast_fp16 = expand_dims(axes = input_1_axes_0, x = x_1_cast_fp16)[name = string("input_1_cast_fp16")];
             string input_3_pad_type_0 = const()[name = string("input_3_pad_type_0"), val = string("custom")];
             tensor<int32, [4]> input_3_pad_0 = const()[name = string("input_3_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
             tensor<int32, [2]> input_3_strides_0 = const()[name = string("input_3_strides_0"), val = tensor<int32, [2]>([2, 2])];
@@ -39,8 +39,8 @@ program(1.3)
             int32 input_3_groups_0 = const()[name = string("input_3_groups_0"), val = int32(1)];
             tensor<fp16, [256, 1, 3, 3]> module_pre_encode_conv_0_weight_to_fp16 = const()[name = string("module_pre_encode_conv_0_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
             tensor<fp16, [256]> module_pre_encode_conv_0_bias_to_fp16 = const()[name = string("module_pre_encode_conv_0_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4736)))];
-            tensor<fp16, [1, 256, 751, 64]> input_3_cast_fp16 = conv(bias = module_pre_encode_conv_0_bias_to_fp16, dilations = input_3_dilations_0, groups = input_3_groups_0, pad = input_3_pad_0, pad_type = input_3_pad_type_0, strides = input_3_strides_0, weight = module_pre_encode_conv_0_weight_to_fp16, x = input_1_cast_fp16)[name = string("input_3_cast_fp16")];
-            tensor<fp16, [1, 256, 751, 64]> input_5_cast_fp16 = relu(x = input_3_cast_fp16)[name = string("input_5_cast_fp16")];
+            tensor<fp16, [1, 256, 509, 64]> input_3_cast_fp16 = conv(bias = module_pre_encode_conv_0_bias_to_fp16, dilations = input_3_dilations_0, groups = input_3_groups_0, pad = input_3_pad_0, pad_type = input_3_pad_type_0, strides = input_3_strides_0, weight = module_pre_encode_conv_0_weight_to_fp16, x = input_1_cast_fp16)[name = string("input_3_cast_fp16")];
+            tensor<fp16, [1, 256, 509, 64]> input_5_cast_fp16 = relu(x = input_3_cast_fp16)[name = string("input_5_cast_fp16")];
             string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("custom")];
             tensor<int32, [4]> input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
             tensor<int32, [2]> input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor<int32, [2]>([2, 2])];
@@ -48,7 +48,7 @@ program(1.3)
             tensor<int32, [2]> input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
             tensor<fp16, [256, 1, 3, 3]> module_pre_encode_conv_2_weight_to_fp16 = const()[name = string("module_pre_encode_conv_2_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5312)))];
             tensor<fp16, [256]> module_pre_encode_conv_2_bias_to_fp16 = const()[name = string("module_pre_encode_conv_2_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9984)))];
-            tensor<fp16, [1, 256, 376, 32]> input_7_cast_fp16 = conv(bias = module_pre_encode_conv_2_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = module_pre_encode_conv_2_weight_to_fp16, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")];
+            tensor<fp16, [1, 256, 255, 32]> input_7_cast_fp16 = conv(bias = module_pre_encode_conv_2_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = module_pre_encode_conv_2_weight_to_fp16, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")];
             string input_9_pad_type_0 = const()[name = string("input_9_pad_type_0"), val = string("valid")];
             tensor<int32, [2]> input_9_strides_0 = const()[name = string("input_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
             tensor<int32, [4]> input_9_pad_0 = const()[name = string("input_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
@@ -56,8 +56,8 @@ program(1.3)
             int32 input_9_groups_0 = const()[name = string("input_9_groups_0"), val = int32(1)];
             tensor<fp16, [256, 256, 1, 1]> module_pre_encode_conv_3_weight_to_fp16 = const()[name = string("module_pre_encode_conv_3_weight_to_fp16"), val = tensor<fp16, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10560)))];
             tensor<fp16, [256]> module_pre_encode_conv_3_bias_to_fp16 = const()[name = string("module_pre_encode_conv_3_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141696)))];
-            tensor<fp16, [1, 256, 376, 32]> input_9_cast_fp16 = conv(bias = module_pre_encode_conv_3_bias_to_fp16, dilations = input_9_dilations_0, groups = input_9_groups_0, pad = input_9_pad_0, pad_type = input_9_pad_type_0, strides = input_9_strides_0, weight = module_pre_encode_conv_3_weight_to_fp16, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")];
-            tensor<fp16, [1, 256, 376, 32]> input_11_cast_fp16 = relu(x = input_9_cast_fp16)[name = string("input_11_cast_fp16")];
+            tensor<fp16, [1, 256, 255, 32]> input_9_cast_fp16 = conv(bias = module_pre_encode_conv_3_bias_to_fp16, dilations = input_9_dilations_0, groups = input_9_groups_0, pad = input_9_pad_0, pad_type = input_9_pad_type_0, strides = input_9_strides_0, weight = module_pre_encode_conv_3_weight_to_fp16, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")];
+            tensor<fp16, [1, 256, 255, 32]> input_11_cast_fp16 = relu(x = input_9_cast_fp16)[name = string("input_11_cast_fp16")];
             string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("custom")];
             tensor<int32, [4]> input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
             tensor<int32, [2]> input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor<int32, [2]>([2, 2])];
@@ -65,7 +65,7 @@ program(1.3)
             tensor<int32, [2]> input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
             tensor<fp16, [256, 1, 3, 3]> module_pre_encode_conv_5_weight_to_fp16 = const()[name = string("module_pre_encode_conv_5_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142272)))];
             tensor<fp16, [256]> module_pre_encode_conv_5_bias_to_fp16 = const()[name = string("module_pre_encode_conv_5_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146944)))];
-            tensor<fp16, [1, 256, 188, 16]> input_13_cast_fp16 = conv(bias = module_pre_encode_conv_5_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = module_pre_encode_conv_5_weight_to_fp16, x = input_11_cast_fp16)[name = string("input_13_cast_fp16")];
+            tensor<fp16, [1, 256, 128, 16]> input_13_cast_fp16 = conv(bias = module_pre_encode_conv_5_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = module_pre_encode_conv_5_weight_to_fp16, x = input_11_cast_fp16)[name = string("input_13_cast_fp16")];
             string input_15_pad_type_0 = const()[name = string("input_15_pad_type_0"), val = string("valid")];
             tensor<int32, [2]> input_15_strides_0 = const()[name = string("input_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
             tensor<int32, [4]> input_15_pad_0 = const()[name = string("input_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
@@ -73,3572 +73,10852 @@ program(1.3)
             int32 input_15_groups_0 = const()[name = string("input_15_groups_0"), val = int32(1)];
             tensor<fp16, [256, 256, 1, 1]> module_pre_encode_conv_6_weight_to_fp16 = const()[name = string("module_pre_encode_conv_6_weight_to_fp16"), val = tensor<fp16, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147520)))];
             tensor<fp16, [256]> module_pre_encode_conv_6_bias_to_fp16 = const()[name = string("module_pre_encode_conv_6_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278656)))];
-            tensor<fp16, [1, 256, 188, 16]> input_15_cast_fp16 = conv(bias = module_pre_encode_conv_6_bias_to_fp16, dilations = input_15_dilations_0, groups = input_15_groups_0, pad = input_15_pad_0, pad_type = input_15_pad_type_0, strides = input_15_strides_0, weight = module_pre_encode_conv_6_weight_to_fp16, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")];
-            tensor<fp16, [1, 256, 188, 16]> x_3_cast_fp16 = relu(x = input_15_cast_fp16)[name = string("x_3_cast_fp16")];
+            tensor<fp16, [1, 256, 128, 16]> input_15_cast_fp16 = conv(bias = module_pre_encode_conv_6_bias_to_fp16, dilations = input_15_dilations_0, groups = input_15_groups_0, pad = input_15_pad_0, pad_type = input_15_pad_type_0, strides = input_15_strides_0, weight = module_pre_encode_conv_6_weight_to_fp16, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")];
+            tensor<fp16, [1, 256, 128, 16]> x_3_cast_fp16 = relu(x = input_15_cast_fp16)[name = string("x_3_cast_fp16")];
             tensor<int32, [4]> var_157_perm_0 = const()[name = string("op_157_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
-            tensor<int32, [3]> var_158 = const()[name = string("op_158"), val = tensor<int32, [3]>([1, 188, -1])];
-            tensor<fp16, [1, 188, 256, 16]> var_157_cast_fp16 = transpose(perm = var_157_perm_0, x = x_3_cast_fp16)[name = string("transpose_314")];
-            tensor<fp16, [1, 188, 4096]> input_17_cast_fp16 = reshape(shape = var_158, x = var_157_cast_fp16)[name = string("input_17_cast_fp16")];
+            tensor<int32, [3]> var_158 = const()[name = string("op_158"), val = tensor<int32, [3]>([1, 128, -1])];
+            tensor<fp16, [1, 128, 256, 16]> var_157_cast_fp16 = transpose(perm = var_157_perm_0, x = x_3_cast_fp16)[name = string("transpose_314")];
+            tensor<fp16, [1, 128, 4096]> input_17_cast_fp16 = reshape(shape = var_158, x = var_157_cast_fp16)[name = string("input_17_cast_fp16")];
             tensor<fp16, [1024, 4096]> module_pre_encode_out_weight_to_fp16 = const()[name = string("module_pre_encode_out_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279232)))];
             tensor<fp16, [1024]> module_pre_encode_out_bias_to_fp16 = const()[name = string("module_pre_encode_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8667904)))];
-            tensor<fp16, [1, 188, 1024]> linear_0_cast_fp16 = linear(bias = module_pre_encode_out_bias_to_fp16, weight = module_pre_encode_out_weight_to_fp16, x = input_17_cast_fp16)[name = string("linear_0_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_0_cast_fp16 = linear(bias = module_pre_encode_out_bias_to_fp16, weight = module_pre_encode_out_weight_to_fp16, x = input_17_cast_fp16)[name = string("linear_0_cast_fp16")];
             string padding_length_dtype_0 = const()[name = string("padding_length_dtype_0"), val = string("int32")];
-            tensor<int32, [1, 188]> expand_dims_0 = const()[name = string("expand_dims_0"), val = tensor<int32, [1, 188]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8670016)))];
+            tensor<int32, [1, 128]> expand_dims_0 = const()[name = string("expand_dims_0"), val = tensor<int32, [1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307421248)))];
             tensor<int32, [1]> var_196_axes_0 = const()[name = string("op_196_axes_0"), val = tensor<int32, [1]>([-1])];
             tensor<int32, [1]> encoder_length = cast(dtype = padding_length_dtype_0, x = lengths_cast_fp16)[name = string("cast_1")];
             tensor<int32, [1, 1]> var_196 = expand_dims(axes = var_196_axes_0, x = encoder_length)[name = string("op_196")];
-            tensor<bool, [1, 188]> pad_mask_1 = less(x = expand_dims_0, y = var_196)[name = string("pad_mask_1")];
+            tensor<bool, [1, 128]> pad_mask_1 = less(x = expand_dims_0, y = var_196)[name = string("pad_mask_1")];
             tensor<int32, [1]> var_198_axes_0 = const()[name = string("op_198_axes_0"), val = tensor<int32, [1]>([1])];
-            tensor<bool, [1, 1, 188]> var_198 = expand_dims(axes = var_198_axes_0, x = pad_mask_1)[name = string("op_198")];
-            tensor<int32, [3]> var_199 = const()[name = string("op_199"), val = tensor<int32, [3]>([1, 188, 1])];
-            tensor<bool, [1, 188, 188]> pad_mask_for_att_mask_1 = tile(reps = var_199, x = var_198)[name = string("pad_mask_for_att_mask_1")];
+            tensor<bool, [1, 1, 128]> var_198 = expand_dims(axes = var_198_axes_0, x = pad_mask_1)[name = string("op_198")];
+            tensor<int32, [3]> var_199 = const()[name = string("op_199"), val = tensor<int32, [3]>([1, 128, 1])];
+            tensor<bool, [1, 128, 128]> pad_mask_for_att_mask_1 = tile(reps = var_199, x = var_198)[name = string("pad_mask_for_att_mask_1")];
             tensor<int32, [3]> var_201_perm_0 = const()[name = string("op_201_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<bool, [1, 188, 188]> var_201 = transpose(perm = var_201_perm_0, x = pad_mask_for_att_mask_1)[name = string("transpose_313")];
-            tensor<bool, [1, 188, 188]> pad_mask_for_att_mask = logical_and(x = pad_mask_for_att_mask_1, y = var_201)[name = string("pad_mask_for_att_mask")];
-            tensor<bool, [1, 188, 188]> const_7 = const()[name = string("const_7"), val = tensor<bool, [1, 188, 188]>([[[true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true]]])];
-            tensor<bool, [1, 188, 188]> att_mask = logical_and(x = pad_mask_for_att_mask, y = const_7)[name = string("att_mask")];
-            tensor<bool, [1, 188, 188]> mask_1 = logical_not(x = att_mask)[name = string("mask_1")];
-            tensor<bool, [1, 188]> pad_mask = logical_not(x = pad_mask_1)[name = string("pad_mask")];
+            tensor<bool, [1, 128, 128]> var_201 = transpose(perm = var_201_perm_0, x = pad_mask_for_att_mask_1)[name = string("transpose_313")];
+            tensor<bool, [1, 128, 128]> pad_mask_for_att_mask = logical_and(x = pad_mask_for_att_mask_1, y = var_201)[name = string("pad_mask_for_att_mask")];
+            tensor<bool, [1, 128, 128]> const_7 = const()[name = string("const_7"), val = tensor<bool, [1, 128, 128]>([[[true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true]]])];
+            tensor<bool, [1, 128, 128]> att_mask = logical_and(x = pad_mask_for_att_mask, y = const_7)[name = string("att_mask")];
+            tensor<bool, [1, 128, 128]> mask_1 = logical_not(x = att_mask)[name = string("mask_1")];
+            tensor<bool, [1, 128]> pad_mask = logical_not(x = pad_mask_1)[name = string("pad_mask")];
             tensor<int32, [1]> input_21_axes_0 = const()[name = string("input_21_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_0_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_0_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8670848)))];
-            tensor<fp16, [1024]> module_layers_0_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_0_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8672960)))];
+            tensor<fp16, [1024]> module_layers_0_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_0_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8670336)))];
+            tensor<fp16, [1024]> module_layers_0_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_0_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8672448)))];
             fp16 var_9_to_fp16 = const()[name = string("op_9_to_fp16"), val = fp16(0x1.5p-17)];
-            tensor<fp16, [1, 188, 1024]> input_21_cast_fp16 = layer_norm(axes = input_21_axes_0, beta = module_layers_0_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_feed_forward1_weight_to_fp16, x = linear_0_cast_fp16)[name = string("input_21_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2072390]> module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8675072))), indices_nonzero_data = tensor<uint6, [2072390]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9199424))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10753792))))[name = string("module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 1024]> input_21_cast_fp16 = layer_norm(axes = input_21_axes_0, beta = module_layers_0_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_feed_forward1_weight_to_fp16, x = linear_0_cast_fp16)[name = string("input_21_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2072390]> module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8674560))), indices_nonzero_data = tensor<uint6, [2072390]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9198912))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10753280))))[name = string("module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [4096]> linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10819392)))];
-            tensor<fp16, [1, 188, 4096]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_21_cast_fp16)[name = string("linear_1_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_25_cast_fp16 = silu(x = linear_1_cast_fp16)[name = string("input_25_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2038634]> module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10827648))), indices_nonzero_data = tensor<uint6, [2038634]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11352000))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12881088))))[name = string("module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096]> linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10818880)))];
+            tensor<fp16, [1, 128, 4096]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_21_cast_fp16)[name = string("linear_1_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_25_cast_fp16 = silu(x = linear_1_cast_fp16)[name = string("input_25_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2038634]> module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10827136))), indices_nonzero_data = tensor<uint6, [2038634]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11351488))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12880576))))[name = string("module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1024]> linear_2_bias_0_to_fp16 = const()[name = string("linear_2_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12897536)))];
-            tensor<fp16, [1, 188, 1024]> linear_2_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_25_cast_fp16)[name = string("linear_2_cast_fp16")];
+            tensor<fp16, [1024]> linear_2_bias_0_to_fp16 = const()[name = string("linear_2_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12897024)))];
+            tensor<fp16, [1, 128, 1024]> linear_2_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_25_cast_fp16)[name = string("linear_2_cast_fp16")];
             fp16 var_232_to_fp16 = const()[name = string("op_232_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_233_cast_fp16 = mul(x = linear_2_cast_fp16, y = var_232_to_fp16)[name = string("op_233_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_31_cast_fp16 = add(x = linear_0_cast_fp16, y = var_233_cast_fp16)[name = string("input_31_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_233_cast_fp16 = mul(x = linear_2_cast_fp16, y = var_232_to_fp16)[name = string("op_233_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_31_cast_fp16 = add(x = linear_0_cast_fp16, y = var_233_cast_fp16)[name = string("input_31_cast_fp16")];
             tensor<int32, [1]> query_1_axes_0 = const()[name = string("query_1_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_0_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_0_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12899648)))];
-            tensor<fp16, [1024]> module_layers_0_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_0_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12901760)))];
-            tensor<fp16, [1, 188, 1024]> query_1_cast_fp16 = layer_norm(axes = query_1_axes_0, beta = module_layers_0_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_self_att_weight_to_fp16, x = input_31_cast_fp16)[name = string("query_1_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522455]> module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12903872))), indices_nonzero_data = tensor<uint6, [522455]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13035008))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13426944))))[name = string("module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_0_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_0_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12899136)))];
+            tensor<fp16, [1024]> module_layers_0_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_0_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12901248)))];
+            tensor<fp16, [1, 128, 1024]> query_1_cast_fp16 = layer_norm(axes = query_1_axes_0, beta = module_layers_0_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_self_att_weight_to_fp16, x = input_31_cast_fp16)[name = string("query_1_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522455]> module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12903360))), indices_nonzero_data = tensor<uint6, [522455]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13034496))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13426432))))[name = string("module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_3_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_1_cast_fp16)[name = string("linear_3_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_3_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_1_cast_fp16)[name = string("linear_3_cast_fp16")];
             tensor<int32, [4]> var_249 = const()[name = string("op_249"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_1_cast_fp16 = reshape(shape = var_249, x = linear_3_cast_fp16)[name = string("q_1_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [515024]> module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13443392))), indices_nonzero_data = tensor<uint6, [515024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13574528))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13960896))))[name = string("module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> q_1_cast_fp16 = reshape(shape = var_249, x = linear_3_cast_fp16)[name = string("q_1_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [515024]> module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13442880))), indices_nonzero_data = tensor<uint6, [515024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13574016))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13960384))))[name = string("module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_4_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_1_cast_fp16)[name = string("linear_4_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_4_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_1_cast_fp16)[name = string("linear_4_cast_fp16")];
             tensor<int32, [4]> var_253 = const()[name = string("op_253"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_1_cast_fp16 = reshape(shape = var_253, x = linear_4_cast_fp16)[name = string("k_1_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [505889]> module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13977344))), indices_nonzero_data = tensor<uint6, [505889]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14108480))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14488000))))[name = string("module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> k_1_cast_fp16 = reshape(shape = var_253, x = linear_4_cast_fp16)[name = string("k_1_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [505889]> module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13976832))), indices_nonzero_data = tensor<uint6, [505889]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14107968))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14487488))))[name = string("module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_5_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_1_cast_fp16)[name = string("linear_5_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_5_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_1_cast_fp16)[name = string("linear_5_cast_fp16")];
             tensor<int32, [4]> var_257 = const()[name = string("op_257"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_1_cast_fp16 = reshape(shape = var_257, x = linear_5_cast_fp16)[name = string("v_1_cast_fp16")];
+            tensor<fp16, [1, 128, 8, 128]> v_1_cast_fp16 = reshape(shape = var_257, x = linear_5_cast_fp16)[name = string("v_1_cast_fp16")];
             tensor<int32, [4]> value_3_perm_0 = const()[name = string("value_3_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_0_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_0_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14504448)))];
-            tensor<fp16, [1, 188, 8, 128]> var_269_cast_fp16 = add(x = q_1_cast_fp16, y = module_layers_0_self_attn_pos_bias_u_to_fp16)[name = string("op_269_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_0_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_0_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14506560)))];
-            tensor<fp16, [1, 188, 8, 128]> var_271_cast_fp16 = add(x = q_1_cast_fp16, y = module_layers_0_self_attn_pos_bias_v_to_fp16)[name = string("op_271_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_0_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_0_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14503936)))];
+            tensor<fp16, [1, 128, 8, 128]> var_269_cast_fp16 = add(x = q_1_cast_fp16, y = module_layers_0_self_attn_pos_bias_u_to_fp16)[name = string("op_269_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_0_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_0_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14506048)))];
+            tensor<fp16, [1, 128, 8, 128]> var_271_cast_fp16 = add(x = q_1_cast_fp16, y = module_layers_0_self_attn_pos_bias_v_to_fp16)[name = string("op_271_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_1_perm_0 = const()[name = string("q_with_bias_v_1_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_7_transpose_x_0 = const()[name = string("x_7_transpose_x_0"), val = bool(false)];
             bool x_7_transpose_y_0 = const()[name = string("x_7_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_273_to_fp16 = const()[name = string("op_273_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14508672)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_1_cast_fp16 = transpose(perm = q_with_bias_v_1_perm_0, x = var_271_cast_fp16)[name = string("transpose_312")];
-            tensor<fp16, [1, 8, 188, 375]> x_7_cast_fp16 = matmul(transpose_x = x_7_transpose_x_0, transpose_y = x_7_transpose_y_0, x = q_with_bias_v_1_cast_fp16, y = var_273_to_fp16)[name = string("x_7_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 255]> var_273_to_fp16 = const()[name = string("op_273_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307421824)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_1_cast_fp16 = transpose(perm = q_with_bias_v_1_perm_0, x = var_271_cast_fp16)[name = string("transpose_312")];
+            tensor<fp16, [1, 8, 128, 255]> x_7_cast_fp16 = matmul(transpose_x = x_7_transpose_x_0, transpose_y = x_7_transpose_y_0, x = q_with_bias_v_1_cast_fp16, y = var_273_to_fp16)[name = string("x_7_cast_fp16")];
             tensor<int32, [8]> x_9_pad_0 = const()[name = string("x_9_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_9_mode_0 = const()[name = string("x_9_mode_0"), val = string("constant")];
             fp16 const_14_to_fp16 = const()[name = string("const_14_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_9_cast_fp16 = pad(constant_val = const_14_to_fp16, mode = x_9_mode_0, pad = x_9_pad_0, x = x_7_cast_fp16)[name = string("x_9_cast_fp16")];
-            tensor<int32, [4]> var_281 = const()[name = string("op_281"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_11_cast_fp16 = reshape(shape = var_281, x = x_9_cast_fp16)[name = string("x_11_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 256]> x_9_cast_fp16 = pad(constant_val = const_14_to_fp16, mode = x_9_mode_0, pad = x_9_pad_0, x = x_7_cast_fp16)[name = string("x_9_cast_fp16")];
+            tensor<int32, [4]> var_281 = const()[name = string("op_281"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_11_cast_fp16 = reshape(shape = var_281, x = x_9_cast_fp16)[name = string("x_11_cast_fp16")];
             tensor<int32, [4]> var_285_begin_0 = const()[name = string("op_285_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_285_end_0 = const()[name = string("op_285_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_285_end_0 = const()[name = string("op_285_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
             tensor<bool, [4]> var_285_end_mask_0 = const()[name = string("op_285_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_285_cast_fp16 = slice_by_index(begin = var_285_begin_0, end = var_285_end_0, end_mask = var_285_end_mask_0, x = x_11_cast_fp16)[name = string("op_285_cast_fp16")];
-            tensor<int32, [4]> var_286 = const()[name = string("op_286"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_1_cast_fp16 = reshape(shape = var_286, x = var_285_cast_fp16)[name = string("matrix_bd_1_cast_fp16")];
+            tensor<fp16, [1, 8, 255, 128]> var_285_cast_fp16 = slice_by_index(begin = var_285_begin_0, end = var_285_end_0, end_mask = var_285_end_mask_0, x = x_11_cast_fp16)[name = string("op_285_cast_fp16")];
+            tensor<int32, [4]> var_286 = const()[name = string("op_286"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_1_cast_fp16 = reshape(shape = var_286, x = var_285_cast_fp16)[name = string("matrix_bd_1_cast_fp16")];
             bool matrix_ac_1_transpose_x_0 = const()[name = string("matrix_ac_1_transpose_x_0"), val = bool(false)];
             bool matrix_ac_1_transpose_y_0 = const()[name = string("matrix_ac_1_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_96_perm_0 = const()[name = string("transpose_96_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_97_perm_0 = const()[name = string("transpose_97_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_97 = transpose(perm = transpose_97_perm_0, x = k_1_cast_fp16)[name = string("transpose_310")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_96 = transpose(perm = transpose_96_perm_0, x = var_269_cast_fp16)[name = string("transpose_311")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_1_cast_fp16 = matmul(transpose_x = matrix_ac_1_transpose_x_0, transpose_y = matrix_ac_1_transpose_y_0, x = transpose_96, y = transpose_97)[name = string("matrix_ac_1_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_97 = transpose(perm = transpose_97_perm_0, x = k_1_cast_fp16)[name = string("transpose_310")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_96 = transpose(perm = transpose_96_perm_0, x = var_269_cast_fp16)[name = string("transpose_311")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_1_cast_fp16 = matmul(transpose_x = matrix_ac_1_transpose_x_0, transpose_y = matrix_ac_1_transpose_y_0, x = transpose_96, y = transpose_97)[name = string("matrix_ac_1_cast_fp16")];
             tensor<int32, [4]> matrix_bd_3_begin_0 = const()[name = string("matrix_bd_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_3_end_0 = const()[name = string("matrix_bd_3_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_3_end_0 = const()[name = string("matrix_bd_3_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
             tensor<bool, [4]> matrix_bd_3_end_mask_0 = const()[name = string("matrix_bd_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_3_cast_fp16 = slice_by_index(begin = matrix_bd_3_begin_0, end = matrix_bd_3_end_0, end_mask = matrix_bd_3_end_mask_0, x = matrix_bd_1_cast_fp16)[name = string("matrix_bd_3_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_295_cast_fp16 = add(x = matrix_ac_1_cast_fp16, y = matrix_bd_3_cast_fp16)[name = string("op_295_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_3_cast_fp16 = slice_by_index(begin = matrix_bd_3_begin_0, end = matrix_bd_3_end_0, end_mask = matrix_bd_3_end_mask_0, x = matrix_bd_1_cast_fp16)[name = string("matrix_bd_3_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_295_cast_fp16 = add(x = matrix_ac_1_cast_fp16, y = matrix_bd_3_cast_fp16)[name = string("op_295_cast_fp16")];
             fp16 _inversed_scores_1_y_0_to_fp16 = const()[name = string("_inversed_scores_1_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_1_cast_fp16 = mul(x = var_295_cast_fp16, y = _inversed_scores_1_y_0_to_fp16)[name = string("_inversed_scores_1_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_1_cast_fp16 = mul(x = var_295_cast_fp16, y = _inversed_scores_1_y_0_to_fp16)[name = string("_inversed_scores_1_cast_fp16")];
             tensor<int32, [1]> mask_3_axes_0 = const()[name = string("mask_3_axes_0"), val = tensor<int32, [1]>([1])];
-            tensor<bool, [1, 1, 188, 188]> mask_3 = expand_dims(axes = mask_3_axes_0, x = mask_1)[name = string("mask_3")];
+            tensor<bool, [1, 1, 128, 128]> mask_3 = expand_dims(axes = mask_3_axes_0, x = mask_1)[name = string("mask_3")];
             fp16 var_12_to_fp16 = const()[name = string("op_12_to_fp16"), val = fp16(-0x1.388p+13)];
-            tensor<fp16, [1, 8, 188, 188]> scores_3_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_1_cast_fp16, cond = mask_3)[name = string("scores_3_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_301_cast_fp16 = softmax(axis = var_30, x = scores_3_cast_fp16)[name = string("op_301_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_3_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_1_cast_fp16, cond = mask_3)[name = string("scores_3_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_301_cast_fp16 = softmax(axis = var_30, x = scores_3_cast_fp16)[name = string("op_301_cast_fp16")];
             fp16 var_11_to_fp16 = const()[name = string("op_11_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 188]> input_33_cast_fp16 = select(a = var_11_to_fp16, b = var_301_cast_fp16, cond = mask_3)[name = string("input_33_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_33_cast_fp16 = select(a = var_11_to_fp16, b = var_301_cast_fp16, cond = mask_3)[name = string("input_33_cast_fp16")];
             bool x_13_transpose_x_0 = const()[name = string("x_13_transpose_x_0"), val = bool(false)];
             bool x_13_transpose_y_0 = const()[name = string("x_13_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_3_cast_fp16 = transpose(perm = value_3_perm_0, x = v_1_cast_fp16)[name = string("transpose_309")];
-            tensor<fp16, [1, 8, 188, 128]> x_13_cast_fp16 = matmul(transpose_x = x_13_transpose_x_0, transpose_y = x_13_transpose_y_0, x = input_33_cast_fp16, y = value_3_cast_fp16)[name = string("x_13_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> value_3_cast_fp16 = transpose(perm = value_3_perm_0, x = v_1_cast_fp16)[name = string("transpose_309")];
+            tensor<fp16, [1, 8, 128, 128]> x_13_cast_fp16 = matmul(transpose_x = x_13_transpose_x_0, transpose_y = x_13_transpose_y_0, x = input_33_cast_fp16, y = value_3_cast_fp16)[name = string("x_13_cast_fp16")];
             tensor<int32, [4]> var_305_perm_0 = const()[name = string("op_305_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_306 = const()[name = string("op_306"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_305_cast_fp16 = transpose(perm = var_305_perm_0, x = x_13_cast_fp16)[name = string("transpose_308")];
-            tensor<fp16, [1, 188, 1024]> input_35_cast_fp16 = reshape(shape = var_306, x = var_305_cast_fp16)[name = string("input_35_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [517310]> module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15276736))), indices_nonzero_data = tensor<uint6, [517310]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15407872))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15795968))))[name = string("module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> var_305_cast_fp16 = transpose(perm = var_305_perm_0, x = x_13_cast_fp16)[name = string("transpose_308")];
+            tensor<fp16, [1, 128, 1024]> input_35_cast_fp16 = reshape(shape = var_306, x = var_305_cast_fp16)[name = string("input_35_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [517310]> module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14768320))), indices_nonzero_data = tensor<uint6, [517310]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14899456))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15287552))))[name = string("module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_7_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_35_cast_fp16)[name = string("linear_7_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_39_cast_fp16 = add(x = input_31_cast_fp16, y = linear_7_cast_fp16)[name = string("input_39_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_7_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_35_cast_fp16)[name = string("linear_7_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_39_cast_fp16 = add(x = input_31_cast_fp16, y = linear_7_cast_fp16)[name = string("input_39_cast_fp16")];
             tensor<int32, [1]> x_17_axes_0 = const()[name = string("x_17_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_0_norm_conv_weight_to_fp16 = const()[name = string("module_layers_0_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15812416)))];
-            tensor<fp16, [1024]> module_layers_0_norm_conv_bias_to_fp16 = const()[name = string("module_layers_0_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15814528)))];
-            tensor<fp16, [1, 188, 1024]> x_17_cast_fp16 = layer_norm(axes = x_17_axes_0, beta = module_layers_0_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_conv_weight_to_fp16, x = input_39_cast_fp16)[name = string("x_17_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_0_norm_conv_weight_to_fp16 = const()[name = string("module_layers_0_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15304000)))];
+            tensor<fp16, [1024]> module_layers_0_norm_conv_bias_to_fp16 = const()[name = string("module_layers_0_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15306112)))];
+            tensor<fp16, [1, 128, 1024]> x_17_cast_fp16 = layer_norm(axes = x_17_axes_0, beta = module_layers_0_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_conv_weight_to_fp16, x = input_39_cast_fp16)[name = string("x_17_cast_fp16")];
             tensor<int32, [3]> input_41_perm_0 = const()[name = string("input_41_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_43_pad_type_0 = const()[name = string("input_43_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_43_strides_0 = const()[name = string("input_43_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_43_pad_0 = const()[name = string("input_43_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_43_dilations_0 = const()[name = string("input_43_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_43_groups_0 = const()[name = string("input_43_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1033148]> module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15816640))), indices_nonzero_data = tensor<uint6, [1033148]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16078848))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16853824))))[name = string("module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1033148]> module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15308224))), indices_nonzero_data = tensor<uint6, [1033148]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15570432))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16345408))))[name = string("module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_41_cast_fp16 = transpose(perm = input_41_perm_0, x = x_17_cast_fp16)[name = string("transpose_307")];
-            tensor<fp16, [1, 2048, 188]> input_43_cast_fp16 = conv(dilations = input_43_dilations_0, groups = input_43_groups_0, pad = input_43_pad_0, pad_type = input_43_pad_type_0, strides = input_43_strides_0, weight = module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_41_cast_fp16)[name = string("input_43_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_41_cast_fp16 = transpose(perm = input_41_perm_0, x = x_17_cast_fp16)[name = string("transpose_307")];
+            tensor<fp16, [1, 2048, 128]> input_43_cast_fp16 = conv(dilations = input_43_dilations_0, groups = input_43_groups_0, pad = input_43_pad_0, pad_type = input_43_pad_type_0, strides = input_43_strides_0, weight = module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_41_cast_fp16)[name = string("input_43_cast_fp16")];
             int32 x_19_split_num_splits_0 = const()[name = string("x_19_split_num_splits_0"), val = int32(2)];
             int32 x_19_split_axis_0 = const()[name = string("x_19_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_19_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_19_split_cast_fp16_1 = split(axis = x_19_split_axis_0, num_splits = x_19_split_num_splits_0, x = input_43_cast_fp16)[name = string("x_19_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_19_split_1_sigmoid_cast_fp16 = sigmoid(x = x_19_split_cast_fp16_1)[name = string("x_19_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_19_cast_fp16 = mul(x = x_19_split_cast_fp16_0, y = x_19_split_1_sigmoid_cast_fp16)[name = string("x_19_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_19_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_19_split_cast_fp16_1 = split(axis = x_19_split_axis_0, num_splits = x_19_split_num_splits_0, x = input_43_cast_fp16)[name = string("x_19_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_19_split_1_sigmoid_cast_fp16 = sigmoid(x = x_19_split_cast_fp16_1)[name = string("x_19_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_19_cast_fp16 = mul(x = x_19_split_cast_fp16_0, y = x_19_split_1_sigmoid_cast_fp16)[name = string("x_19_cast_fp16")];
             tensor<int32, [1]> var_328_axes_0 = const()[name = string("op_328_axes_0"), val = tensor<int32, [1]>([1])];
-            tensor<bool, [1, 1, 188]> var_328 = expand_dims(axes = var_328_axes_0, x = pad_mask)[name = string("op_328")];
-            tensor<fp16, [1, 1024, 188]> input_45_cast_fp16 = select(a = var_11_to_fp16, b = x_19_cast_fp16, cond = var_328)[name = string("input_45_cast_fp16")];
+            tensor<bool, [1, 1, 128]> var_328 = expand_dims(axes = var_328_axes_0, x = pad_mask)[name = string("op_328")];
+            tensor<fp16, [1, 1024, 128]> input_45_cast_fp16 = select(a = var_11_to_fp16, b = x_19_cast_fp16, cond = var_328)[name = string("input_45_cast_fp16")];
             tensor<int32, [6]> input_47_pad_0 = const()[name = string("input_47_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_47_mode_0 = const()[name = string("input_47_mode_0"), val = string("constant")];
             fp16 const_17_to_fp16 = const()[name = string("const_17_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_47_cast_fp16 = pad(constant_val = const_17_to_fp16, mode = input_47_mode_0, pad = input_47_pad_0, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")];
+            tensor<fp16, [1, 1024, 136]> input_47_cast_fp16 = pad(constant_val = const_17_to_fp16, mode = input_47_mode_0, pad = input_47_pad_0, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")];
             string input_49_pad_type_0 = const()[name = string("input_49_pad_type_0"), val = string("valid")];
             int32 input_49_groups_0 = const()[name = string("input_49_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_49_strides_0 = const()[name = string("input_49_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_49_pad_0 = const()[name = string("input_49_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_49_dilations_0 = const()[name = string("input_49_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_248_to_fp16 = const()[name = string("const_248_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16886656)))];
-            tensor<fp16, [1024]> const_249_to_fp16 = const()[name = string("const_249_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16905152)))];
-            tensor<fp16, [1, 1024, 188]> input_51_cast_fp16 = conv(bias = const_249_to_fp16, dilations = input_49_dilations_0, groups = input_49_groups_0, pad = input_49_pad_0, pad_type = input_49_pad_type_0, strides = input_49_strides_0, weight = const_248_to_fp16, x = input_47_cast_fp16)[name = string("input_51_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_53_cast_fp16 = silu(x = input_51_cast_fp16)[name = string("input_53_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_248_to_fp16 = const()[name = string("const_248_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16378240)))];
+            tensor<fp16, [1024]> const_249_to_fp16 = const()[name = string("const_249_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16396736)))];
+            tensor<fp16, [1, 1024, 128]> input_51_cast_fp16 = conv(bias = const_249_to_fp16, dilations = input_49_dilations_0, groups = input_49_groups_0, pad = input_49_pad_0, pad_type = input_49_pad_type_0, strides = input_49_strides_0, weight = const_248_to_fp16, x = input_47_cast_fp16)[name = string("input_51_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_53_cast_fp16 = silu(x = input_51_cast_fp16)[name = string("input_53_cast_fp16")];
             string x_21_pad_type_0 = const()[name = string("x_21_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_21_strides_0 = const()[name = string("x_21_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_21_pad_0 = const()[name = string("x_21_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_21_dilations_0 = const()[name = string("x_21_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_21_groups_0 = const()[name = string("x_21_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [512831]> module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16907264))), indices_nonzero_data = tensor<uint6, [512831]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17038400))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17423104))))[name = string("module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [512831]> module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16398848))), indices_nonzero_data = tensor<uint6, [512831]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16529984))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16914688))))[name = string("module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_21_cast_fp16 = conv(dilations = x_21_dilations_0, groups = x_21_groups_0, pad = x_21_pad_0, pad_type = x_21_pad_type_0, strides = x_21_strides_0, weight = module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_53_cast_fp16)[name = string("x_21_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_21_cast_fp16 = conv(dilations = x_21_dilations_0, groups = x_21_groups_0, pad = x_21_pad_0, pad_type = x_21_pad_type_0, strides = x_21_strides_0, weight = module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_53_cast_fp16)[name = string("x_21_cast_fp16")];
             tensor<int32, [3]> input_55_perm_0 = const()[name = string("input_55_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_55_cast_fp16 = transpose(perm = input_55_perm_0, x = x_21_cast_fp16)[name = string("transpose_306")];
-            tensor<fp16, [1, 188, 1024]> input_57_cast_fp16 = add(x = input_39_cast_fp16, y = input_55_cast_fp16)[name = string("input_57_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_55_cast_fp16 = transpose(perm = input_55_perm_0, x = x_21_cast_fp16)[name = string("transpose_306")];
+            tensor<fp16, [1, 128, 1024]> input_57_cast_fp16 = add(x = input_39_cast_fp16, y = input_55_cast_fp16)[name = string("input_57_cast_fp16")];
             tensor<int32, [1]> input_59_axes_0 = const()[name = string("input_59_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_0_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_0_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17439552)))];
-            tensor<fp16, [1024]> module_layers_0_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_0_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17441664)))];
-            tensor<fp16, [1, 188, 1024]> input_59_cast_fp16 = layer_norm(axes = input_59_axes_0, beta = module_layers_0_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_feed_forward2_weight_to_fp16, x = input_57_cast_fp16)[name = string("input_59_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2071218]> module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17443776))), indices_nonzero_data = tensor<uint6, [2071218]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17968128))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19521664))))[name = string("module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_0_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_0_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16931136)))];
+            tensor<fp16, [1024]> module_layers_0_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_0_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16933248)))];
+            tensor<fp16, [1, 128, 1024]> input_59_cast_fp16 = layer_norm(axes = input_59_axes_0, beta = module_layers_0_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_feed_forward2_weight_to_fp16, x = input_57_cast_fp16)[name = string("input_59_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2071218]> module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16935360))), indices_nonzero_data = tensor<uint6, [2071218]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17459712))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19013248))))[name = string("module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_8_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_59_cast_fp16)[name = string("linear_8_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_63_cast_fp16 = silu(x = linear_8_cast_fp16)[name = string("input_63_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2031949]> module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19587264))), indices_nonzero_data = tensor<uint6, [2031949]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20111616))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21635648))))[name = string("module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_8_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_59_cast_fp16)[name = string("linear_8_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_63_cast_fp16 = silu(x = linear_8_cast_fp16)[name = string("input_63_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2031949]> module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19078848))), indices_nonzero_data = tensor<uint6, [2031949]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19603200))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21127232))))[name = string("module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_9_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_63_cast_fp16)[name = string("linear_9_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_9_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_63_cast_fp16)[name = string("linear_9_cast_fp16")];
             fp16 var_366_to_fp16 = const()[name = string("op_366_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_367_cast_fp16 = mul(x = linear_9_cast_fp16, y = var_366_to_fp16)[name = string("op_367_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_69_cast_fp16 = add(x = input_57_cast_fp16, y = var_367_cast_fp16)[name = string("input_69_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_367_cast_fp16 = mul(x = linear_9_cast_fp16, y = var_366_to_fp16)[name = string("op_367_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_69_cast_fp16 = add(x = input_57_cast_fp16, y = var_367_cast_fp16)[name = string("input_69_cast_fp16")];
             tensor<int32, [1]> input_71_axes_0 = const()[name = string("input_71_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_0_norm_out_weight_to_fp16 = const()[name = string("module_layers_0_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21652096)))];
-            tensor<fp16, [1024]> module_layers_0_norm_out_bias_to_fp16 = const()[name = string("module_layers_0_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21654208)))];
-            tensor<fp16, [1, 188, 1024]> input_71_cast_fp16 = layer_norm(axes = input_71_axes_0, beta = module_layers_0_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_out_weight_to_fp16, x = input_69_cast_fp16)[name = string("input_71_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_0_norm_out_weight_to_fp16 = const()[name = string("module_layers_0_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21143680)))];
+            tensor<fp16, [1024]> module_layers_0_norm_out_bias_to_fp16 = const()[name = string("module_layers_0_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21145792)))];
+            tensor<fp16, [1, 128, 1024]> input_71_cast_fp16 = layer_norm(axes = input_71_axes_0, beta = module_layers_0_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_out_weight_to_fp16, x = input_69_cast_fp16)[name = string("input_71_cast_fp16")];
             tensor<int32, [1]> input_73_axes_0 = const()[name = string("input_73_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_1_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_1_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21656320)))];
-            tensor<fp16, [1024]> module_layers_1_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_1_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21658432)))];
-            tensor<fp16, [1, 188, 1024]> input_73_cast_fp16 = layer_norm(axes = input_73_axes_0, beta = module_layers_1_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_feed_forward1_weight_to_fp16, x = input_71_cast_fp16)[name = string("input_73_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2039600]> module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21660544))), indices_nonzero_data = tensor<uint6, [2039600]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22184896))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23714688))))[name = string("module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_1_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_1_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21147904)))];
+            tensor<fp16, [1024]> module_layers_1_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_1_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21150016)))];
+            tensor<fp16, [1, 128, 1024]> input_73_cast_fp16 = layer_norm(axes = input_73_axes_0, beta = module_layers_1_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_feed_forward1_weight_to_fp16, x = input_71_cast_fp16)[name = string("input_73_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2039600]> module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21152128))), indices_nonzero_data = tensor<uint6, [2039600]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21676480))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23206272))))[name = string("module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_10_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_73_cast_fp16)[name = string("linear_10_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_77_cast_fp16 = silu(x = linear_10_cast_fp16)[name = string("input_77_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1944084]> module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23780288))), indices_nonzero_data = tensor<uint6, [1944084]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24304640))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25762816))))[name = string("module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_10_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_73_cast_fp16)[name = string("linear_10_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_77_cast_fp16 = silu(x = linear_10_cast_fp16)[name = string("input_77_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1944084]> module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23271872))), indices_nonzero_data = tensor<uint6, [1944084]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23796224))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25254400))))[name = string("module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_11_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_77_cast_fp16)[name = string("linear_11_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_11_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_77_cast_fp16)[name = string("linear_11_cast_fp16")];
             fp16 var_395_to_fp16 = const()[name = string("op_395_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_396_cast_fp16 = mul(x = linear_11_cast_fp16, y = var_395_to_fp16)[name = string("op_396_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_83_cast_fp16 = add(x = input_71_cast_fp16, y = var_396_cast_fp16)[name = string("input_83_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_396_cast_fp16 = mul(x = linear_11_cast_fp16, y = var_395_to_fp16)[name = string("op_396_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_83_cast_fp16 = add(x = input_71_cast_fp16, y = var_396_cast_fp16)[name = string("input_83_cast_fp16")];
             tensor<int32, [1]> query_3_axes_0 = const()[name = string("query_3_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_1_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_1_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25779264)))];
-            tensor<fp16, [1024]> module_layers_1_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_1_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25781376)))];
-            tensor<fp16, [1, 188, 1024]> query_3_cast_fp16 = layer_norm(axes = query_3_axes_0, beta = module_layers_1_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_self_att_weight_to_fp16, x = input_83_cast_fp16)[name = string("query_3_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523158]> module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25783488))), indices_nonzero_data = tensor<uint6, [523158]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25914624))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26307072))))[name = string("module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_1_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_1_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25270848)))];
+            tensor<fp16, [1024]> module_layers_1_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_1_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25272960)))];
+            tensor<fp16, [1, 128, 1024]> query_3_cast_fp16 = layer_norm(axes = query_3_axes_0, beta = module_layers_1_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_self_att_weight_to_fp16, x = input_83_cast_fp16)[name = string("query_3_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523158]> module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25275072))), indices_nonzero_data = tensor<uint6, [523158]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25406208))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25798656))))[name = string("module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_12_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_3_cast_fp16)[name = string("linear_12_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_12_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_3_cast_fp16)[name = string("linear_12_cast_fp16")];
             tensor<int32, [4]> var_412 = const()[name = string("op_412"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_7_cast_fp16 = reshape(shape = var_412, x = linear_12_cast_fp16)[name = string("q_7_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [515603]> module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26323520))), indices_nonzero_data = tensor<uint6, [515603]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26454656))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26841472))))[name = string("module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> q_7_cast_fp16 = reshape(shape = var_412, x = linear_12_cast_fp16)[name = string("q_7_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [515603]> module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25815104))), indices_nonzero_data = tensor<uint6, [515603]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25946240))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26333056))))[name = string("module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_13_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_3_cast_fp16)[name = string("linear_13_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_13_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_3_cast_fp16)[name = string("linear_13_cast_fp16")];
             tensor<int32, [4]> var_416 = const()[name = string("op_416"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_5_cast_fp16 = reshape(shape = var_416, x = linear_13_cast_fp16)[name = string("k_5_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [511483]> module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26857920))), indices_nonzero_data = tensor<uint6, [511483]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26989056))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27372736))))[name = string("module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> k_5_cast_fp16 = reshape(shape = var_416, x = linear_13_cast_fp16)[name = string("k_5_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [511483]> module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26349504))), indices_nonzero_data = tensor<uint6, [511483]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26480640))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26864320))))[name = string("module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_14_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_3_cast_fp16)[name = string("linear_14_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_14_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_3_cast_fp16)[name = string("linear_14_cast_fp16")];
             tensor<int32, [4]> var_420 = const()[name = string("op_420"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_3_cast_fp16 = reshape(shape = var_420, x = linear_14_cast_fp16)[name = string("v_3_cast_fp16")];
+            tensor<fp16, [1, 128, 8, 128]> v_3_cast_fp16 = reshape(shape = var_420, x = linear_14_cast_fp16)[name = string("v_3_cast_fp16")];
             tensor<int32, [4]> value_5_perm_0 = const()[name = string("value_5_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_1_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_1_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27389184)))];
-            tensor<fp16, [1, 188, 8, 128]> var_432_cast_fp16 = add(x = q_7_cast_fp16, y = module_layers_1_self_attn_pos_bias_u_to_fp16)[name = string("op_432_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_1_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_1_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27391296)))];
-            tensor<fp16, [1, 188, 8, 128]> var_434_cast_fp16 = add(x = q_7_cast_fp16, y = module_layers_1_self_attn_pos_bias_v_to_fp16)[name = string("op_434_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_1_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_1_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26880768)))];
+            tensor<fp16, [1, 128, 8, 128]> var_432_cast_fp16 = add(x = q_7_cast_fp16, y = module_layers_1_self_attn_pos_bias_u_to_fp16)[name = string("op_432_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_1_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_1_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26882880)))];
+            tensor<fp16, [1, 128, 8, 128]> var_434_cast_fp16 = add(x = q_7_cast_fp16, y = module_layers_1_self_attn_pos_bias_v_to_fp16)[name = string("op_434_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_3_perm_0 = const()[name = string("q_with_bias_v_3_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_29_transpose_x_0 = const()[name = string("x_29_transpose_x_0"), val = bool(false)];
             bool x_29_transpose_y_0 = const()[name = string("x_29_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_436_to_fp16 = const()[name = string("op_436_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27393408)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_3_cast_fp16 = transpose(perm = q_with_bias_v_3_perm_0, x = var_434_cast_fp16)[name = string("transpose_305")];
-            tensor<fp16, [1, 8, 188, 375]> x_29_cast_fp16 = matmul(transpose_x = x_29_transpose_x_0, transpose_y = x_29_transpose_y_0, x = q_with_bias_v_3_cast_fp16, y = var_436_to_fp16)[name = string("x_29_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 255]> var_436_to_fp16 = const()[name = string("op_436_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307944128)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_3_cast_fp16 = transpose(perm = q_with_bias_v_3_perm_0, x = var_434_cast_fp16)[name = string("transpose_305")];
+            tensor<fp16, [1, 8, 128, 255]> x_29_cast_fp16 = matmul(transpose_x = x_29_transpose_x_0, transpose_y = x_29_transpose_y_0, x = q_with_bias_v_3_cast_fp16, y = var_436_to_fp16)[name = string("x_29_cast_fp16")];
             tensor<int32, [8]> x_31_pad_0 = const()[name = string("x_31_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_31_mode_0 = const()[name = string("x_31_mode_0"), val = string("constant")];
             fp16 const_24_to_fp16 = const()[name = string("const_24_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_31_cast_fp16 = pad(constant_val = const_24_to_fp16, mode = x_31_mode_0, pad = x_31_pad_0, x = x_29_cast_fp16)[name = string("x_31_cast_fp16")];
-            tensor<int32, [4]> var_444 = const()[name = string("op_444"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_33_cast_fp16 = reshape(shape = var_444, x = x_31_cast_fp16)[name = string("x_33_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 256]> x_31_cast_fp16 = pad(constant_val = const_24_to_fp16, mode = x_31_mode_0, pad = x_31_pad_0, x = x_29_cast_fp16)[name = string("x_31_cast_fp16")];
+            tensor<int32, [4]> var_444 = const()[name = string("op_444"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_33_cast_fp16 = reshape(shape = var_444, x = x_31_cast_fp16)[name = string("x_33_cast_fp16")];
             tensor<int32, [4]> var_448_begin_0 = const()[name = string("op_448_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_448_end_0 = const()[name = string("op_448_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_448_end_0 = const()[name = string("op_448_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
             tensor<bool, [4]> var_448_end_mask_0 = const()[name = string("op_448_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_448_cast_fp16 = slice_by_index(begin = var_448_begin_0, end = var_448_end_0, end_mask = var_448_end_mask_0, x = x_33_cast_fp16)[name = string("op_448_cast_fp16")];
-            tensor<int32, [4]> var_449 = const()[name = string("op_449"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_5_cast_fp16 = reshape(shape = var_449, x = var_448_cast_fp16)[name = string("matrix_bd_5_cast_fp16")];
+            tensor<fp16, [1, 8, 255, 128]> var_448_cast_fp16 = slice_by_index(begin = var_448_begin_0, end = var_448_end_0, end_mask = var_448_end_mask_0, x = x_33_cast_fp16)[name = string("op_448_cast_fp16")];
+            tensor<int32, [4]> var_449 = const()[name = string("op_449"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_5_cast_fp16 = reshape(shape = var_449, x = var_448_cast_fp16)[name = string("matrix_bd_5_cast_fp16")];
             bool matrix_ac_3_transpose_x_0 = const()[name = string("matrix_ac_3_transpose_x_0"), val = bool(false)];
             bool matrix_ac_3_transpose_y_0 = const()[name = string("matrix_ac_3_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_98_perm_0 = const()[name = string("transpose_98_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_99_perm_0 = const()[name = string("transpose_99_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_99 = transpose(perm = transpose_99_perm_0, x = k_5_cast_fp16)[name = string("transpose_303")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_98 = transpose(perm = transpose_98_perm_0, x = var_432_cast_fp16)[name = string("transpose_304")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_3_cast_fp16 = matmul(transpose_x = matrix_ac_3_transpose_x_0, transpose_y = matrix_ac_3_transpose_y_0, x = transpose_98, y = transpose_99)[name = string("matrix_ac_3_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_99 = transpose(perm = transpose_99_perm_0, x = k_5_cast_fp16)[name = string("transpose_303")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_98 = transpose(perm = transpose_98_perm_0, x = var_432_cast_fp16)[name = string("transpose_304")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_3_cast_fp16 = matmul(transpose_x = matrix_ac_3_transpose_x_0, transpose_y = matrix_ac_3_transpose_y_0, x = transpose_98, y = transpose_99)[name = string("matrix_ac_3_cast_fp16")];
             tensor<int32, [4]> matrix_bd_7_begin_0 = const()[name = string("matrix_bd_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_7_end_0 = const()[name = string("matrix_bd_7_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_7_end_0 = const()[name = string("matrix_bd_7_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
             tensor<bool, [4]> matrix_bd_7_end_mask_0 = const()[name = string("matrix_bd_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_7_cast_fp16 = slice_by_index(begin = matrix_bd_7_begin_0, end = matrix_bd_7_end_0, end_mask = matrix_bd_7_end_mask_0, x = matrix_bd_5_cast_fp16)[name = string("matrix_bd_7_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_458_cast_fp16 = add(x = matrix_ac_3_cast_fp16, y = matrix_bd_7_cast_fp16)[name = string("op_458_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_7_cast_fp16 = slice_by_index(begin = matrix_bd_7_begin_0, end = matrix_bd_7_end_0, end_mask = matrix_bd_7_end_mask_0, x = matrix_bd_5_cast_fp16)[name = string("matrix_bd_7_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_458_cast_fp16 = add(x = matrix_ac_3_cast_fp16, y = matrix_bd_7_cast_fp16)[name = string("op_458_cast_fp16")];
             fp16 _inversed_scores_5_y_0_to_fp16 = const()[name = string("_inversed_scores_5_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_5_cast_fp16 = mul(x = var_458_cast_fp16, y = _inversed_scores_5_y_0_to_fp16)[name = string("_inversed_scores_5_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_7_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_5_cast_fp16, cond = mask_3)[name = string("scores_7_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_464_cast_fp16 = softmax(axis = var_30, x = scores_7_cast_fp16)[name = string("op_464_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_85_cast_fp16 = select(a = var_11_to_fp16, b = var_464_cast_fp16, cond = mask_3)[name = string("input_85_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_5_cast_fp16 = mul(x = var_458_cast_fp16, y = _inversed_scores_5_y_0_to_fp16)[name = string("_inversed_scores_5_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_7_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_5_cast_fp16, cond = mask_3)[name = string("scores_7_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_464_cast_fp16 = softmax(axis = var_30, x = scores_7_cast_fp16)[name = string("op_464_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_85_cast_fp16 = select(a = var_11_to_fp16, b = var_464_cast_fp16, cond = mask_3)[name = string("input_85_cast_fp16")];
             bool x_35_transpose_x_0 = const()[name = string("x_35_transpose_x_0"), val = bool(false)];
             bool x_35_transpose_y_0 = const()[name = string("x_35_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_5_cast_fp16 = transpose(perm = value_5_perm_0, x = v_3_cast_fp16)[name = string("transpose_302")];
-            tensor<fp16, [1, 8, 188, 128]> x_35_cast_fp16 = matmul(transpose_x = x_35_transpose_x_0, transpose_y = x_35_transpose_y_0, x = input_85_cast_fp16, y = value_5_cast_fp16)[name = string("x_35_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> value_5_cast_fp16 = transpose(perm = value_5_perm_0, x = v_3_cast_fp16)[name = string("transpose_302")];
+            tensor<fp16, [1, 8, 128, 128]> x_35_cast_fp16 = matmul(transpose_x = x_35_transpose_x_0, transpose_y = x_35_transpose_y_0, x = input_85_cast_fp16, y = value_5_cast_fp16)[name = string("x_35_cast_fp16")];
             tensor<int32, [4]> var_468_perm_0 = const()[name = string("op_468_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_469 = const()[name = string("op_469"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_468_cast_fp16 = transpose(perm = var_468_perm_0, x = x_35_cast_fp16)[name = string("transpose_301")];
-            tensor<fp16, [1, 188, 1024]> input_87_cast_fp16 = reshape(shape = var_469, x = var_468_cast_fp16)[name = string("input_87_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523005]> module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28161472))), indices_nonzero_data = tensor<uint6, [523005]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28292608))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28684928))))[name = string("module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> var_468_cast_fp16 = transpose(perm = var_468_perm_0, x = x_35_cast_fp16)[name = string("transpose_301")];
+            tensor<fp16, [1, 128, 1024]> input_87_cast_fp16 = reshape(shape = var_469, x = var_468_cast_fp16)[name = string("input_87_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523005]> module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27145152))), indices_nonzero_data = tensor<uint6, [523005]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27276288))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27668608))))[name = string("module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_16_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_87_cast_fp16)[name = string("linear_16_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_91_cast_fp16 = add(x = input_83_cast_fp16, y = linear_16_cast_fp16)[name = string("input_91_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_16_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_87_cast_fp16)[name = string("linear_16_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_91_cast_fp16 = add(x = input_83_cast_fp16, y = linear_16_cast_fp16)[name = string("input_91_cast_fp16")];
             tensor<int32, [1]> x_39_axes_0 = const()[name = string("x_39_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_1_norm_conv_weight_to_fp16 = const()[name = string("module_layers_1_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28701376)))];
-            tensor<fp16, [1024]> module_layers_1_norm_conv_bias_to_fp16 = const()[name = string("module_layers_1_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28703488)))];
-            tensor<fp16, [1, 188, 1024]> x_39_cast_fp16 = layer_norm(axes = x_39_axes_0, beta = module_layers_1_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_conv_weight_to_fp16, x = input_91_cast_fp16)[name = string("x_39_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_1_norm_conv_weight_to_fp16 = const()[name = string("module_layers_1_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27685056)))];
+            tensor<fp16, [1024]> module_layers_1_norm_conv_bias_to_fp16 = const()[name = string("module_layers_1_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27687168)))];
+            tensor<fp16, [1, 128, 1024]> x_39_cast_fp16 = layer_norm(axes = x_39_axes_0, beta = module_layers_1_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_conv_weight_to_fp16, x = input_91_cast_fp16)[name = string("x_39_cast_fp16")];
             tensor<int32, [3]> input_93_perm_0 = const()[name = string("input_93_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_95_pad_type_0 = const()[name = string("input_95_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_95_strides_0 = const()[name = string("input_95_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_95_pad_0 = const()[name = string("input_95_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_95_dilations_0 = const()[name = string("input_95_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_95_groups_0 = const()[name = string("input_95_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1039540]> module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28705600))), indices_nonzero_data = tensor<uint6, [1039540]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28967808))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29747584))))[name = string("module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1039540]> module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27689280))), indices_nonzero_data = tensor<uint6, [1039540]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27951488))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28731264))))[name = string("module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_93_cast_fp16 = transpose(perm = input_93_perm_0, x = x_39_cast_fp16)[name = string("transpose_300")];
-            tensor<fp16, [1, 2048, 188]> input_95_cast_fp16 = conv(dilations = input_95_dilations_0, groups = input_95_groups_0, pad = input_95_pad_0, pad_type = input_95_pad_type_0, strides = input_95_strides_0, weight = module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_93_cast_fp16)[name = string("input_95_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_93_cast_fp16 = transpose(perm = input_93_perm_0, x = x_39_cast_fp16)[name = string("transpose_300")];
+            tensor<fp16, [1, 2048, 128]> input_95_cast_fp16 = conv(dilations = input_95_dilations_0, groups = input_95_groups_0, pad = input_95_pad_0, pad_type = input_95_pad_type_0, strides = input_95_strides_0, weight = module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_93_cast_fp16)[name = string("input_95_cast_fp16")];
             int32 x_41_split_num_splits_0 = const()[name = string("x_41_split_num_splits_0"), val = int32(2)];
             int32 x_41_split_axis_0 = const()[name = string("x_41_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_41_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_41_split_cast_fp16_1 = split(axis = x_41_split_axis_0, num_splits = x_41_split_num_splits_0, x = input_95_cast_fp16)[name = string("x_41_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_41_split_1_sigmoid_cast_fp16 = sigmoid(x = x_41_split_cast_fp16_1)[name = string("x_41_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_41_cast_fp16 = mul(x = x_41_split_cast_fp16_0, y = x_41_split_1_sigmoid_cast_fp16)[name = string("x_41_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_97_cast_fp16 = select(a = var_11_to_fp16, b = x_41_cast_fp16, cond = var_328)[name = string("input_97_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_41_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_41_split_cast_fp16_1 = split(axis = x_41_split_axis_0, num_splits = x_41_split_num_splits_0, x = input_95_cast_fp16)[name = string("x_41_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_41_split_1_sigmoid_cast_fp16 = sigmoid(x = x_41_split_cast_fp16_1)[name = string("x_41_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_41_cast_fp16 = mul(x = x_41_split_cast_fp16_0, y = x_41_split_1_sigmoid_cast_fp16)[name = string("x_41_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_97_cast_fp16 = select(a = var_11_to_fp16, b = x_41_cast_fp16, cond = var_328)[name = string("input_97_cast_fp16")];
             tensor<int32, [6]> input_99_pad_0 = const()[name = string("input_99_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_99_mode_0 = const()[name = string("input_99_mode_0"), val = string("constant")];
             fp16 const_27_to_fp16 = const()[name = string("const_27_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_99_cast_fp16 = pad(constant_val = const_27_to_fp16, mode = input_99_mode_0, pad = input_99_pad_0, x = input_97_cast_fp16)[name = string("input_99_cast_fp16")];
+            tensor<fp16, [1, 1024, 136]> input_99_cast_fp16 = pad(constant_val = const_27_to_fp16, mode = input_99_mode_0, pad = input_99_pad_0, x = input_97_cast_fp16)[name = string("input_99_cast_fp16")];
             string input_101_pad_type_0 = const()[name = string("input_101_pad_type_0"), val = string("valid")];
             int32 input_101_groups_0 = const()[name = string("input_101_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_101_strides_0 = const()[name = string("input_101_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_101_pad_0 = const()[name = string("input_101_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_101_dilations_0 = const()[name = string("input_101_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_250_to_fp16 = const()[name = string("const_250_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29780416)))];
-            tensor<fp16, [1024]> const_251_to_fp16 = const()[name = string("const_251_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29798912)))];
-            tensor<fp16, [1, 1024, 188]> input_103_cast_fp16 = conv(bias = const_251_to_fp16, dilations = input_101_dilations_0, groups = input_101_groups_0, pad = input_101_pad_0, pad_type = input_101_pad_type_0, strides = input_101_strides_0, weight = const_250_to_fp16, x = input_99_cast_fp16)[name = string("input_103_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_105_cast_fp16 = silu(x = input_103_cast_fp16)[name = string("input_105_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_250_to_fp16 = const()[name = string("const_250_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28764096)))];
+            tensor<fp16, [1024]> const_251_to_fp16 = const()[name = string("const_251_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28782592)))];
+            tensor<fp16, [1, 1024, 128]> input_103_cast_fp16 = conv(bias = const_251_to_fp16, dilations = input_101_dilations_0, groups = input_101_groups_0, pad = input_101_pad_0, pad_type = input_101_pad_type_0, strides = input_101_strides_0, weight = const_250_to_fp16, x = input_99_cast_fp16)[name = string("input_103_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_105_cast_fp16 = silu(x = input_103_cast_fp16)[name = string("input_105_cast_fp16")];
             string x_43_pad_type_0 = const()[name = string("x_43_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_43_strides_0 = const()[name = string("x_43_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_43_pad_0 = const()[name = string("x_43_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_43_dilations_0 = const()[name = string("x_43_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_43_groups_0 = const()[name = string("x_43_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [515869]> module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29801024))), indices_nonzero_data = tensor<uint6, [515869]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29932160))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30319168))))[name = string("module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [515869]> module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28784704))), indices_nonzero_data = tensor<uint6, [515869]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28915840))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29302848))))[name = string("module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_43_cast_fp16 = conv(dilations = x_43_dilations_0, groups = x_43_groups_0, pad = x_43_pad_0, pad_type = x_43_pad_type_0, strides = x_43_strides_0, weight = module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_105_cast_fp16)[name = string("x_43_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_43_cast_fp16 = conv(dilations = x_43_dilations_0, groups = x_43_groups_0, pad = x_43_pad_0, pad_type = x_43_pad_type_0, strides = x_43_strides_0, weight = module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_105_cast_fp16)[name = string("x_43_cast_fp16")];
             tensor<int32, [3]> input_107_perm_0 = const()[name = string("input_107_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_107_cast_fp16 = transpose(perm = input_107_perm_0, x = x_43_cast_fp16)[name = string("transpose_299")];
-            tensor<fp16, [1, 188, 1024]> input_109_cast_fp16 = add(x = input_91_cast_fp16, y = input_107_cast_fp16)[name = string("input_109_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_107_cast_fp16 = transpose(perm = input_107_perm_0, x = x_43_cast_fp16)[name = string("transpose_299")];
+            tensor<fp16, [1, 128, 1024]> input_109_cast_fp16 = add(x = input_91_cast_fp16, y = input_107_cast_fp16)[name = string("input_109_cast_fp16")];
             tensor<int32, [1]> input_111_axes_0 = const()[name = string("input_111_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_1_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_1_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30335616)))];
-            tensor<fp16, [1024]> module_layers_1_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_1_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30337728)))];
-            tensor<fp16, [1, 188, 1024]> input_111_cast_fp16 = layer_norm(axes = input_111_axes_0, beta = module_layers_1_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_feed_forward2_weight_to_fp16, x = input_109_cast_fp16)[name = string("input_111_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2057265]> module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30339840))), indices_nonzero_data = tensor<uint6, [2057265]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30864192))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32407232))))[name = string("module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_1_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_1_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29319296)))];
+            tensor<fp16, [1024]> module_layers_1_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_1_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29321408)))];
+            tensor<fp16, [1, 128, 1024]> input_111_cast_fp16 = layer_norm(axes = input_111_axes_0, beta = module_layers_1_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_feed_forward2_weight_to_fp16, x = input_109_cast_fp16)[name = string("input_111_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2057265]> module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29323520))), indices_nonzero_data = tensor<uint6, [2057265]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29847872))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31390912))))[name = string("module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_17_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_111_cast_fp16)[name = string("linear_17_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_115_cast_fp16 = silu(x = linear_17_cast_fp16)[name = string("input_115_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1937635]> module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32472832))), indices_nonzero_data = tensor<uint6, [1937635]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32997184))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34450496))))[name = string("module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_17_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_111_cast_fp16)[name = string("linear_17_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_115_cast_fp16 = silu(x = linear_17_cast_fp16)[name = string("input_115_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1937635]> module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31456512))), indices_nonzero_data = tensor<uint6, [1937635]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31980864))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33434176))))[name = string("module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_18_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_115_cast_fp16)[name = string("linear_18_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_18_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_115_cast_fp16)[name = string("linear_18_cast_fp16")];
             fp16 var_529_to_fp16 = const()[name = string("op_529_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_530_cast_fp16 = mul(x = linear_18_cast_fp16, y = var_529_to_fp16)[name = string("op_530_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_121_cast_fp16 = add(x = input_109_cast_fp16, y = var_530_cast_fp16)[name = string("input_121_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_530_cast_fp16 = mul(x = linear_18_cast_fp16, y = var_529_to_fp16)[name = string("op_530_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_121_cast_fp16 = add(x = input_109_cast_fp16, y = var_530_cast_fp16)[name = string("input_121_cast_fp16")];
             tensor<int32, [1]> input_123_axes_0 = const()[name = string("input_123_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_1_norm_out_weight_to_fp16 = const()[name = string("module_layers_1_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34466944)))];
-            tensor<fp16, [1024]> module_layers_1_norm_out_bias_to_fp16 = const()[name = string("module_layers_1_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34469056)))];
-            tensor<fp16, [1, 188, 1024]> input_123_cast_fp16 = layer_norm(axes = input_123_axes_0, beta = module_layers_1_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_out_weight_to_fp16, x = input_121_cast_fp16)[name = string("input_123_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_1_norm_out_weight_to_fp16 = const()[name = string("module_layers_1_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33450624)))];
+            tensor<fp16, [1024]> module_layers_1_norm_out_bias_to_fp16 = const()[name = string("module_layers_1_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33452736)))];
+            tensor<fp16, [1, 128, 1024]> input_123_cast_fp16 = layer_norm(axes = input_123_axes_0, beta = module_layers_1_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_out_weight_to_fp16, x = input_121_cast_fp16)[name = string("input_123_cast_fp16")];
             tensor<int32, [1]> input_125_axes_0 = const()[name = string("input_125_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_2_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_2_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34471168)))];
-            tensor<fp16, [1024]> module_layers_2_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_2_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34473280)))];
-            tensor<fp16, [1, 188, 1024]> input_125_cast_fp16 = layer_norm(axes = input_125_axes_0, beta = module_layers_2_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_feed_forward1_weight_to_fp16, x = input_123_cast_fp16)[name = string("input_125_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2079554]> module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34475392))), indices_nonzero_data = tensor<uint6, [2079554]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34999744))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36559488))))[name = string("module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_2_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_2_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33454848)))];
+            tensor<fp16, [1024]> module_layers_2_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_2_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33456960)))];
+            tensor<fp16, [1, 128, 1024]> input_125_cast_fp16 = layer_norm(axes = input_125_axes_0, beta = module_layers_2_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_feed_forward1_weight_to_fp16, x = input_123_cast_fp16)[name = string("input_125_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2079554]> module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33459072))), indices_nonzero_data = tensor<uint6, [2079554]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33983424))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35543168))))[name = string("module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_125_cast_fp16)[name = string("linear_19_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_129_cast_fp16 = silu(x = linear_19_cast_fp16)[name = string("input_129_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1998259]> module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36625088))), indices_nonzero_data = tensor<uint6, [1998259]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37149440))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38648256))))[name = string("module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_125_cast_fp16)[name = string("linear_19_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_129_cast_fp16 = silu(x = linear_19_cast_fp16)[name = string("input_129_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1998259]> module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35608768))), indices_nonzero_data = tensor<uint6, [1998259]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36133120))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37631936))))[name = string("module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_20_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_129_cast_fp16)[name = string("linear_20_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_20_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_129_cast_fp16)[name = string("linear_20_cast_fp16")];
             fp16 var_558_to_fp16 = const()[name = string("op_558_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_559_cast_fp16 = mul(x = linear_20_cast_fp16, y = var_558_to_fp16)[name = string("op_559_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_135_cast_fp16 = add(x = input_123_cast_fp16, y = var_559_cast_fp16)[name = string("input_135_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_559_cast_fp16 = mul(x = linear_20_cast_fp16, y = var_558_to_fp16)[name = string("op_559_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_135_cast_fp16 = add(x = input_123_cast_fp16, y = var_559_cast_fp16)[name = string("input_135_cast_fp16")];
             tensor<int32, [1]> query_5_axes_0 = const()[name = string("query_5_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_2_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_2_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38664704)))];
-            tensor<fp16, [1024]> module_layers_2_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_2_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38666816)))];
-            tensor<fp16, [1, 188, 1024]> query_5_cast_fp16 = layer_norm(axes = query_5_axes_0, beta = module_layers_2_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_self_att_weight_to_fp16, x = input_135_cast_fp16)[name = string("query_5_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522858]> module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38668928))), indices_nonzero_data = tensor<uint6, [522858]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38800064))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39192320))))[name = string("module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_2_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_2_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37648384)))];
+            tensor<fp16, [1024]> module_layers_2_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_2_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37650496)))];
+            tensor<fp16, [1, 128, 1024]> query_5_cast_fp16 = layer_norm(axes = query_5_axes_0, beta = module_layers_2_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_self_att_weight_to_fp16, x = input_135_cast_fp16)[name = string("query_5_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522858]> module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37652608))), indices_nonzero_data = tensor<uint6, [522858]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37783744))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38176000))))[name = string("module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_21_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_5_cast_fp16)[name = string("linear_21_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_21_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_5_cast_fp16)[name = string("linear_21_cast_fp16")];
             tensor<int32, [4]> var_575 = const()[name = string("op_575"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_13_cast_fp16 = reshape(shape = var_575, x = linear_21_cast_fp16)[name = string("q_13_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521195]> module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39208768))), indices_nonzero_data = tensor<uint6, [521195]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39339904))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39730880))))[name = string("module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> q_13_cast_fp16 = reshape(shape = var_575, x = linear_21_cast_fp16)[name = string("q_13_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521195]> module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38192448))), indices_nonzero_data = tensor<uint6, [521195]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38323584))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38714560))))[name = string("module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_22_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_5_cast_fp16)[name = string("linear_22_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_22_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_5_cast_fp16)[name = string("linear_22_cast_fp16")];
             tensor<int32, [4]> var_579 = const()[name = string("op_579"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_9_cast_fp16 = reshape(shape = var_579, x = linear_22_cast_fp16)[name = string("k_9_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521755]> module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39747328))), indices_nonzero_data = tensor<uint6, [521755]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39878464))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40269888))))[name = string("module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> k_9_cast_fp16 = reshape(shape = var_579, x = linear_22_cast_fp16)[name = string("k_9_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521755]> module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38731008))), indices_nonzero_data = tensor<uint6, [521755]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38862144))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39253568))))[name = string("module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_23_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_5_cast_fp16)[name = string("linear_23_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_23_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_5_cast_fp16)[name = string("linear_23_cast_fp16")];
             tensor<int32, [4]> var_583 = const()[name = string("op_583"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_5_cast_fp16 = reshape(shape = var_583, x = linear_23_cast_fp16)[name = string("v_5_cast_fp16")];
+            tensor<fp16, [1, 128, 8, 128]> v_5_cast_fp16 = reshape(shape = var_583, x = linear_23_cast_fp16)[name = string("v_5_cast_fp16")];
             tensor<int32, [4]> value_7_perm_0 = const()[name = string("value_7_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_2_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_2_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40286336)))];
-            tensor<fp16, [1, 188, 8, 128]> var_595_cast_fp16 = add(x = q_13_cast_fp16, y = module_layers_2_self_attn_pos_bias_u_to_fp16)[name = string("op_595_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_2_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_2_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40288448)))];
-            tensor<fp16, [1, 188, 8, 128]> var_597_cast_fp16 = add(x = q_13_cast_fp16, y = module_layers_2_self_attn_pos_bias_v_to_fp16)[name = string("op_597_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_2_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_2_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39270016)))];
+            tensor<fp16, [1, 128, 8, 128]> var_595_cast_fp16 = add(x = q_13_cast_fp16, y = module_layers_2_self_attn_pos_bias_u_to_fp16)[name = string("op_595_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_2_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_2_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39272128)))];
+            tensor<fp16, [1, 128, 8, 128]> var_597_cast_fp16 = add(x = q_13_cast_fp16, y = module_layers_2_self_attn_pos_bias_v_to_fp16)[name = string("op_597_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_5_perm_0 = const()[name = string("q_with_bias_v_5_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_51_transpose_x_0 = const()[name = string("x_51_transpose_x_0"), val = bool(false)];
             bool x_51_transpose_y_0 = const()[name = string("x_51_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_599_to_fp16 = const()[name = string("op_599_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40290560)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_5_cast_fp16 = transpose(perm = q_with_bias_v_5_perm_0, x = var_597_cast_fp16)[name = string("transpose_298")];
-            tensor<fp16, [1, 8, 188, 375]> x_51_cast_fp16 = matmul(transpose_x = x_51_transpose_x_0, transpose_y = x_51_transpose_y_0, x = q_with_bias_v_5_cast_fp16, y = var_599_to_fp16)[name = string("x_51_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 255]> var_599_to_fp16 = const()[name = string("op_599_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(308466432)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_5_cast_fp16 = transpose(perm = q_with_bias_v_5_perm_0, x = var_597_cast_fp16)[name = string("transpose_298")];
+            tensor<fp16, [1, 8, 128, 255]> x_51_cast_fp16 = matmul(transpose_x = x_51_transpose_x_0, transpose_y = x_51_transpose_y_0, x = q_with_bias_v_5_cast_fp16, y = var_599_to_fp16)[name = string("x_51_cast_fp16")];
             tensor<int32, [8]> x_53_pad_0 = const()[name = string("x_53_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_53_mode_0 = const()[name = string("x_53_mode_0"), val = string("constant")];
             fp16 const_34_to_fp16 = const()[name = string("const_34_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_53_cast_fp16 = pad(constant_val = const_34_to_fp16, mode = x_53_mode_0, pad = x_53_pad_0, x = x_51_cast_fp16)[name = string("x_53_cast_fp16")];
-            tensor<int32, [4]> var_607 = const()[name = string("op_607"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_55_cast_fp16 = reshape(shape = var_607, x = x_53_cast_fp16)[name = string("x_55_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 256]> x_53_cast_fp16 = pad(constant_val = const_34_to_fp16, mode = x_53_mode_0, pad = x_53_pad_0, x = x_51_cast_fp16)[name = string("x_53_cast_fp16")];
+            tensor<int32, [4]> var_607 = const()[name = string("op_607"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_55_cast_fp16 = reshape(shape = var_607, x = x_53_cast_fp16)[name = string("x_55_cast_fp16")];
             tensor<int32, [4]> var_611_begin_0 = const()[name = string("op_611_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_611_end_0 = const()[name = string("op_611_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_611_end_0 = const()[name = string("op_611_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
             tensor<bool, [4]> var_611_end_mask_0 = const()[name = string("op_611_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_611_cast_fp16 = slice_by_index(begin = var_611_begin_0, end = var_611_end_0, end_mask = var_611_end_mask_0, x = x_55_cast_fp16)[name = string("op_611_cast_fp16")];
-            tensor<int32, [4]> var_612 = const()[name = string("op_612"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_9_cast_fp16 = reshape(shape = var_612, x = var_611_cast_fp16)[name = string("matrix_bd_9_cast_fp16")];
+            tensor<fp16, [1, 8, 255, 128]> var_611_cast_fp16 = slice_by_index(begin = var_611_begin_0, end = var_611_end_0, end_mask = var_611_end_mask_0, x = x_55_cast_fp16)[name = string("op_611_cast_fp16")];
+            tensor<int32, [4]> var_612 = const()[name = string("op_612"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_9_cast_fp16 = reshape(shape = var_612, x = var_611_cast_fp16)[name = string("matrix_bd_9_cast_fp16")];
             bool matrix_ac_5_transpose_x_0 = const()[name = string("matrix_ac_5_transpose_x_0"), val = bool(false)];
             bool matrix_ac_5_transpose_y_0 = const()[name = string("matrix_ac_5_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_100_perm_0 = const()[name = string("transpose_100_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_101_perm_0 = const()[name = string("transpose_101_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_101 = transpose(perm = transpose_101_perm_0, x = k_9_cast_fp16)[name = string("transpose_296")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_100 = transpose(perm = transpose_100_perm_0, x = var_595_cast_fp16)[name = string("transpose_297")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_5_cast_fp16 = matmul(transpose_x = matrix_ac_5_transpose_x_0, transpose_y = matrix_ac_5_transpose_y_0, x = transpose_100, y = transpose_101)[name = string("matrix_ac_5_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_101 = transpose(perm = transpose_101_perm_0, x = k_9_cast_fp16)[name = string("transpose_296")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_100 = transpose(perm = transpose_100_perm_0, x = var_595_cast_fp16)[name = string("transpose_297")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_5_cast_fp16 = matmul(transpose_x = matrix_ac_5_transpose_x_0, transpose_y = matrix_ac_5_transpose_y_0, x = transpose_100, y = transpose_101)[name = string("matrix_ac_5_cast_fp16")];
             tensor<int32, [4]> matrix_bd_11_begin_0 = const()[name = string("matrix_bd_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_11_end_0 = const()[name = string("matrix_bd_11_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_11_end_0 = const()[name = string("matrix_bd_11_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
             tensor<bool, [4]> matrix_bd_11_end_mask_0 = const()[name = string("matrix_bd_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_11_cast_fp16 = slice_by_index(begin = matrix_bd_11_begin_0, end = matrix_bd_11_end_0, end_mask = matrix_bd_11_end_mask_0, x = matrix_bd_9_cast_fp16)[name = string("matrix_bd_11_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_621_cast_fp16 = add(x = matrix_ac_5_cast_fp16, y = matrix_bd_11_cast_fp16)[name = string("op_621_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_11_cast_fp16 = slice_by_index(begin = matrix_bd_11_begin_0, end = matrix_bd_11_end_0, end_mask = matrix_bd_11_end_mask_0, x = matrix_bd_9_cast_fp16)[name = string("matrix_bd_11_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_621_cast_fp16 = add(x = matrix_ac_5_cast_fp16, y = matrix_bd_11_cast_fp16)[name = string("op_621_cast_fp16")];
             fp16 _inversed_scores_9_y_0_to_fp16 = const()[name = string("_inversed_scores_9_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_9_cast_fp16 = mul(x = var_621_cast_fp16, y = _inversed_scores_9_y_0_to_fp16)[name = string("_inversed_scores_9_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_11_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_9_cast_fp16, cond = mask_3)[name = string("scores_11_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_627_cast_fp16 = softmax(axis = var_30, x = scores_11_cast_fp16)[name = string("op_627_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_137_cast_fp16 = select(a = var_11_to_fp16, b = var_627_cast_fp16, cond = mask_3)[name = string("input_137_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_9_cast_fp16 = mul(x = var_621_cast_fp16, y = _inversed_scores_9_y_0_to_fp16)[name = string("_inversed_scores_9_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_11_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_9_cast_fp16, cond = mask_3)[name = string("scores_11_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_627_cast_fp16 = softmax(axis = var_30, x = scores_11_cast_fp16)[name = string("op_627_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_137_cast_fp16 = select(a = var_11_to_fp16, b = var_627_cast_fp16, cond = mask_3)[name = string("input_137_cast_fp16")];
             bool x_57_transpose_x_0 = const()[name = string("x_57_transpose_x_0"), val = bool(false)];
             bool x_57_transpose_y_0 = const()[name = string("x_57_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_7_cast_fp16 = transpose(perm = value_7_perm_0, x = v_5_cast_fp16)[name = string("transpose_295")];
-            tensor<fp16, [1, 8, 188, 128]> x_57_cast_fp16 = matmul(transpose_x = x_57_transpose_x_0, transpose_y = x_57_transpose_y_0, x = input_137_cast_fp16, y = value_7_cast_fp16)[name = string("x_57_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> value_7_cast_fp16 = transpose(perm = value_7_perm_0, x = v_5_cast_fp16)[name = string("transpose_295")];
+            tensor<fp16, [1, 8, 128, 128]> x_57_cast_fp16 = matmul(transpose_x = x_57_transpose_x_0, transpose_y = x_57_transpose_y_0, x = input_137_cast_fp16, y = value_7_cast_fp16)[name = string("x_57_cast_fp16")];
             tensor<int32, [4]> var_631_perm_0 = const()[name = string("op_631_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_632 = const()[name = string("op_632"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_631_cast_fp16 = transpose(perm = var_631_perm_0, x = x_57_cast_fp16)[name = string("transpose_294")];
-            tensor<fp16, [1, 188, 1024]> input_139_cast_fp16 = reshape(shape = var_632, x = var_631_cast_fp16)[name = string("input_139_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525056]> module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41058624))), indices_nonzero_data = tensor<uint6, [525056]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41189760))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41583616))))[name = string("module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> var_631_cast_fp16 = transpose(perm = var_631_perm_0, x = x_57_cast_fp16)[name = string("transpose_294")];
+            tensor<fp16, [1, 128, 1024]> input_139_cast_fp16 = reshape(shape = var_632, x = var_631_cast_fp16)[name = string("input_139_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525056]> module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39534400))), indices_nonzero_data = tensor<uint6, [525056]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39665536))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40059392))))[name = string("module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_25_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_139_cast_fp16)[name = string("linear_25_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_143_cast_fp16 = add(x = input_135_cast_fp16, y = linear_25_cast_fp16)[name = string("input_143_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_25_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_139_cast_fp16)[name = string("linear_25_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_143_cast_fp16 = add(x = input_135_cast_fp16, y = linear_25_cast_fp16)[name = string("input_143_cast_fp16")];
             tensor<int32, [1]> x_61_axes_0 = const()[name = string("x_61_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_2_norm_conv_weight_to_fp16 = const()[name = string("module_layers_2_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41600064)))];
-            tensor<fp16, [1024]> module_layers_2_norm_conv_bias_to_fp16 = const()[name = string("module_layers_2_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41602176)))];
-            tensor<fp16, [1, 188, 1024]> x_61_cast_fp16 = layer_norm(axes = x_61_axes_0, beta = module_layers_2_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_conv_weight_to_fp16, x = input_143_cast_fp16)[name = string("x_61_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_2_norm_conv_weight_to_fp16 = const()[name = string("module_layers_2_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40075840)))];
+            tensor<fp16, [1024]> module_layers_2_norm_conv_bias_to_fp16 = const()[name = string("module_layers_2_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40077952)))];
+            tensor<fp16, [1, 128, 1024]> x_61_cast_fp16 = layer_norm(axes = x_61_axes_0, beta = module_layers_2_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_conv_weight_to_fp16, x = input_143_cast_fp16)[name = string("x_61_cast_fp16")];
             tensor<int32, [3]> input_145_perm_0 = const()[name = string("input_145_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_147_pad_type_0 = const()[name = string("input_147_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_147_strides_0 = const()[name = string("input_147_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_147_pad_0 = const()[name = string("input_147_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_147_dilations_0 = const()[name = string("input_147_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_147_groups_0 = const()[name = string("input_147_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1042766]> module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41604288))), indices_nonzero_data = tensor<uint6, [1042766]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41866496))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42648640))))[name = string("module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1042766]> module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40080064))), indices_nonzero_data = tensor<uint6, [1042766]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40342272))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41124416))))[name = string("module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_145_cast_fp16 = transpose(perm = input_145_perm_0, x = x_61_cast_fp16)[name = string("transpose_293")];
-            tensor<fp16, [1, 2048, 188]> input_147_cast_fp16 = conv(dilations = input_147_dilations_0, groups = input_147_groups_0, pad = input_147_pad_0, pad_type = input_147_pad_type_0, strides = input_147_strides_0, weight = module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_145_cast_fp16)[name = string("input_147_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_145_cast_fp16 = transpose(perm = input_145_perm_0, x = x_61_cast_fp16)[name = string("transpose_293")];
+            tensor<fp16, [1, 2048, 128]> input_147_cast_fp16 = conv(dilations = input_147_dilations_0, groups = input_147_groups_0, pad = input_147_pad_0, pad_type = input_147_pad_type_0, strides = input_147_strides_0, weight = module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_145_cast_fp16)[name = string("input_147_cast_fp16")];
             int32 x_63_split_num_splits_0 = const()[name = string("x_63_split_num_splits_0"), val = int32(2)];
             int32 x_63_split_axis_0 = const()[name = string("x_63_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_63_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_63_split_cast_fp16_1 = split(axis = x_63_split_axis_0, num_splits = x_63_split_num_splits_0, x = input_147_cast_fp16)[name = string("x_63_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_63_split_1_sigmoid_cast_fp16 = sigmoid(x = x_63_split_cast_fp16_1)[name = string("x_63_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_63_cast_fp16 = mul(x = x_63_split_cast_fp16_0, y = x_63_split_1_sigmoid_cast_fp16)[name = string("x_63_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_149_cast_fp16 = select(a = var_11_to_fp16, b = x_63_cast_fp16, cond = var_328)[name = string("input_149_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_63_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_63_split_cast_fp16_1 = split(axis = x_63_split_axis_0, num_splits = x_63_split_num_splits_0, x = input_147_cast_fp16)[name = string("x_63_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_63_split_1_sigmoid_cast_fp16 = sigmoid(x = x_63_split_cast_fp16_1)[name = string("x_63_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_63_cast_fp16 = mul(x = x_63_split_cast_fp16_0, y = x_63_split_1_sigmoid_cast_fp16)[name = string("x_63_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_149_cast_fp16 = select(a = var_11_to_fp16, b = x_63_cast_fp16, cond = var_328)[name = string("input_149_cast_fp16")];
             tensor<int32, [6]> input_151_pad_0 = const()[name = string("input_151_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_151_mode_0 = const()[name = string("input_151_mode_0"), val = string("constant")];
             fp16 const_37_to_fp16 = const()[name = string("const_37_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_151_cast_fp16 = pad(constant_val = const_37_to_fp16, mode = input_151_mode_0, pad = input_151_pad_0, x = input_149_cast_fp16)[name = string("input_151_cast_fp16")];
+            tensor<fp16, [1, 1024, 136]> input_151_cast_fp16 = pad(constant_val = const_37_to_fp16, mode = input_151_mode_0, pad = input_151_pad_0, x = input_149_cast_fp16)[name = string("input_151_cast_fp16")];
             string input_153_pad_type_0 = const()[name = string("input_153_pad_type_0"), val = string("valid")];
             int32 input_153_groups_0 = const()[name = string("input_153_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_153_strides_0 = const()[name = string("input_153_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_153_pad_0 = const()[name = string("input_153_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_153_dilations_0 = const()[name = string("input_153_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_252_to_fp16 = const()[name = string("const_252_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42681472)))];
-            tensor<fp16, [1024]> const_253_to_fp16 = const()[name = string("const_253_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42699968)))];
-            tensor<fp16, [1, 1024, 188]> input_155_cast_fp16 = conv(bias = const_253_to_fp16, dilations = input_153_dilations_0, groups = input_153_groups_0, pad = input_153_pad_0, pad_type = input_153_pad_type_0, strides = input_153_strides_0, weight = const_252_to_fp16, x = input_151_cast_fp16)[name = string("input_155_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_157_cast_fp16 = silu(x = input_155_cast_fp16)[name = string("input_157_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_252_to_fp16 = const()[name = string("const_252_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41157248)))];
+            tensor<fp16, [1024]> const_253_to_fp16 = const()[name = string("const_253_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41175744)))];
+            tensor<fp16, [1, 1024, 128]> input_155_cast_fp16 = conv(bias = const_253_to_fp16, dilations = input_153_dilations_0, groups = input_153_groups_0, pad = input_153_pad_0, pad_type = input_153_pad_type_0, strides = input_153_strides_0, weight = const_252_to_fp16, x = input_151_cast_fp16)[name = string("input_155_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_157_cast_fp16 = silu(x = input_155_cast_fp16)[name = string("input_157_cast_fp16")];
             string x_65_pad_type_0 = const()[name = string("x_65_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_65_strides_0 = const()[name = string("x_65_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_65_pad_0 = const()[name = string("x_65_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_65_dilations_0 = const()[name = string("x_65_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_65_groups_0 = const()[name = string("x_65_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519385]> module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42702080))), indices_nonzero_data = tensor<uint6, [519385]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42833216))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43222848))))[name = string("module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519385]> module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41177856))), indices_nonzero_data = tensor<uint6, [519385]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41308992))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41698624))))[name = string("module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_65_cast_fp16 = conv(dilations = x_65_dilations_0, groups = x_65_groups_0, pad = x_65_pad_0, pad_type = x_65_pad_type_0, strides = x_65_strides_0, weight = module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_157_cast_fp16)[name = string("x_65_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_65_cast_fp16 = conv(dilations = x_65_dilations_0, groups = x_65_groups_0, pad = x_65_pad_0, pad_type = x_65_pad_type_0, strides = x_65_strides_0, weight = module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_157_cast_fp16)[name = string("x_65_cast_fp16")];
             tensor<int32, [3]> input_159_perm_0 = const()[name = string("input_159_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_159_cast_fp16 = transpose(perm = input_159_perm_0, x = x_65_cast_fp16)[name = string("transpose_292")];
-            tensor<fp16, [1, 188, 1024]> input_161_cast_fp16 = add(x = input_143_cast_fp16, y = input_159_cast_fp16)[name = string("input_161_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_159_cast_fp16 = transpose(perm = input_159_perm_0, x = x_65_cast_fp16)[name = string("transpose_292")];
+            tensor<fp16, [1, 128, 1024]> input_161_cast_fp16 = add(x = input_143_cast_fp16, y = input_159_cast_fp16)[name = string("input_161_cast_fp16")];
             tensor<int32, [1]> input_163_axes_0 = const()[name = string("input_163_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_2_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_2_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43239296)))];
-            tensor<fp16, [1024]> module_layers_2_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_2_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43241408)))];
-            tensor<fp16, [1, 188, 1024]> input_163_cast_fp16 = layer_norm(axes = input_163_axes_0, beta = module_layers_2_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_feed_forward2_weight_to_fp16, x = input_161_cast_fp16)[name = string("input_163_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2065499]> module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43243520))), indices_nonzero_data = tensor<uint6, [2065499]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43767872))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45317120))))[name = string("module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_2_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_2_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41715072)))];
+            tensor<fp16, [1024]> module_layers_2_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_2_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41717184)))];
+            tensor<fp16, [1, 128, 1024]> input_163_cast_fp16 = layer_norm(axes = input_163_axes_0, beta = module_layers_2_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_feed_forward2_weight_to_fp16, x = input_161_cast_fp16)[name = string("input_163_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2065499]> module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41719296))), indices_nonzero_data = tensor<uint6, [2065499]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42243648))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43792896))))[name = string("module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_26_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_163_cast_fp16)[name = string("linear_26_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_167_cast_fp16 = silu(x = linear_26_cast_fp16)[name = string("input_167_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1965997]> module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45382720))), indices_nonzero_data = tensor<uint6, [1965997]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45907072))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47381696))))[name = string("module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_26_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_163_cast_fp16)[name = string("linear_26_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_167_cast_fp16 = silu(x = linear_26_cast_fp16)[name = string("input_167_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1965997]> module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43858496))), indices_nonzero_data = tensor<uint6, [1965997]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44382848))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45857472))))[name = string("module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_27_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_167_cast_fp16)[name = string("linear_27_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_27_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_167_cast_fp16)[name = string("linear_27_cast_fp16")];
             fp16 var_692_to_fp16 = const()[name = string("op_692_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_693_cast_fp16 = mul(x = linear_27_cast_fp16, y = var_692_to_fp16)[name = string("op_693_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_173_cast_fp16 = add(x = input_161_cast_fp16, y = var_693_cast_fp16)[name = string("input_173_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_693_cast_fp16 = mul(x = linear_27_cast_fp16, y = var_692_to_fp16)[name = string("op_693_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_173_cast_fp16 = add(x = input_161_cast_fp16, y = var_693_cast_fp16)[name = string("input_173_cast_fp16")];
             tensor<int32, [1]> input_175_axes_0 = const()[name = string("input_175_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_2_norm_out_weight_to_fp16 = const()[name = string("module_layers_2_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47398144)))];
-            tensor<fp16, [1024]> module_layers_2_norm_out_bias_to_fp16 = const()[name = string("module_layers_2_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47400256)))];
-            tensor<fp16, [1, 188, 1024]> input_175_cast_fp16 = layer_norm(axes = input_175_axes_0, beta = module_layers_2_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_out_weight_to_fp16, x = input_173_cast_fp16)[name = string("input_175_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_2_norm_out_weight_to_fp16 = const()[name = string("module_layers_2_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45873920)))];
+            tensor<fp16, [1024]> module_layers_2_norm_out_bias_to_fp16 = const()[name = string("module_layers_2_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45876032)))];
+            tensor<fp16, [1, 128, 1024]> input_175_cast_fp16 = layer_norm(axes = input_175_axes_0, beta = module_layers_2_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_out_weight_to_fp16, x = input_173_cast_fp16)[name = string("input_175_cast_fp16")];
             tensor<int32, [1]> input_177_axes_0 = const()[name = string("input_177_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_3_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_3_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47402368)))];
-            tensor<fp16, [1024]> module_layers_3_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_3_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47404480)))];
-            tensor<fp16, [1, 188, 1024]> input_177_cast_fp16 = layer_norm(axes = input_177_axes_0, beta = module_layers_3_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_feed_forward1_weight_to_fp16, x = input_175_cast_fp16)[name = string("input_177_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2075730]> module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47406592))), indices_nonzero_data = tensor<uint6, [2075730]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47930944))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49487808))))[name = string("module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_3_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_3_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45878144)))];
+            tensor<fp16, [1024]> module_layers_3_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_3_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45880256)))];
+            tensor<fp16, [1, 128, 1024]> input_177_cast_fp16 = layer_norm(axes = input_177_axes_0, beta = module_layers_3_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_feed_forward1_weight_to_fp16, x = input_175_cast_fp16)[name = string("input_177_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2075730]> module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45882368))), indices_nonzero_data = tensor<uint6, [2075730]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46406720))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47963584))))[name = string("module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_28_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_177_cast_fp16)[name = string("linear_28_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_181_cast_fp16 = silu(x = linear_28_cast_fp16)[name = string("input_181_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1953322]> module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49553408))), indices_nonzero_data = tensor<uint6, [1953322]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50077760))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51542848))))[name = string("module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_28_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_177_cast_fp16)[name = string("linear_28_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_181_cast_fp16 = silu(x = linear_28_cast_fp16)[name = string("input_181_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1953322]> module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48029184))), indices_nonzero_data = tensor<uint6, [1953322]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48553536))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50018624))))[name = string("module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_29_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_181_cast_fp16)[name = string("linear_29_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_29_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_181_cast_fp16)[name = string("linear_29_cast_fp16")];
             fp16 var_721_to_fp16 = const()[name = string("op_721_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_722_cast_fp16 = mul(x = linear_29_cast_fp16, y = var_721_to_fp16)[name = string("op_722_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_187_cast_fp16 = add(x = input_175_cast_fp16, y = var_722_cast_fp16)[name = string("input_187_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_722_cast_fp16 = mul(x = linear_29_cast_fp16, y = var_721_to_fp16)[name = string("op_722_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_187_cast_fp16 = add(x = input_175_cast_fp16, y = var_722_cast_fp16)[name = string("input_187_cast_fp16")];
             tensor<int32, [1]> query_7_axes_0 = const()[name = string("query_7_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_3_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_3_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51559296)))];
-            tensor<fp16, [1024]> module_layers_3_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_3_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51561408)))];
-            tensor<fp16, [1, 188, 1024]> query_7_cast_fp16 = layer_norm(axes = query_7_axes_0, beta = module_layers_3_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_self_att_weight_to_fp16, x = input_187_cast_fp16)[name = string("query_7_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522877]> module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51563520))), indices_nonzero_data = tensor<uint6, [522877]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51694656))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52086912))))[name = string("module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_3_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_3_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50035072)))];
+            tensor<fp16, [1024]> module_layers_3_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_3_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50037184)))];
+            tensor<fp16, [1, 128, 1024]> query_7_cast_fp16 = layer_norm(axes = query_7_axes_0, beta = module_layers_3_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_self_att_weight_to_fp16, x = input_187_cast_fp16)[name = string("query_7_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522877]> module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50039296))), indices_nonzero_data = tensor<uint6, [522877]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50170432))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50562688))))[name = string("module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_30_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_7_cast_fp16)[name = string("linear_30_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_30_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_7_cast_fp16)[name = string("linear_30_cast_fp16")];
             tensor<int32, [4]> var_738 = const()[name = string("op_738"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_19_cast_fp16 = reshape(shape = var_738, x = linear_30_cast_fp16)[name = string("q_19_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520403]> module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52103360))), indices_nonzero_data = tensor<uint6, [520403]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52234496))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52624896))))[name = string("module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> q_19_cast_fp16 = reshape(shape = var_738, x = linear_30_cast_fp16)[name = string("q_19_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520403]> module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50579136))), indices_nonzero_data = tensor<uint6, [520403]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50710272))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51100672))))[name = string("module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_31_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_7_cast_fp16)[name = string("linear_31_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_31_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_7_cast_fp16)[name = string("linear_31_cast_fp16")];
             tensor<int32, [4]> var_742 = const()[name = string("op_742"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_13_cast_fp16 = reshape(shape = var_742, x = linear_31_cast_fp16)[name = string("k_13_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521127]> module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52641344))), indices_nonzero_data = tensor<uint6, [521127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52772480))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53163392))))[name = string("module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> k_13_cast_fp16 = reshape(shape = var_742, x = linear_31_cast_fp16)[name = string("k_13_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521127]> module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51117120))), indices_nonzero_data = tensor<uint6, [521127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51248256))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51639168))))[name = string("module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_32_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_7_cast_fp16)[name = string("linear_32_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_32_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_7_cast_fp16)[name = string("linear_32_cast_fp16")];
             tensor<int32, [4]> var_746 = const()[name = string("op_746"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_7_cast_fp16 = reshape(shape = var_746, x = linear_32_cast_fp16)[name = string("v_7_cast_fp16")];
+            tensor<fp16, [1, 128, 8, 128]> v_7_cast_fp16 = reshape(shape = var_746, x = linear_32_cast_fp16)[name = string("v_7_cast_fp16")];
             tensor<int32, [4]> value_9_perm_0 = const()[name = string("value_9_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_3_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_3_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53179840)))];
-            tensor<fp16, [1, 188, 8, 128]> var_758_cast_fp16 = add(x = q_19_cast_fp16, y = module_layers_3_self_attn_pos_bias_u_to_fp16)[name = string("op_758_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_3_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_3_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53181952)))];
-            tensor<fp16, [1, 188, 8, 128]> var_760_cast_fp16 = add(x = q_19_cast_fp16, y = module_layers_3_self_attn_pos_bias_v_to_fp16)[name = string("op_760_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_3_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_3_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51655616)))];
+            tensor<fp16, [1, 128, 8, 128]> var_758_cast_fp16 = add(x = q_19_cast_fp16, y = module_layers_3_self_attn_pos_bias_u_to_fp16)[name = string("op_758_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_3_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_3_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51657728)))];
+            tensor<fp16, [1, 128, 8, 128]> var_760_cast_fp16 = add(x = q_19_cast_fp16, y = module_layers_3_self_attn_pos_bias_v_to_fp16)[name = string("op_760_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_7_perm_0 = const()[name = string("q_with_bias_v_7_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_73_transpose_x_0 = const()[name = string("x_73_transpose_x_0"), val = bool(false)];
             bool x_73_transpose_y_0 = const()[name = string("x_73_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_762_to_fp16 = const()[name = string("op_762_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53184064)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_7_cast_fp16 = transpose(perm = q_with_bias_v_7_perm_0, x = var_760_cast_fp16)[name = string("transpose_291")];
-            tensor<fp16, [1, 8, 188, 375]> x_73_cast_fp16 = matmul(transpose_x = x_73_transpose_x_0, transpose_y = x_73_transpose_y_0, x = q_with_bias_v_7_cast_fp16, y = var_762_to_fp16)[name = string("x_73_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 255]> var_762_to_fp16 = const()[name = string("op_762_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(308988736)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_7_cast_fp16 = transpose(perm = q_with_bias_v_7_perm_0, x = var_760_cast_fp16)[name = string("transpose_291")];
+            tensor<fp16, [1, 8, 128, 255]> x_73_cast_fp16 = matmul(transpose_x = x_73_transpose_x_0, transpose_y = x_73_transpose_y_0, x = q_with_bias_v_7_cast_fp16, y = var_762_to_fp16)[name = string("x_73_cast_fp16")];
             tensor<int32, [8]> x_75_pad_0 = const()[name = string("x_75_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_75_mode_0 = const()[name = string("x_75_mode_0"), val = string("constant")];
             fp16 const_44_to_fp16 = const()[name = string("const_44_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_75_cast_fp16 = pad(constant_val = const_44_to_fp16, mode = x_75_mode_0, pad = x_75_pad_0, x = x_73_cast_fp16)[name = string("x_75_cast_fp16")];
-            tensor<int32, [4]> var_770 = const()[name = string("op_770"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_77_cast_fp16 = reshape(shape = var_770, x = x_75_cast_fp16)[name = string("x_77_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 256]> x_75_cast_fp16 = pad(constant_val = const_44_to_fp16, mode = x_75_mode_0, pad = x_75_pad_0, x = x_73_cast_fp16)[name = string("x_75_cast_fp16")];
+            tensor<int32, [4]> var_770 = const()[name = string("op_770"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_77_cast_fp16 = reshape(shape = var_770, x = x_75_cast_fp16)[name = string("x_77_cast_fp16")];
             tensor<int32, [4]> var_774_begin_0 = const()[name = string("op_774_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_774_end_0 = const()[name = string("op_774_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_774_end_0 = const()[name = string("op_774_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
             tensor<bool, [4]> var_774_end_mask_0 = const()[name = string("op_774_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_774_cast_fp16 = slice_by_index(begin = var_774_begin_0, end = var_774_end_0, end_mask = var_774_end_mask_0, x = x_77_cast_fp16)[name = string("op_774_cast_fp16")];
-            tensor<int32, [4]> var_775 = const()[name = string("op_775"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_13_cast_fp16 = reshape(shape = var_775, x = var_774_cast_fp16)[name = string("matrix_bd_13_cast_fp16")];
+            tensor<fp16, [1, 8, 255, 128]> var_774_cast_fp16 = slice_by_index(begin = var_774_begin_0, end = var_774_end_0, end_mask = var_774_end_mask_0, x = x_77_cast_fp16)[name = string("op_774_cast_fp16")];
+            tensor<int32, [4]> var_775 = const()[name = string("op_775"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_13_cast_fp16 = reshape(shape = var_775, x = var_774_cast_fp16)[name = string("matrix_bd_13_cast_fp16")];
             bool matrix_ac_7_transpose_x_0 = const()[name = string("matrix_ac_7_transpose_x_0"), val = bool(false)];
             bool matrix_ac_7_transpose_y_0 = const()[name = string("matrix_ac_7_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_102_perm_0 = const()[name = string("transpose_102_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_103_perm_0 = const()[name = string("transpose_103_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_103 = transpose(perm = transpose_103_perm_0, x = k_13_cast_fp16)[name = string("transpose_289")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_102 = transpose(perm = transpose_102_perm_0, x = var_758_cast_fp16)[name = string("transpose_290")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_7_cast_fp16 = matmul(transpose_x = matrix_ac_7_transpose_x_0, transpose_y = matrix_ac_7_transpose_y_0, x = transpose_102, y = transpose_103)[name = string("matrix_ac_7_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_103 = transpose(perm = transpose_103_perm_0, x = k_13_cast_fp16)[name = string("transpose_289")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_102 = transpose(perm = transpose_102_perm_0, x = var_758_cast_fp16)[name = string("transpose_290")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_7_cast_fp16 = matmul(transpose_x = matrix_ac_7_transpose_x_0, transpose_y = matrix_ac_7_transpose_y_0, x = transpose_102, y = transpose_103)[name = string("matrix_ac_7_cast_fp16")];
             tensor<int32, [4]> matrix_bd_15_begin_0 = const()[name = string("matrix_bd_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_15_end_0 = const()[name = string("matrix_bd_15_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_15_end_0 = const()[name = string("matrix_bd_15_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
             tensor<bool, [4]> matrix_bd_15_end_mask_0 = const()[name = string("matrix_bd_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_15_cast_fp16 = slice_by_index(begin = matrix_bd_15_begin_0, end = matrix_bd_15_end_0, end_mask = matrix_bd_15_end_mask_0, x = matrix_bd_13_cast_fp16)[name = string("matrix_bd_15_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_784_cast_fp16 = add(x = matrix_ac_7_cast_fp16, y = matrix_bd_15_cast_fp16)[name = string("op_784_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_15_cast_fp16 = slice_by_index(begin = matrix_bd_15_begin_0, end = matrix_bd_15_end_0, end_mask = matrix_bd_15_end_mask_0, x = matrix_bd_13_cast_fp16)[name = string("matrix_bd_15_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_784_cast_fp16 = add(x = matrix_ac_7_cast_fp16, y = matrix_bd_15_cast_fp16)[name = string("op_784_cast_fp16")];
             fp16 _inversed_scores_13_y_0_to_fp16 = const()[name = string("_inversed_scores_13_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_13_cast_fp16 = mul(x = var_784_cast_fp16, y = _inversed_scores_13_y_0_to_fp16)[name = string("_inversed_scores_13_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_15_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_13_cast_fp16, cond = mask_3)[name = string("scores_15_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_790_cast_fp16 = softmax(axis = var_30, x = scores_15_cast_fp16)[name = string("op_790_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_189_cast_fp16 = select(a = var_11_to_fp16, b = var_790_cast_fp16, cond = mask_3)[name = string("input_189_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_13_cast_fp16 = mul(x = var_784_cast_fp16, y = _inversed_scores_13_y_0_to_fp16)[name = string("_inversed_scores_13_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_15_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_13_cast_fp16, cond = mask_3)[name = string("scores_15_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_790_cast_fp16 = softmax(axis = var_30, x = scores_15_cast_fp16)[name = string("op_790_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_189_cast_fp16 = select(a = var_11_to_fp16, b = var_790_cast_fp16, cond = mask_3)[name = string("input_189_cast_fp16")];
             bool x_79_transpose_x_0 = const()[name = string("x_79_transpose_x_0"), val = bool(false)];
             bool x_79_transpose_y_0 = const()[name = string("x_79_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_9_cast_fp16 = transpose(perm = value_9_perm_0, x = v_7_cast_fp16)[name = string("transpose_288")];
-            tensor<fp16, [1, 8, 188, 128]> x_79_cast_fp16 = matmul(transpose_x = x_79_transpose_x_0, transpose_y = x_79_transpose_y_0, x = input_189_cast_fp16, y = value_9_cast_fp16)[name = string("x_79_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> value_9_cast_fp16 = transpose(perm = value_9_perm_0, x = v_7_cast_fp16)[name = string("transpose_288")];
+            tensor<fp16, [1, 8, 128, 128]> x_79_cast_fp16 = matmul(transpose_x = x_79_transpose_x_0, transpose_y = x_79_transpose_y_0, x = input_189_cast_fp16, y = value_9_cast_fp16)[name = string("x_79_cast_fp16")];
             tensor<int32, [4]> var_794_perm_0 = const()[name = string("op_794_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_795 = const()[name = string("op_795"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_794_cast_fp16 = transpose(perm = var_794_perm_0, x = x_79_cast_fp16)[name = string("transpose_287")];
-            tensor<fp16, [1, 188, 1024]> input_191_cast_fp16 = reshape(shape = var_795, x = var_794_cast_fp16)[name = string("input_191_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522589]> module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53952128))), indices_nonzero_data = tensor<uint6, [522589]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54083264))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54475328))))[name = string("module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> var_794_cast_fp16 = transpose(perm = var_794_perm_0, x = x_79_cast_fp16)[name = string("transpose_287")];
+            tensor<fp16, [1, 128, 1024]> input_191_cast_fp16 = reshape(shape = var_795, x = var_794_cast_fp16)[name = string("input_191_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522589]> module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51920000))), indices_nonzero_data = tensor<uint6, [522589]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52051136))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52443200))))[name = string("module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_34_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_191_cast_fp16)[name = string("linear_34_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_195_cast_fp16 = add(x = input_187_cast_fp16, y = linear_34_cast_fp16)[name = string("input_195_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_34_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_191_cast_fp16)[name = string("linear_34_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_195_cast_fp16 = add(x = input_187_cast_fp16, y = linear_34_cast_fp16)[name = string("input_195_cast_fp16")];
             tensor<int32, [1]> x_83_axes_0 = const()[name = string("x_83_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_3_norm_conv_weight_to_fp16 = const()[name = string("module_layers_3_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54491776)))];
-            tensor<fp16, [1024]> module_layers_3_norm_conv_bias_to_fp16 = const()[name = string("module_layers_3_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54493888)))];
-            tensor<fp16, [1, 188, 1024]> x_83_cast_fp16 = layer_norm(axes = x_83_axes_0, beta = module_layers_3_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_conv_weight_to_fp16, x = input_195_cast_fp16)[name = string("x_83_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_3_norm_conv_weight_to_fp16 = const()[name = string("module_layers_3_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52459648)))];
+            tensor<fp16, [1024]> module_layers_3_norm_conv_bias_to_fp16 = const()[name = string("module_layers_3_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52461760)))];
+            tensor<fp16, [1, 128, 1024]> x_83_cast_fp16 = layer_norm(axes = x_83_axes_0, beta = module_layers_3_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_conv_weight_to_fp16, x = input_195_cast_fp16)[name = string("x_83_cast_fp16")];
             tensor<int32, [3]> input_197_perm_0 = const()[name = string("input_197_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_199_pad_type_0 = const()[name = string("input_199_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_199_strides_0 = const()[name = string("input_199_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_199_pad_0 = const()[name = string("input_199_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_199_dilations_0 = const()[name = string("input_199_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_199_groups_0 = const()[name = string("input_199_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1044665]> module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54496000))), indices_nonzero_data = tensor<uint6, [1044665]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54758208))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55541824))))[name = string("module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1044665]> module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52463872))), indices_nonzero_data = tensor<uint6, [1044665]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52726080))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53509696))))[name = string("module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_197_cast_fp16 = transpose(perm = input_197_perm_0, x = x_83_cast_fp16)[name = string("transpose_286")];
-            tensor<fp16, [1, 2048, 188]> input_199_cast_fp16 = conv(dilations = input_199_dilations_0, groups = input_199_groups_0, pad = input_199_pad_0, pad_type = input_199_pad_type_0, strides = input_199_strides_0, weight = module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_197_cast_fp16)[name = string("input_199_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_197_cast_fp16 = transpose(perm = input_197_perm_0, x = x_83_cast_fp16)[name = string("transpose_286")];
+            tensor<fp16, [1, 2048, 128]> input_199_cast_fp16 = conv(dilations = input_199_dilations_0, groups = input_199_groups_0, pad = input_199_pad_0, pad_type = input_199_pad_type_0, strides = input_199_strides_0, weight = module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_197_cast_fp16)[name = string("input_199_cast_fp16")];
             int32 x_85_split_num_splits_0 = const()[name = string("x_85_split_num_splits_0"), val = int32(2)];
             int32 x_85_split_axis_0 = const()[name = string("x_85_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_85_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_85_split_cast_fp16_1 = split(axis = x_85_split_axis_0, num_splits = x_85_split_num_splits_0, x = input_199_cast_fp16)[name = string("x_85_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_85_split_1_sigmoid_cast_fp16 = sigmoid(x = x_85_split_cast_fp16_1)[name = string("x_85_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_85_cast_fp16 = mul(x = x_85_split_cast_fp16_0, y = x_85_split_1_sigmoid_cast_fp16)[name = string("x_85_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_201_cast_fp16 = select(a = var_11_to_fp16, b = x_85_cast_fp16, cond = var_328)[name = string("input_201_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_85_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_85_split_cast_fp16_1 = split(axis = x_85_split_axis_0, num_splits = x_85_split_num_splits_0, x = input_199_cast_fp16)[name = string("x_85_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_85_split_1_sigmoid_cast_fp16 = sigmoid(x = x_85_split_cast_fp16_1)[name = string("x_85_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_85_cast_fp16 = mul(x = x_85_split_cast_fp16_0, y = x_85_split_1_sigmoid_cast_fp16)[name = string("x_85_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_201_cast_fp16 = select(a = var_11_to_fp16, b = x_85_cast_fp16, cond = var_328)[name = string("input_201_cast_fp16")];
             tensor<int32, [6]> input_203_pad_0 = const()[name = string("input_203_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_203_mode_0 = const()[name = string("input_203_mode_0"), val = string("constant")];
             fp16 const_47_to_fp16 = const()[name = string("const_47_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_203_cast_fp16 = pad(constant_val = const_47_to_fp16, mode = input_203_mode_0, pad = input_203_pad_0, x = input_201_cast_fp16)[name = string("input_203_cast_fp16")];
+            tensor<fp16, [1, 1024, 136]> input_203_cast_fp16 = pad(constant_val = const_47_to_fp16, mode = input_203_mode_0, pad = input_203_pad_0, x = input_201_cast_fp16)[name = string("input_203_cast_fp16")];
             string input_205_pad_type_0 = const()[name = string("input_205_pad_type_0"), val = string("valid")];
             int32 input_205_groups_0 = const()[name = string("input_205_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_205_strides_0 = const()[name = string("input_205_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_205_pad_0 = const()[name = string("input_205_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_205_dilations_0 = const()[name = string("input_205_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_254_to_fp16 = const()[name = string("const_254_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55574656)))];
-            tensor<fp16, [1024]> const_255_to_fp16 = const()[name = string("const_255_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55593152)))];
-            tensor<fp16, [1, 1024, 188]> input_207_cast_fp16 = conv(bias = const_255_to_fp16, dilations = input_205_dilations_0, groups = input_205_groups_0, pad = input_205_pad_0, pad_type = input_205_pad_type_0, strides = input_205_strides_0, weight = const_254_to_fp16, x = input_203_cast_fp16)[name = string("input_207_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_209_cast_fp16 = silu(x = input_207_cast_fp16)[name = string("input_209_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_254_to_fp16 = const()[name = string("const_254_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53542528)))];
+            tensor<fp16, [1024]> const_255_to_fp16 = const()[name = string("const_255_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53561024)))];
+            tensor<fp16, [1, 1024, 128]> input_207_cast_fp16 = conv(bias = const_255_to_fp16, dilations = input_205_dilations_0, groups = input_205_groups_0, pad = input_205_pad_0, pad_type = input_205_pad_type_0, strides = input_205_strides_0, weight = const_254_to_fp16, x = input_203_cast_fp16)[name = string("input_207_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_209_cast_fp16 = silu(x = input_207_cast_fp16)[name = string("input_209_cast_fp16")];
             string x_87_pad_type_0 = const()[name = string("x_87_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_87_strides_0 = const()[name = string("x_87_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_87_pad_0 = const()[name = string("x_87_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_87_dilations_0 = const()[name = string("x_87_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_87_groups_0 = const()[name = string("x_87_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519219]> module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55595264))), indices_nonzero_data = tensor<uint6, [519219]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55726400))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56115904))))[name = string("module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519219]> module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53563136))), indices_nonzero_data = tensor<uint6, [519219]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53694272))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54083776))))[name = string("module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_87_cast_fp16 = conv(dilations = x_87_dilations_0, groups = x_87_groups_0, pad = x_87_pad_0, pad_type = x_87_pad_type_0, strides = x_87_strides_0, weight = module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_209_cast_fp16)[name = string("x_87_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_87_cast_fp16 = conv(dilations = x_87_dilations_0, groups = x_87_groups_0, pad = x_87_pad_0, pad_type = x_87_pad_type_0, strides = x_87_strides_0, weight = module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_209_cast_fp16)[name = string("x_87_cast_fp16")];
             tensor<int32, [3]> input_211_perm_0 = const()[name = string("input_211_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_211_cast_fp16 = transpose(perm = input_211_perm_0, x = x_87_cast_fp16)[name = string("transpose_285")];
-            tensor<fp16, [1, 188, 1024]> input_213_cast_fp16 = add(x = input_195_cast_fp16, y = input_211_cast_fp16)[name = string("input_213_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_211_cast_fp16 = transpose(perm = input_211_perm_0, x = x_87_cast_fp16)[name = string("transpose_285")];
+            tensor<fp16, [1, 128, 1024]> input_213_cast_fp16 = add(x = input_195_cast_fp16, y = input_211_cast_fp16)[name = string("input_213_cast_fp16")];
             tensor<int32, [1]> input_215_axes_0 = const()[name = string("input_215_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_3_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_3_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56132352)))];
-            tensor<fp16, [1024]> module_layers_3_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_3_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56134464)))];
-            tensor<fp16, [1, 188, 1024]> input_215_cast_fp16 = layer_norm(axes = input_215_axes_0, beta = module_layers_3_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_feed_forward2_weight_to_fp16, x = input_213_cast_fp16)[name = string("input_215_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2059316]> module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56136576))), indices_nonzero_data = tensor<uint6, [2059316]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56660928))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58205504))))[name = string("module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_3_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_3_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54100224)))];
+            tensor<fp16, [1024]> module_layers_3_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_3_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54102336)))];
+            tensor<fp16, [1, 128, 1024]> input_215_cast_fp16 = layer_norm(axes = input_215_axes_0, beta = module_layers_3_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_feed_forward2_weight_to_fp16, x = input_213_cast_fp16)[name = string("input_215_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2059316]> module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54104448))), indices_nonzero_data = tensor<uint6, [2059316]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54628800))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56173376))))[name = string("module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_35_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_215_cast_fp16)[name = string("linear_35_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_219_cast_fp16 = silu(x = linear_35_cast_fp16)[name = string("input_219_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1938642]> module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58271104))), indices_nonzero_data = tensor<uint6, [1938642]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58795456))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60249536))))[name = string("module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_35_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_215_cast_fp16)[name = string("linear_35_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_219_cast_fp16 = silu(x = linear_35_cast_fp16)[name = string("input_219_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1938642]> module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56238976))), indices_nonzero_data = tensor<uint6, [1938642]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56763328))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58217408))))[name = string("module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_36_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_219_cast_fp16)[name = string("linear_36_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_36_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_219_cast_fp16)[name = string("linear_36_cast_fp16")];
             fp16 var_855_to_fp16 = const()[name = string("op_855_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_856_cast_fp16 = mul(x = linear_36_cast_fp16, y = var_855_to_fp16)[name = string("op_856_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_225_cast_fp16 = add(x = input_213_cast_fp16, y = var_856_cast_fp16)[name = string("input_225_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_856_cast_fp16 = mul(x = linear_36_cast_fp16, y = var_855_to_fp16)[name = string("op_856_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_225_cast_fp16 = add(x = input_213_cast_fp16, y = var_856_cast_fp16)[name = string("input_225_cast_fp16")];
             tensor<int32, [1]> input_227_axes_0 = const()[name = string("input_227_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_3_norm_out_weight_to_fp16 = const()[name = string("module_layers_3_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60265984)))];
-            tensor<fp16, [1024]> module_layers_3_norm_out_bias_to_fp16 = const()[name = string("module_layers_3_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60268096)))];
-            tensor<fp16, [1, 188, 1024]> input_227_cast_fp16 = layer_norm(axes = input_227_axes_0, beta = module_layers_3_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_out_weight_to_fp16, x = input_225_cast_fp16)[name = string("input_227_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_3_norm_out_weight_to_fp16 = const()[name = string("module_layers_3_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58233856)))];
+            tensor<fp16, [1024]> module_layers_3_norm_out_bias_to_fp16 = const()[name = string("module_layers_3_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58235968)))];
+            tensor<fp16, [1, 128, 1024]> input_227_cast_fp16 = layer_norm(axes = input_227_axes_0, beta = module_layers_3_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_out_weight_to_fp16, x = input_225_cast_fp16)[name = string("input_227_cast_fp16")];
             tensor<int32, [1]> input_229_axes_0 = const()[name = string("input_229_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_4_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_4_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60270208)))];
-            tensor<fp16, [1024]> module_layers_4_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_4_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60272320)))];
-            tensor<fp16, [1, 188, 1024]> input_229_cast_fp16 = layer_norm(axes = input_229_axes_0, beta = module_layers_4_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_feed_forward1_weight_to_fp16, x = input_227_cast_fp16)[name = string("input_229_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2071628]> module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60274432))), indices_nonzero_data = tensor<uint6, [2071628]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60798784))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62352576))))[name = string("module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_4_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_4_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58238080)))];
+            tensor<fp16, [1024]> module_layers_4_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_4_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58240192)))];
+            tensor<fp16, [1, 128, 1024]> input_229_cast_fp16 = layer_norm(axes = input_229_axes_0, beta = module_layers_4_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_feed_forward1_weight_to_fp16, x = input_227_cast_fp16)[name = string("input_229_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2071628]> module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58242304))), indices_nonzero_data = tensor<uint6, [2071628]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58766656))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60320448))))[name = string("module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_37_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_229_cast_fp16)[name = string("linear_37_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_233_cast_fp16 = silu(x = linear_37_cast_fp16)[name = string("input_233_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1939129]> module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62418176))), indices_nonzero_data = tensor<uint6, [1939129]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62942528))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64396992))))[name = string("module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_37_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_229_cast_fp16)[name = string("linear_37_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_233_cast_fp16 = silu(x = linear_37_cast_fp16)[name = string("input_233_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1939129]> module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60386048))), indices_nonzero_data = tensor<uint6, [1939129]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60910400))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62364864))))[name = string("module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_38_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_233_cast_fp16)[name = string("linear_38_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_38_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_233_cast_fp16)[name = string("linear_38_cast_fp16")];
             fp16 var_884_to_fp16 = const()[name = string("op_884_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_885_cast_fp16 = mul(x = linear_38_cast_fp16, y = var_884_to_fp16)[name = string("op_885_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_239_cast_fp16 = add(x = input_227_cast_fp16, y = var_885_cast_fp16)[name = string("input_239_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_885_cast_fp16 = mul(x = linear_38_cast_fp16, y = var_884_to_fp16)[name = string("op_885_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_239_cast_fp16 = add(x = input_227_cast_fp16, y = var_885_cast_fp16)[name = string("input_239_cast_fp16")];
             tensor<int32, [1]> query_9_axes_0 = const()[name = string("query_9_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_4_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_4_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64413440)))];
-            tensor<fp16, [1024]> module_layers_4_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_4_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64415552)))];
-            tensor<fp16, [1, 188, 1024]> query_9_cast_fp16 = layer_norm(axes = query_9_axes_0, beta = module_layers_4_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_self_att_weight_to_fp16, x = input_239_cast_fp16)[name = string("query_9_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523421]> module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64417664))), indices_nonzero_data = tensor<uint6, [523421]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64548800))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64941440))))[name = string("module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_4_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_4_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62381312)))];
+            tensor<fp16, [1024]> module_layers_4_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_4_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62383424)))];
+            tensor<fp16, [1, 128, 1024]> query_9_cast_fp16 = layer_norm(axes = query_9_axes_0, beta = module_layers_4_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_self_att_weight_to_fp16, x = input_239_cast_fp16)[name = string("query_9_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523421]> module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62385536))), indices_nonzero_data = tensor<uint6, [523421]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62516672))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62909312))))[name = string("module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_39_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_9_cast_fp16)[name = string("linear_39_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_39_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_9_cast_fp16)[name = string("linear_39_cast_fp16")];
             tensor<int32, [4]> var_901 = const()[name = string("op_901"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_25_cast_fp16 = reshape(shape = var_901, x = linear_39_cast_fp16)[name = string("q_25_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521081]> module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64957888))), indices_nonzero_data = tensor<uint6, [521081]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65089024))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65479936))))[name = string("module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> q_25_cast_fp16 = reshape(shape = var_901, x = linear_39_cast_fp16)[name = string("q_25_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521081]> module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62925760))), indices_nonzero_data = tensor<uint6, [521081]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63056896))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63447808))))[name = string("module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_40_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_9_cast_fp16)[name = string("linear_40_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_40_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_9_cast_fp16)[name = string("linear_40_cast_fp16")];
             tensor<int32, [4]> var_905 = const()[name = string("op_905"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_17_cast_fp16 = reshape(shape = var_905, x = linear_40_cast_fp16)[name = string("k_17_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520224]> module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65496384))), indices_nonzero_data = tensor<uint6, [520224]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65627520))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66017792))))[name = string("module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> k_17_cast_fp16 = reshape(shape = var_905, x = linear_40_cast_fp16)[name = string("k_17_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520224]> module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63464256))), indices_nonzero_data = tensor<uint6, [520224]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63595392))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63985664))))[name = string("module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_41_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_9_cast_fp16)[name = string("linear_41_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_41_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_9_cast_fp16)[name = string("linear_41_cast_fp16")];
             tensor<int32, [4]> var_909 = const()[name = string("op_909"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_9_cast_fp16 = reshape(shape = var_909, x = linear_41_cast_fp16)[name = string("v_9_cast_fp16")];
+            tensor<fp16, [1, 128, 8, 128]> v_9_cast_fp16 = reshape(shape = var_909, x = linear_41_cast_fp16)[name = string("v_9_cast_fp16")];
             tensor<int32, [4]> value_11_perm_0 = const()[name = string("value_11_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_4_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_4_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66034240)))];
-            tensor<fp16, [1, 188, 8, 128]> var_921_cast_fp16 = add(x = q_25_cast_fp16, y = module_layers_4_self_attn_pos_bias_u_to_fp16)[name = string("op_921_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_4_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_4_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66036352)))];
-            tensor<fp16, [1, 188, 8, 128]> var_923_cast_fp16 = add(x = q_25_cast_fp16, y = module_layers_4_self_attn_pos_bias_v_to_fp16)[name = string("op_923_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_4_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_4_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64002112)))];
+            tensor<fp16, [1, 128, 8, 128]> var_921_cast_fp16 = add(x = q_25_cast_fp16, y = module_layers_4_self_attn_pos_bias_u_to_fp16)[name = string("op_921_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_4_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_4_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64004224)))];
+            tensor<fp16, [1, 128, 8, 128]> var_923_cast_fp16 = add(x = q_25_cast_fp16, y = module_layers_4_self_attn_pos_bias_v_to_fp16)[name = string("op_923_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_9_perm_0 = const()[name = string("q_with_bias_v_9_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_95_transpose_x_0 = const()[name = string("x_95_transpose_x_0"), val = bool(false)];
             bool x_95_transpose_y_0 = const()[name = string("x_95_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_925_to_fp16 = const()[name = string("op_925_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66038464)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_9_cast_fp16 = transpose(perm = q_with_bias_v_9_perm_0, x = var_923_cast_fp16)[name = string("transpose_284")];
-            tensor<fp16, [1, 8, 188, 375]> x_95_cast_fp16 = matmul(transpose_x = x_95_transpose_x_0, transpose_y = x_95_transpose_y_0, x = q_with_bias_v_9_cast_fp16, y = var_925_to_fp16)[name = string("x_95_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 255]> var_925_to_fp16 = const()[name = string("op_925_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(309511040)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_9_cast_fp16 = transpose(perm = q_with_bias_v_9_perm_0, x = var_923_cast_fp16)[name = string("transpose_284")];
+            tensor<fp16, [1, 8, 128, 255]> x_95_cast_fp16 = matmul(transpose_x = x_95_transpose_x_0, transpose_y = x_95_transpose_y_0, x = q_with_bias_v_9_cast_fp16, y = var_925_to_fp16)[name = string("x_95_cast_fp16")];
             tensor<int32, [8]> x_97_pad_0 = const()[name = string("x_97_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_97_mode_0 = const()[name = string("x_97_mode_0"), val = string("constant")];
             fp16 const_54_to_fp16 = const()[name = string("const_54_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_97_cast_fp16 = pad(constant_val = const_54_to_fp16, mode = x_97_mode_0, pad = x_97_pad_0, x = x_95_cast_fp16)[name = string("x_97_cast_fp16")];
-            tensor<int32, [4]> var_933 = const()[name = string("op_933"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_99_cast_fp16 = reshape(shape = var_933, x = x_97_cast_fp16)[name = string("x_99_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 256]> x_97_cast_fp16 = pad(constant_val = const_54_to_fp16, mode = x_97_mode_0, pad = x_97_pad_0, x = x_95_cast_fp16)[name = string("x_97_cast_fp16")];
+            tensor<int32, [4]> var_933 = const()[name = string("op_933"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_99_cast_fp16 = reshape(shape = var_933, x = x_97_cast_fp16)[name = string("x_99_cast_fp16")];
             tensor<int32, [4]> var_937_begin_0 = const()[name = string("op_937_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_937_end_0 = const()[name = string("op_937_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_937_end_0 = const()[name = string("op_937_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
             tensor<bool, [4]> var_937_end_mask_0 = const()[name = string("op_937_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_937_cast_fp16 = slice_by_index(begin = var_937_begin_0, end = var_937_end_0, end_mask = var_937_end_mask_0, x = x_99_cast_fp16)[name = string("op_937_cast_fp16")];
-            tensor<int32, [4]> var_938 = const()[name = string("op_938"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_17_cast_fp16 = reshape(shape = var_938, x = var_937_cast_fp16)[name = string("matrix_bd_17_cast_fp16")];
+            tensor<fp16, [1, 8, 255, 128]> var_937_cast_fp16 = slice_by_index(begin = var_937_begin_0, end = var_937_end_0, end_mask = var_937_end_mask_0, x = x_99_cast_fp16)[name = string("op_937_cast_fp16")];
+            tensor<int32, [4]> var_938 = const()[name = string("op_938"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_17_cast_fp16 = reshape(shape = var_938, x = var_937_cast_fp16)[name = string("matrix_bd_17_cast_fp16")];
             bool matrix_ac_9_transpose_x_0 = const()[name = string("matrix_ac_9_transpose_x_0"), val = bool(false)];
             bool matrix_ac_9_transpose_y_0 = const()[name = string("matrix_ac_9_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_104_perm_0 = const()[name = string("transpose_104_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_105_perm_0 = const()[name = string("transpose_105_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_105 = transpose(perm = transpose_105_perm_0, x = k_17_cast_fp16)[name = string("transpose_282")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_104 = transpose(perm = transpose_104_perm_0, x = var_921_cast_fp16)[name = string("transpose_283")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_9_cast_fp16 = matmul(transpose_x = matrix_ac_9_transpose_x_0, transpose_y = matrix_ac_9_transpose_y_0, x = transpose_104, y = transpose_105)[name = string("matrix_ac_9_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_105 = transpose(perm = transpose_105_perm_0, x = k_17_cast_fp16)[name = string("transpose_282")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_104 = transpose(perm = transpose_104_perm_0, x = var_921_cast_fp16)[name = string("transpose_283")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_9_cast_fp16 = matmul(transpose_x = matrix_ac_9_transpose_x_0, transpose_y = matrix_ac_9_transpose_y_0, x = transpose_104, y = transpose_105)[name = string("matrix_ac_9_cast_fp16")];
             tensor<int32, [4]> matrix_bd_19_begin_0 = const()[name = string("matrix_bd_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_19_end_0 = const()[name = string("matrix_bd_19_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_19_end_0 = const()[name = string("matrix_bd_19_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
             tensor<bool, [4]> matrix_bd_19_end_mask_0 = const()[name = string("matrix_bd_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_19_cast_fp16 = slice_by_index(begin = matrix_bd_19_begin_0, end = matrix_bd_19_end_0, end_mask = matrix_bd_19_end_mask_0, x = matrix_bd_17_cast_fp16)[name = string("matrix_bd_19_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_947_cast_fp16 = add(x = matrix_ac_9_cast_fp16, y = matrix_bd_19_cast_fp16)[name = string("op_947_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_19_cast_fp16 = slice_by_index(begin = matrix_bd_19_begin_0, end = matrix_bd_19_end_0, end_mask = matrix_bd_19_end_mask_0, x = matrix_bd_17_cast_fp16)[name = string("matrix_bd_19_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_947_cast_fp16 = add(x = matrix_ac_9_cast_fp16, y = matrix_bd_19_cast_fp16)[name = string("op_947_cast_fp16")];
             fp16 _inversed_scores_17_y_0_to_fp16 = const()[name = string("_inversed_scores_17_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_17_cast_fp16 = mul(x = var_947_cast_fp16, y = _inversed_scores_17_y_0_to_fp16)[name = string("_inversed_scores_17_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_19_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_17_cast_fp16, cond = mask_3)[name = string("scores_19_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_953_cast_fp16 = softmax(axis = var_30, x = scores_19_cast_fp16)[name = string("op_953_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_241_cast_fp16 = select(a = var_11_to_fp16, b = var_953_cast_fp16, cond = mask_3)[name = string("input_241_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_17_cast_fp16 = mul(x = var_947_cast_fp16, y = _inversed_scores_17_y_0_to_fp16)[name = string("_inversed_scores_17_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_19_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_17_cast_fp16, cond = mask_3)[name = string("scores_19_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_953_cast_fp16 = softmax(axis = var_30, x = scores_19_cast_fp16)[name = string("op_953_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_241_cast_fp16 = select(a = var_11_to_fp16, b = var_953_cast_fp16, cond = mask_3)[name = string("input_241_cast_fp16")];
             bool x_101_transpose_x_0 = const()[name = string("x_101_transpose_x_0"), val = bool(false)];
             bool x_101_transpose_y_0 = const()[name = string("x_101_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_11_cast_fp16 = transpose(perm = value_11_perm_0, x = v_9_cast_fp16)[name = string("transpose_281")];
-            tensor<fp16, [1, 8, 188, 128]> x_101_cast_fp16 = matmul(transpose_x = x_101_transpose_x_0, transpose_y = x_101_transpose_y_0, x = input_241_cast_fp16, y = value_11_cast_fp16)[name = string("x_101_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> value_11_cast_fp16 = transpose(perm = value_11_perm_0, x = v_9_cast_fp16)[name = string("transpose_281")];
+            tensor<fp16, [1, 8, 128, 128]> x_101_cast_fp16 = matmul(transpose_x = x_101_transpose_x_0, transpose_y = x_101_transpose_y_0, x = input_241_cast_fp16, y = value_11_cast_fp16)[name = string("x_101_cast_fp16")];
             tensor<int32, [4]> var_957_perm_0 = const()[name = string("op_957_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_958 = const()[name = string("op_958"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_957_cast_fp16 = transpose(perm = var_957_perm_0, x = x_101_cast_fp16)[name = string("transpose_280")];
-            tensor<fp16, [1, 188, 1024]> input_243_cast_fp16 = reshape(shape = var_958, x = var_957_cast_fp16)[name = string("input_243_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523771]> module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66806528))), indices_nonzero_data = tensor<uint6, [523771]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66937664))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67330560))))[name = string("module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> var_957_cast_fp16 = transpose(perm = var_957_perm_0, x = x_101_cast_fp16)[name = string("transpose_280")];
+            tensor<fp16, [1, 128, 1024]> input_243_cast_fp16 = reshape(shape = var_958, x = var_957_cast_fp16)[name = string("input_243_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523771]> module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64266496))), indices_nonzero_data = tensor<uint6, [523771]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64397632))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64790528))))[name = string("module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_43_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_243_cast_fp16)[name = string("linear_43_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_247_cast_fp16 = add(x = input_239_cast_fp16, y = linear_43_cast_fp16)[name = string("input_247_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_43_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_243_cast_fp16)[name = string("linear_43_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_247_cast_fp16 = add(x = input_239_cast_fp16, y = linear_43_cast_fp16)[name = string("input_247_cast_fp16")];
             tensor<int32, [1]> x_105_axes_0 = const()[name = string("x_105_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_4_norm_conv_weight_to_fp16 = const()[name = string("module_layers_4_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67347008)))];
-            tensor<fp16, [1024]> module_layers_4_norm_conv_bias_to_fp16 = const()[name = string("module_layers_4_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67349120)))];
-            tensor<fp16, [1, 188, 1024]> x_105_cast_fp16 = layer_norm(axes = x_105_axes_0, beta = module_layers_4_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_conv_weight_to_fp16, x = input_247_cast_fp16)[name = string("x_105_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_4_norm_conv_weight_to_fp16 = const()[name = string("module_layers_4_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64806976)))];
+            tensor<fp16, [1024]> module_layers_4_norm_conv_bias_to_fp16 = const()[name = string("module_layers_4_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64809088)))];
+            tensor<fp16, [1, 128, 1024]> x_105_cast_fp16 = layer_norm(axes = x_105_axes_0, beta = module_layers_4_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_conv_weight_to_fp16, x = input_247_cast_fp16)[name = string("x_105_cast_fp16")];
             tensor<int32, [3]> input_249_perm_0 = const()[name = string("input_249_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_251_pad_type_0 = const()[name = string("input_251_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_251_strides_0 = const()[name = string("input_251_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_251_pad_0 = const()[name = string("input_251_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_251_dilations_0 = const()[name = string("input_251_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_251_groups_0 = const()[name = string("input_251_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1046993]> module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67351232))), indices_nonzero_data = tensor<uint6, [1046993]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67613440))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68398784))))[name = string("module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1046993]> module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64811200))), indices_nonzero_data = tensor<uint6, [1046993]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65073408))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65858752))))[name = string("module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_249_cast_fp16 = transpose(perm = input_249_perm_0, x = x_105_cast_fp16)[name = string("transpose_279")];
-            tensor<fp16, [1, 2048, 188]> input_251_cast_fp16 = conv(dilations = input_251_dilations_0, groups = input_251_groups_0, pad = input_251_pad_0, pad_type = input_251_pad_type_0, strides = input_251_strides_0, weight = module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_249_cast_fp16)[name = string("input_251_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_249_cast_fp16 = transpose(perm = input_249_perm_0, x = x_105_cast_fp16)[name = string("transpose_279")];
+            tensor<fp16, [1, 2048, 128]> input_251_cast_fp16 = conv(dilations = input_251_dilations_0, groups = input_251_groups_0, pad = input_251_pad_0, pad_type = input_251_pad_type_0, strides = input_251_strides_0, weight = module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_249_cast_fp16)[name = string("input_251_cast_fp16")];
             int32 x_107_split_num_splits_0 = const()[name = string("x_107_split_num_splits_0"), val = int32(2)];
             int32 x_107_split_axis_0 = const()[name = string("x_107_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_107_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_107_split_cast_fp16_1 = split(axis = x_107_split_axis_0, num_splits = x_107_split_num_splits_0, x = input_251_cast_fp16)[name = string("x_107_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_107_split_1_sigmoid_cast_fp16 = sigmoid(x = x_107_split_cast_fp16_1)[name = string("x_107_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_107_cast_fp16 = mul(x = x_107_split_cast_fp16_0, y = x_107_split_1_sigmoid_cast_fp16)[name = string("x_107_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_253_cast_fp16 = select(a = var_11_to_fp16, b = x_107_cast_fp16, cond = var_328)[name = string("input_253_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_107_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_107_split_cast_fp16_1 = split(axis = x_107_split_axis_0, num_splits = x_107_split_num_splits_0, x = input_251_cast_fp16)[name = string("x_107_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_107_split_1_sigmoid_cast_fp16 = sigmoid(x = x_107_split_cast_fp16_1)[name = string("x_107_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_107_cast_fp16 = mul(x = x_107_split_cast_fp16_0, y = x_107_split_1_sigmoid_cast_fp16)[name = string("x_107_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_253_cast_fp16 = select(a = var_11_to_fp16, b = x_107_cast_fp16, cond = var_328)[name = string("input_253_cast_fp16")];
             tensor<int32, [6]> input_255_pad_0 = const()[name = string("input_255_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_255_mode_0 = const()[name = string("input_255_mode_0"), val = string("constant")];
             fp16 const_57_to_fp16 = const()[name = string("const_57_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_255_cast_fp16 = pad(constant_val = const_57_to_fp16, mode = input_255_mode_0, pad = input_255_pad_0, x = input_253_cast_fp16)[name = string("input_255_cast_fp16")];
+            tensor<fp16, [1, 1024, 136]> input_255_cast_fp16 = pad(constant_val = const_57_to_fp16, mode = input_255_mode_0, pad = input_255_pad_0, x = input_253_cast_fp16)[name = string("input_255_cast_fp16")];
             string input_257_pad_type_0 = const()[name = string("input_257_pad_type_0"), val = string("valid")];
             int32 input_257_groups_0 = const()[name = string("input_257_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_257_strides_0 = const()[name = string("input_257_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_257_pad_0 = const()[name = string("input_257_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_257_dilations_0 = const()[name = string("input_257_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_256_to_fp16 = const()[name = string("const_256_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68431616)))];
-            tensor<fp16, [1024]> const_257_to_fp16 = const()[name = string("const_257_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68450112)))];
-            tensor<fp16, [1, 1024, 188]> input_259_cast_fp16 = conv(bias = const_257_to_fp16, dilations = input_257_dilations_0, groups = input_257_groups_0, pad = input_257_pad_0, pad_type = input_257_pad_type_0, strides = input_257_strides_0, weight = const_256_to_fp16, x = input_255_cast_fp16)[name = string("input_259_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_261_cast_fp16 = silu(x = input_259_cast_fp16)[name = string("input_261_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_256_to_fp16 = const()[name = string("const_256_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65891584)))];
+            tensor<fp16, [1024]> const_257_to_fp16 = const()[name = string("const_257_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65910080)))];
+            tensor<fp16, [1, 1024, 128]> input_259_cast_fp16 = conv(bias = const_257_to_fp16, dilations = input_257_dilations_0, groups = input_257_groups_0, pad = input_257_pad_0, pad_type = input_257_pad_type_0, strides = input_257_strides_0, weight = const_256_to_fp16, x = input_255_cast_fp16)[name = string("input_259_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_261_cast_fp16 = silu(x = input_259_cast_fp16)[name = string("input_261_cast_fp16")];
             string x_109_pad_type_0 = const()[name = string("x_109_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_109_strides_0 = const()[name = string("x_109_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_109_pad_0 = const()[name = string("x_109_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_109_dilations_0 = const()[name = string("x_109_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_109_groups_0 = const()[name = string("x_109_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520275]> module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68452224))), indices_nonzero_data = tensor<uint6, [520275]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68583360))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68973632))))[name = string("module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520275]> module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65912192))), indices_nonzero_data = tensor<uint6, [520275]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66043328))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66433600))))[name = string("module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_109_cast_fp16 = conv(dilations = x_109_dilations_0, groups = x_109_groups_0, pad = x_109_pad_0, pad_type = x_109_pad_type_0, strides = x_109_strides_0, weight = module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_261_cast_fp16)[name = string("x_109_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_109_cast_fp16 = conv(dilations = x_109_dilations_0, groups = x_109_groups_0, pad = x_109_pad_0, pad_type = x_109_pad_type_0, strides = x_109_strides_0, weight = module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_261_cast_fp16)[name = string("x_109_cast_fp16")];
             tensor<int32, [3]> input_263_perm_0 = const()[name = string("input_263_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_263_cast_fp16 = transpose(perm = input_263_perm_0, x = x_109_cast_fp16)[name = string("transpose_278")];
-            tensor<fp16, [1, 188, 1024]> input_265_cast_fp16 = add(x = input_247_cast_fp16, y = input_263_cast_fp16)[name = string("input_265_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_263_cast_fp16 = transpose(perm = input_263_perm_0, x = x_109_cast_fp16)[name = string("transpose_278")];
+            tensor<fp16, [1, 128, 1024]> input_265_cast_fp16 = add(x = input_247_cast_fp16, y = input_263_cast_fp16)[name = string("input_265_cast_fp16")];
             tensor<int32, [1]> input_267_axes_0 = const()[name = string("input_267_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_4_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_4_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68990080)))];
-            tensor<fp16, [1024]> module_layers_4_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_4_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68992192)))];
-            tensor<fp16, [1, 188, 1024]> input_267_cast_fp16 = layer_norm(axes = input_267_axes_0, beta = module_layers_4_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_feed_forward2_weight_to_fp16, x = input_265_cast_fp16)[name = string("input_267_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2060977]> module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68994304))), indices_nonzero_data = tensor<uint6, [2060977]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69518656))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71064512))))[name = string("module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_4_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_4_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66450048)))];
+            tensor<fp16, [1024]> module_layers_4_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_4_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66452160)))];
+            tensor<fp16, [1, 128, 1024]> input_267_cast_fp16 = layer_norm(axes = input_267_axes_0, beta = module_layers_4_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_feed_forward2_weight_to_fp16, x = input_265_cast_fp16)[name = string("input_267_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2060977]> module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66454272))), indices_nonzero_data = tensor<uint6, [2060977]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66978624))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68524480))))[name = string("module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_44_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_267_cast_fp16)[name = string("linear_44_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_271_cast_fp16 = silu(x = linear_44_cast_fp16)[name = string("input_271_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1931171]> module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71130112))), indices_nonzero_data = tensor<uint6, [1931171]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71654464))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73102912))))[name = string("module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_44_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_267_cast_fp16)[name = string("linear_44_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_271_cast_fp16 = silu(x = linear_44_cast_fp16)[name = string("input_271_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1931171]> module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68590080))), indices_nonzero_data = tensor<uint6, [1931171]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69114432))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70562880))))[name = string("module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_45_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_271_cast_fp16)[name = string("linear_45_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_45_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_271_cast_fp16)[name = string("linear_45_cast_fp16")];
             fp16 var_1018_to_fp16 = const()[name = string("op_1018_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_1019_cast_fp16 = mul(x = linear_45_cast_fp16, y = var_1018_to_fp16)[name = string("op_1019_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_277_cast_fp16 = add(x = input_265_cast_fp16, y = var_1019_cast_fp16)[name = string("input_277_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_1019_cast_fp16 = mul(x = linear_45_cast_fp16, y = var_1018_to_fp16)[name = string("op_1019_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_277_cast_fp16 = add(x = input_265_cast_fp16, y = var_1019_cast_fp16)[name = string("input_277_cast_fp16")];
             tensor<int32, [1]> input_279_axes_0 = const()[name = string("input_279_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_4_norm_out_weight_to_fp16 = const()[name = string("module_layers_4_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73119360)))];
-            tensor<fp16, [1024]> module_layers_4_norm_out_bias_to_fp16 = const()[name = string("module_layers_4_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73121472)))];
-            tensor<fp16, [1, 188, 1024]> input_279_cast_fp16 = layer_norm(axes = input_279_axes_0, beta = module_layers_4_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_out_weight_to_fp16, x = input_277_cast_fp16)[name = string("input_279_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_4_norm_out_weight_to_fp16 = const()[name = string("module_layers_4_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70579328)))];
+            tensor<fp16, [1024]> module_layers_4_norm_out_bias_to_fp16 = const()[name = string("module_layers_4_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70581440)))];
+            tensor<fp16, [1, 128, 1024]> input_279_cast_fp16 = layer_norm(axes = input_279_axes_0, beta = module_layers_4_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_out_weight_to_fp16, x = input_277_cast_fp16)[name = string("input_279_cast_fp16")];
             tensor<int32, [1]> input_281_axes_0 = const()[name = string("input_281_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_5_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_5_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73123584)))];
-            tensor<fp16, [1024]> module_layers_5_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_5_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73125696)))];
-            tensor<fp16, [1, 188, 1024]> input_281_cast_fp16 = layer_norm(axes = input_281_axes_0, beta = module_layers_5_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_feed_forward1_weight_to_fp16, x = input_279_cast_fp16)[name = string("input_281_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2075089]> module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73127808))), indices_nonzero_data = tensor<uint6, [2075089]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73652160))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75208576))))[name = string("module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_5_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_5_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70583552)))];
+            tensor<fp16, [1024]> module_layers_5_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_5_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70585664)))];
+            tensor<fp16, [1, 128, 1024]> input_281_cast_fp16 = layer_norm(axes = input_281_axes_0, beta = module_layers_5_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_feed_forward1_weight_to_fp16, x = input_279_cast_fp16)[name = string("input_281_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2075089]> module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70587776))), indices_nonzero_data = tensor<uint6, [2075089]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71112128))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72668544))))[name = string("module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_46_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_281_cast_fp16)[name = string("linear_46_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_285_cast_fp16 = silu(x = linear_46_cast_fp16)[name = string("input_285_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1956387]> module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75274176))), indices_nonzero_data = tensor<uint6, [1956387]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75798528))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77265920))))[name = string("module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_46_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_281_cast_fp16)[name = string("linear_46_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_285_cast_fp16 = silu(x = linear_46_cast_fp16)[name = string("input_285_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1956387]> module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72734144))), indices_nonzero_data = tensor<uint6, [1956387]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73258496))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74725888))))[name = string("module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_47_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_285_cast_fp16)[name = string("linear_47_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_47_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_285_cast_fp16)[name = string("linear_47_cast_fp16")];
             fp16 var_1047_to_fp16 = const()[name = string("op_1047_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_1048_cast_fp16 = mul(x = linear_47_cast_fp16, y = var_1047_to_fp16)[name = string("op_1048_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_291_cast_fp16 = add(x = input_279_cast_fp16, y = var_1048_cast_fp16)[name = string("input_291_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_1048_cast_fp16 = mul(x = linear_47_cast_fp16, y = var_1047_to_fp16)[name = string("op_1048_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_291_cast_fp16 = add(x = input_279_cast_fp16, y = var_1048_cast_fp16)[name = string("input_291_cast_fp16")];
             tensor<int32, [1]> query_11_axes_0 = const()[name = string("query_11_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_5_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_5_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77282368)))];
-            tensor<fp16, [1024]> module_layers_5_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_5_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77284480)))];
-            tensor<fp16, [1, 188, 1024]> query_11_cast_fp16 = layer_norm(axes = query_11_axes_0, beta = module_layers_5_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_self_att_weight_to_fp16, x = input_291_cast_fp16)[name = string("query_11_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521972]> module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77286592))), indices_nonzero_data = tensor<uint6, [521972]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77417728))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77809280))))[name = string("module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_5_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_5_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74742336)))];
+            tensor<fp16, [1024]> module_layers_5_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_5_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74744448)))];
+            tensor<fp16, [1, 128, 1024]> query_11_cast_fp16 = layer_norm(axes = query_11_axes_0, beta = module_layers_5_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_self_att_weight_to_fp16, x = input_291_cast_fp16)[name = string("query_11_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521972]> module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74746560))), indices_nonzero_data = tensor<uint6, [521972]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74877696))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75269248))))[name = string("module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_48_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_11_cast_fp16)[name = string("linear_48_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_48_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_11_cast_fp16)[name = string("linear_48_cast_fp16")];
             tensor<int32, [4]> var_1064 = const()[name = string("op_1064"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_31_cast_fp16 = reshape(shape = var_1064, x = linear_48_cast_fp16)[name = string("q_31_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522028]> module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77825728))), indices_nonzero_data = tensor<uint6, [522028]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77956864))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78348480))))[name = string("module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> q_31_cast_fp16 = reshape(shape = var_1064, x = linear_48_cast_fp16)[name = string("q_31_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522028]> module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75285696))), indices_nonzero_data = tensor<uint6, [522028]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75416832))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75808448))))[name = string("module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_49_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_11_cast_fp16)[name = string("linear_49_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_49_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_11_cast_fp16)[name = string("linear_49_cast_fp16")];
             tensor<int32, [4]> var_1068 = const()[name = string("op_1068"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_21_cast_fp16 = reshape(shape = var_1068, x = linear_49_cast_fp16)[name = string("k_21_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519496]> module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78364928))), indices_nonzero_data = tensor<uint6, [519496]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78496064))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78885760))))[name = string("module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> k_21_cast_fp16 = reshape(shape = var_1068, x = linear_49_cast_fp16)[name = string("k_21_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519496]> module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75824896))), indices_nonzero_data = tensor<uint6, [519496]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75956032))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76345728))))[name = string("module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_50_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_11_cast_fp16)[name = string("linear_50_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_50_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_11_cast_fp16)[name = string("linear_50_cast_fp16")];
             tensor<int32, [4]> var_1072 = const()[name = string("op_1072"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_11_cast_fp16 = reshape(shape = var_1072, x = linear_50_cast_fp16)[name = string("v_11_cast_fp16")];
+            tensor<fp16, [1, 128, 8, 128]> v_11_cast_fp16 = reshape(shape = var_1072, x = linear_50_cast_fp16)[name = string("v_11_cast_fp16")];
             tensor<int32, [4]> value_13_perm_0 = const()[name = string("value_13_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_5_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_5_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78902208)))];
-            tensor<fp16, [1, 188, 8, 128]> var_1084_cast_fp16 = add(x = q_31_cast_fp16, y = module_layers_5_self_attn_pos_bias_u_to_fp16)[name = string("op_1084_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_5_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_5_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78904320)))];
-            tensor<fp16, [1, 188, 8, 128]> var_1086_cast_fp16 = add(x = q_31_cast_fp16, y = module_layers_5_self_attn_pos_bias_v_to_fp16)[name = string("op_1086_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_5_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_5_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76362176)))];
+            tensor<fp16, [1, 128, 8, 128]> var_1084_cast_fp16 = add(x = q_31_cast_fp16, y = module_layers_5_self_attn_pos_bias_u_to_fp16)[name = string("op_1084_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_5_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_5_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76364288)))];
+            tensor<fp16, [1, 128, 8, 128]> var_1086_cast_fp16 = add(x = q_31_cast_fp16, y = module_layers_5_self_attn_pos_bias_v_to_fp16)[name = string("op_1086_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_11_perm_0 = const()[name = string("q_with_bias_v_11_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_117_transpose_x_0 = const()[name = string("x_117_transpose_x_0"), val = bool(false)];
             bool x_117_transpose_y_0 = const()[name = string("x_117_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_1088_to_fp16 = const()[name = string("op_1088_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78906432)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_11_cast_fp16 = transpose(perm = q_with_bias_v_11_perm_0, x = var_1086_cast_fp16)[name = string("transpose_277")];
-            tensor<fp16, [1, 8, 188, 375]> x_117_cast_fp16 = matmul(transpose_x = x_117_transpose_x_0, transpose_y = x_117_transpose_y_0, x = q_with_bias_v_11_cast_fp16, y = var_1088_to_fp16)[name = string("x_117_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 255]> var_1088_to_fp16 = const()[name = string("op_1088_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(310033344)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_11_cast_fp16 = transpose(perm = q_with_bias_v_11_perm_0, x = var_1086_cast_fp16)[name = string("transpose_277")];
+            tensor<fp16, [1, 8, 128, 255]> x_117_cast_fp16 = matmul(transpose_x = x_117_transpose_x_0, transpose_y = x_117_transpose_y_0, x = q_with_bias_v_11_cast_fp16, y = var_1088_to_fp16)[name = string("x_117_cast_fp16")];
             tensor<int32, [8]> x_119_pad_0 = const()[name = string("x_119_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_119_mode_0 = const()[name = string("x_119_mode_0"), val = string("constant")];
             fp16 const_64_to_fp16 = const()[name = string("const_64_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_119_cast_fp16 = pad(constant_val = const_64_to_fp16, mode = x_119_mode_0, pad = x_119_pad_0, x = x_117_cast_fp16)[name = string("x_119_cast_fp16")];
-            tensor<int32, [4]> var_1096 = const()[name = string("op_1096"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_121_cast_fp16 = reshape(shape = var_1096, x = x_119_cast_fp16)[name = string("x_121_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 256]> x_119_cast_fp16 = pad(constant_val = const_64_to_fp16, mode = x_119_mode_0, pad = x_119_pad_0, x = x_117_cast_fp16)[name = string("x_119_cast_fp16")];
+            tensor<int32, [4]> var_1096 = const()[name = string("op_1096"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_121_cast_fp16 = reshape(shape = var_1096, x = x_119_cast_fp16)[name = string("x_121_cast_fp16")];
             tensor<int32, [4]> var_1100_begin_0 = const()[name = string("op_1100_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_1100_end_0 = const()[name = string("op_1100_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_1100_end_0 = const()[name = string("op_1100_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
             tensor<bool, [4]> var_1100_end_mask_0 = const()[name = string("op_1100_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_1100_cast_fp16 = slice_by_index(begin = var_1100_begin_0, end = var_1100_end_0, end_mask = var_1100_end_mask_0, x = x_121_cast_fp16)[name = string("op_1100_cast_fp16")];
-            tensor<int32, [4]> var_1101 = const()[name = string("op_1101"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_21_cast_fp16 = reshape(shape = var_1101, x = var_1100_cast_fp16)[name = string("matrix_bd_21_cast_fp16")];
+            tensor<fp16, [1, 8, 255, 128]> var_1100_cast_fp16 = slice_by_index(begin = var_1100_begin_0, end = var_1100_end_0, end_mask = var_1100_end_mask_0, x = x_121_cast_fp16)[name = string("op_1100_cast_fp16")];
+            tensor<int32, [4]> var_1101 = const()[name = string("op_1101"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_21_cast_fp16 = reshape(shape = var_1101, x = var_1100_cast_fp16)[name = string("matrix_bd_21_cast_fp16")];
             bool matrix_ac_11_transpose_x_0 = const()[name = string("matrix_ac_11_transpose_x_0"), val = bool(false)];
             bool matrix_ac_11_transpose_y_0 = const()[name = string("matrix_ac_11_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_106_perm_0 = const()[name = string("transpose_106_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_107_perm_0 = const()[name = string("transpose_107_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_107 = transpose(perm = transpose_107_perm_0, x = k_21_cast_fp16)[name = string("transpose_275")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_106 = transpose(perm = transpose_106_perm_0, x = var_1084_cast_fp16)[name = string("transpose_276")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_11_cast_fp16 = matmul(transpose_x = matrix_ac_11_transpose_x_0, transpose_y = matrix_ac_11_transpose_y_0, x = transpose_106, y = transpose_107)[name = string("matrix_ac_11_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_107 = transpose(perm = transpose_107_perm_0, x = k_21_cast_fp16)[name = string("transpose_275")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_106 = transpose(perm = transpose_106_perm_0, x = var_1084_cast_fp16)[name = string("transpose_276")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_11_cast_fp16 = matmul(transpose_x = matrix_ac_11_transpose_x_0, transpose_y = matrix_ac_11_transpose_y_0, x = transpose_106, y = transpose_107)[name = string("matrix_ac_11_cast_fp16")];
             tensor<int32, [4]> matrix_bd_23_begin_0 = const()[name = string("matrix_bd_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_23_end_0 = const()[name = string("matrix_bd_23_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_23_end_0 = const()[name = string("matrix_bd_23_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
             tensor<bool, [4]> matrix_bd_23_end_mask_0 = const()[name = string("matrix_bd_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_23_cast_fp16 = slice_by_index(begin = matrix_bd_23_begin_0, end = matrix_bd_23_end_0, end_mask = matrix_bd_23_end_mask_0, x = matrix_bd_21_cast_fp16)[name = string("matrix_bd_23_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_1110_cast_fp16 = add(x = matrix_ac_11_cast_fp16, y = matrix_bd_23_cast_fp16)[name = string("op_1110_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_23_cast_fp16 = slice_by_index(begin = matrix_bd_23_begin_0, end = matrix_bd_23_end_0, end_mask = matrix_bd_23_end_mask_0, x = matrix_bd_21_cast_fp16)[name = string("matrix_bd_23_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_1110_cast_fp16 = add(x = matrix_ac_11_cast_fp16, y = matrix_bd_23_cast_fp16)[name = string("op_1110_cast_fp16")];
             fp16 _inversed_scores_21_y_0_to_fp16 = const()[name = string("_inversed_scores_21_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_21_cast_fp16 = mul(x = var_1110_cast_fp16, y = _inversed_scores_21_y_0_to_fp16)[name = string("_inversed_scores_21_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_23_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_21_cast_fp16, cond = mask_3)[name = string("scores_23_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_1116_cast_fp16 = softmax(axis = var_30, x = scores_23_cast_fp16)[name = string("op_1116_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_293_cast_fp16 = select(a = var_11_to_fp16, b = var_1116_cast_fp16, cond = mask_3)[name = string("input_293_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_21_cast_fp16 = mul(x = var_1110_cast_fp16, y = _inversed_scores_21_y_0_to_fp16)[name = string("_inversed_scores_21_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_23_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_21_cast_fp16, cond = mask_3)[name = string("scores_23_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_1116_cast_fp16 = softmax(axis = var_30, x = scores_23_cast_fp16)[name = string("op_1116_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_293_cast_fp16 = select(a = var_11_to_fp16, b = var_1116_cast_fp16, cond = mask_3)[name = string("input_293_cast_fp16")];
             bool x_123_transpose_x_0 = const()[name = string("x_123_transpose_x_0"), val = bool(false)];
             bool x_123_transpose_y_0 = const()[name = string("x_123_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_13_cast_fp16 = transpose(perm = value_13_perm_0, x = v_11_cast_fp16)[name = string("transpose_274")];
-            tensor<fp16, [1, 8, 188, 128]> x_123_cast_fp16 = matmul(transpose_x = x_123_transpose_x_0, transpose_y = x_123_transpose_y_0, x = input_293_cast_fp16, y = value_13_cast_fp16)[name = string("x_123_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> value_13_cast_fp16 = transpose(perm = value_13_perm_0, x = v_11_cast_fp16)[name = string("transpose_274")];
+            tensor<fp16, [1, 8, 128, 128]> x_123_cast_fp16 = matmul(transpose_x = x_123_transpose_x_0, transpose_y = x_123_transpose_y_0, x = input_293_cast_fp16, y = value_13_cast_fp16)[name = string("x_123_cast_fp16")];
             tensor<int32, [4]> var_1120_perm_0 = const()[name = string("op_1120_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_1121 = const()[name = string("op_1121"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_1120_cast_fp16 = transpose(perm = var_1120_perm_0, x = x_123_cast_fp16)[name = string("transpose_273")];
-            tensor<fp16, [1, 188, 1024]> input_295_cast_fp16 = reshape(shape = var_1121, x = var_1120_cast_fp16)[name = string("input_295_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523851]> module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79674496))), indices_nonzero_data = tensor<uint6, [523851]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79805632))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80198592))))[name = string("module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> var_1120_cast_fp16 = transpose(perm = var_1120_perm_0, x = x_123_cast_fp16)[name = string("transpose_273")];
+            tensor<fp16, [1, 128, 1024]> input_295_cast_fp16 = reshape(shape = var_1121, x = var_1120_cast_fp16)[name = string("input_295_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523851]> module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76626560))), indices_nonzero_data = tensor<uint6, [523851]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76757696))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77150656))))[name = string("module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_52_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_295_cast_fp16)[name = string("linear_52_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_299_cast_fp16 = add(x = input_291_cast_fp16, y = linear_52_cast_fp16)[name = string("input_299_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_52_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_295_cast_fp16)[name = string("linear_52_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_299_cast_fp16 = add(x = input_291_cast_fp16, y = linear_52_cast_fp16)[name = string("input_299_cast_fp16")];
             tensor<int32, [1]> x_127_axes_0 = const()[name = string("x_127_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_5_norm_conv_weight_to_fp16 = const()[name = string("module_layers_5_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80215040)))];
-            tensor<fp16, [1024]> module_layers_5_norm_conv_bias_to_fp16 = const()[name = string("module_layers_5_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80217152)))];
-            tensor<fp16, [1, 188, 1024]> x_127_cast_fp16 = layer_norm(axes = x_127_axes_0, beta = module_layers_5_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_conv_weight_to_fp16, x = input_299_cast_fp16)[name = string("x_127_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_5_norm_conv_weight_to_fp16 = const()[name = string("module_layers_5_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77167104)))];
+            tensor<fp16, [1024]> module_layers_5_norm_conv_bias_to_fp16 = const()[name = string("module_layers_5_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77169216)))];
+            tensor<fp16, [1, 128, 1024]> x_127_cast_fp16 = layer_norm(axes = x_127_axes_0, beta = module_layers_5_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_conv_weight_to_fp16, x = input_299_cast_fp16)[name = string("x_127_cast_fp16")];
             tensor<int32, [3]> input_301_perm_0 = const()[name = string("input_301_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_303_pad_type_0 = const()[name = string("input_303_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_303_strides_0 = const()[name = string("input_303_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_303_pad_0 = const()[name = string("input_303_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_303_dilations_0 = const()[name = string("input_303_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_303_groups_0 = const()[name = string("input_303_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047810]> module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80219264))), indices_nonzero_data = tensor<uint6, [1047810]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80481472))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81267456))))[name = string("module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047810]> module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77171328))), indices_nonzero_data = tensor<uint6, [1047810]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77433536))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78219520))))[name = string("module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_301_cast_fp16 = transpose(perm = input_301_perm_0, x = x_127_cast_fp16)[name = string("transpose_272")];
-            tensor<fp16, [1, 2048, 188]> input_303_cast_fp16 = conv(dilations = input_303_dilations_0, groups = input_303_groups_0, pad = input_303_pad_0, pad_type = input_303_pad_type_0, strides = input_303_strides_0, weight = module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_301_cast_fp16)[name = string("input_303_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_301_cast_fp16 = transpose(perm = input_301_perm_0, x = x_127_cast_fp16)[name = string("transpose_272")];
+            tensor<fp16, [1, 2048, 128]> input_303_cast_fp16 = conv(dilations = input_303_dilations_0, groups = input_303_groups_0, pad = input_303_pad_0, pad_type = input_303_pad_type_0, strides = input_303_strides_0, weight = module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_301_cast_fp16)[name = string("input_303_cast_fp16")];
             int32 x_129_split_num_splits_0 = const()[name = string("x_129_split_num_splits_0"), val = int32(2)];
             int32 x_129_split_axis_0 = const()[name = string("x_129_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_129_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_129_split_cast_fp16_1 = split(axis = x_129_split_axis_0, num_splits = x_129_split_num_splits_0, x = input_303_cast_fp16)[name = string("x_129_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_129_split_1_sigmoid_cast_fp16 = sigmoid(x = x_129_split_cast_fp16_1)[name = string("x_129_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_129_cast_fp16 = mul(x = x_129_split_cast_fp16_0, y = x_129_split_1_sigmoid_cast_fp16)[name = string("x_129_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_305_cast_fp16 = select(a = var_11_to_fp16, b = x_129_cast_fp16, cond = var_328)[name = string("input_305_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_129_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_129_split_cast_fp16_1 = split(axis = x_129_split_axis_0, num_splits = x_129_split_num_splits_0, x = input_303_cast_fp16)[name = string("x_129_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_129_split_1_sigmoid_cast_fp16 = sigmoid(x = x_129_split_cast_fp16_1)[name = string("x_129_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_129_cast_fp16 = mul(x = x_129_split_cast_fp16_0, y = x_129_split_1_sigmoid_cast_fp16)[name = string("x_129_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_305_cast_fp16 = select(a = var_11_to_fp16, b = x_129_cast_fp16, cond = var_328)[name = string("input_305_cast_fp16")];
             tensor<int32, [6]> input_307_pad_0 = const()[name = string("input_307_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_307_mode_0 = const()[name = string("input_307_mode_0"), val = string("constant")];
             fp16 const_67_to_fp16 = const()[name = string("const_67_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_307_cast_fp16 = pad(constant_val = const_67_to_fp16, mode = input_307_mode_0, pad = input_307_pad_0, x = input_305_cast_fp16)[name = string("input_307_cast_fp16")];
+            tensor<fp16, [1, 1024, 136]> input_307_cast_fp16 = pad(constant_val = const_67_to_fp16, mode = input_307_mode_0, pad = input_307_pad_0, x = input_305_cast_fp16)[name = string("input_307_cast_fp16")];
             string input_309_pad_type_0 = const()[name = string("input_309_pad_type_0"), val = string("valid")];
             int32 input_309_groups_0 = const()[name = string("input_309_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_309_strides_0 = const()[name = string("input_309_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_309_pad_0 = const()[name = string("input_309_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_309_dilations_0 = const()[name = string("input_309_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_258_to_fp16 = const()[name = string("const_258_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81300288)))];
-            tensor<fp16, [1024]> const_259_to_fp16 = const()[name = string("const_259_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81318784)))];
-            tensor<fp16, [1, 1024, 188]> input_311_cast_fp16 = conv(bias = const_259_to_fp16, dilations = input_309_dilations_0, groups = input_309_groups_0, pad = input_309_pad_0, pad_type = input_309_pad_type_0, strides = input_309_strides_0, weight = const_258_to_fp16, x = input_307_cast_fp16)[name = string("input_311_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_313_cast_fp16 = silu(x = input_311_cast_fp16)[name = string("input_313_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_258_to_fp16 = const()[name = string("const_258_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78252352)))];
+            tensor<fp16, [1024]> const_259_to_fp16 = const()[name = string("const_259_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78270848)))];
+            tensor<fp16, [1, 1024, 128]> input_311_cast_fp16 = conv(bias = const_259_to_fp16, dilations = input_309_dilations_0, groups = input_309_groups_0, pad = input_309_pad_0, pad_type = input_309_pad_type_0, strides = input_309_strides_0, weight = const_258_to_fp16, x = input_307_cast_fp16)[name = string("input_311_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_313_cast_fp16 = silu(x = input_311_cast_fp16)[name = string("input_313_cast_fp16")];
             string x_131_pad_type_0 = const()[name = string("x_131_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_131_strides_0 = const()[name = string("x_131_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_131_pad_0 = const()[name = string("x_131_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_131_dilations_0 = const()[name = string("x_131_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_131_groups_0 = const()[name = string("x_131_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520509]> module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81320896))), indices_nonzero_data = tensor<uint6, [520509]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81452032))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81842496))))[name = string("module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520509]> module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78272960))), indices_nonzero_data = tensor<uint6, [520509]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78404096))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78794560))))[name = string("module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_131_cast_fp16 = conv(dilations = x_131_dilations_0, groups = x_131_groups_0, pad = x_131_pad_0, pad_type = x_131_pad_type_0, strides = x_131_strides_0, weight = module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_313_cast_fp16)[name = string("x_131_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_131_cast_fp16 = conv(dilations = x_131_dilations_0, groups = x_131_groups_0, pad = x_131_pad_0, pad_type = x_131_pad_type_0, strides = x_131_strides_0, weight = module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_313_cast_fp16)[name = string("x_131_cast_fp16")];
             tensor<int32, [3]> input_315_perm_0 = const()[name = string("input_315_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_315_cast_fp16 = transpose(perm = input_315_perm_0, x = x_131_cast_fp16)[name = string("transpose_271")];
-            tensor<fp16, [1, 188, 1024]> input_317_cast_fp16 = add(x = input_299_cast_fp16, y = input_315_cast_fp16)[name = string("input_317_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_315_cast_fp16 = transpose(perm = input_315_perm_0, x = x_131_cast_fp16)[name = string("transpose_271")];
+            tensor<fp16, [1, 128, 1024]> input_317_cast_fp16 = add(x = input_299_cast_fp16, y = input_315_cast_fp16)[name = string("input_317_cast_fp16")];
             tensor<int32, [1]> input_319_axes_0 = const()[name = string("input_319_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_5_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_5_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81858944)))];
-            tensor<fp16, [1024]> module_layers_5_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_5_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81861056)))];
-            tensor<fp16, [1, 188, 1024]> input_319_cast_fp16 = layer_norm(axes = input_319_axes_0, beta = module_layers_5_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_feed_forward2_weight_to_fp16, x = input_317_cast_fp16)[name = string("input_319_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2063618]> module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81863168))), indices_nonzero_data = tensor<uint6, [2063618]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82387520))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83935360))))[name = string("module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_5_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_5_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78811008)))];
+            tensor<fp16, [1024]> module_layers_5_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_5_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78813120)))];
+            tensor<fp16, [1, 128, 1024]> input_319_cast_fp16 = layer_norm(axes = input_319_axes_0, beta = module_layers_5_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_feed_forward2_weight_to_fp16, x = input_317_cast_fp16)[name = string("input_319_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2063618]> module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78815232))), indices_nonzero_data = tensor<uint6, [2063618]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79339584))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80887424))))[name = string("module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_53_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_319_cast_fp16)[name = string("linear_53_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_323_cast_fp16 = silu(x = linear_53_cast_fp16)[name = string("input_323_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1985406]> module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(84000960))), indices_nonzero_data = tensor<uint6, [1985406]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(84525312))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86014464))))[name = string("module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_53_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_319_cast_fp16)[name = string("linear_53_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_323_cast_fp16 = silu(x = linear_53_cast_fp16)[name = string("input_323_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1985406]> module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80953024))), indices_nonzero_data = tensor<uint6, [1985406]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81477376))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82966528))))[name = string("module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_54_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_323_cast_fp16)[name = string("linear_54_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_54_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_323_cast_fp16)[name = string("linear_54_cast_fp16")];
             fp16 var_1181_to_fp16 = const()[name = string("op_1181_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_1182_cast_fp16 = mul(x = linear_54_cast_fp16, y = var_1181_to_fp16)[name = string("op_1182_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_329_cast_fp16 = add(x = input_317_cast_fp16, y = var_1182_cast_fp16)[name = string("input_329_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_1182_cast_fp16 = mul(x = linear_54_cast_fp16, y = var_1181_to_fp16)[name = string("op_1182_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_329_cast_fp16 = add(x = input_317_cast_fp16, y = var_1182_cast_fp16)[name = string("input_329_cast_fp16")];
             tensor<int32, [1]> input_331_axes_0 = const()[name = string("input_331_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_5_norm_out_weight_to_fp16 = const()[name = string("module_layers_5_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86030912)))];
-            tensor<fp16, [1024]> module_layers_5_norm_out_bias_to_fp16 = const()[name = string("module_layers_5_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86033024)))];
-            tensor<fp16, [1, 188, 1024]> input_331_cast_fp16 = layer_norm(axes = input_331_axes_0, beta = module_layers_5_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_out_weight_to_fp16, x = input_329_cast_fp16)[name = string("input_331_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_5_norm_out_weight_to_fp16 = const()[name = string("module_layers_5_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82982976)))];
+            tensor<fp16, [1024]> module_layers_5_norm_out_bias_to_fp16 = const()[name = string("module_layers_5_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82985088)))];
+            tensor<fp16, [1, 128, 1024]> input_331_cast_fp16 = layer_norm(axes = input_331_axes_0, beta = module_layers_5_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_out_weight_to_fp16, x = input_329_cast_fp16)[name = string("input_331_cast_fp16")];
             tensor<int32, [1]> input_333_axes_0 = const()[name = string("input_333_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_6_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_6_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86035136)))];
-            tensor<fp16, [1024]> module_layers_6_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_6_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86037248)))];
-            tensor<fp16, [1, 188, 1024]> input_333_cast_fp16 = layer_norm(axes = input_333_axes_0, beta = module_layers_6_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_feed_forward1_weight_to_fp16, x = input_331_cast_fp16)[name = string("input_333_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2073214]> module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86039360))), indices_nonzero_data = tensor<uint6, [2073214]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86563712))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88118720))))[name = string("module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_6_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_6_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82987200)))];
+            tensor<fp16, [1024]> module_layers_6_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_6_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82989312)))];
+            tensor<fp16, [1, 128, 1024]> input_333_cast_fp16 = layer_norm(axes = input_333_axes_0, beta = module_layers_6_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_feed_forward1_weight_to_fp16, x = input_331_cast_fp16)[name = string("input_333_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2073214]> module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82991424))), indices_nonzero_data = tensor<uint6, [2073214]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83515776))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85070784))))[name = string("module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_333_cast_fp16)[name = string("linear_55_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_337_cast_fp16 = silu(x = linear_55_cast_fp16)[name = string("input_337_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1956749]> module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88184320))), indices_nonzero_data = tensor<uint6, [1956749]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88708672))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90176320))))[name = string("module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_333_cast_fp16)[name = string("linear_55_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_337_cast_fp16 = silu(x = linear_55_cast_fp16)[name = string("input_337_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1956749]> module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85136384))), indices_nonzero_data = tensor<uint6, [1956749]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85660736))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87128384))))[name = string("module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_56_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_337_cast_fp16)[name = string("linear_56_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_56_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_337_cast_fp16)[name = string("linear_56_cast_fp16")];
             fp16 var_1210_to_fp16 = const()[name = string("op_1210_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_1211_cast_fp16 = mul(x = linear_56_cast_fp16, y = var_1210_to_fp16)[name = string("op_1211_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_343_cast_fp16 = add(x = input_331_cast_fp16, y = var_1211_cast_fp16)[name = string("input_343_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_1211_cast_fp16 = mul(x = linear_56_cast_fp16, y = var_1210_to_fp16)[name = string("op_1211_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_343_cast_fp16 = add(x = input_331_cast_fp16, y = var_1211_cast_fp16)[name = string("input_343_cast_fp16")];
             tensor<int32, [1]> query_13_axes_0 = const()[name = string("query_13_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_6_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_6_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90192768)))];
-            tensor<fp16, [1024]> module_layers_6_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_6_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90194880)))];
-            tensor<fp16, [1, 188, 1024]> query_13_cast_fp16 = layer_norm(axes = query_13_axes_0, beta = module_layers_6_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_self_att_weight_to_fp16, x = input_343_cast_fp16)[name = string("query_13_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522973]> module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90196992))), indices_nonzero_data = tensor<uint6, [522973]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90328128))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90720448))))[name = string("module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_6_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_6_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87144832)))];
+            tensor<fp16, [1024]> module_layers_6_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_6_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87146944)))];
+            tensor<fp16, [1, 128, 1024]> query_13_cast_fp16 = layer_norm(axes = query_13_axes_0, beta = module_layers_6_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_self_att_weight_to_fp16, x = input_343_cast_fp16)[name = string("query_13_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522973]> module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87149056))), indices_nonzero_data = tensor<uint6, [522973]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87280192))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87672512))))[name = string("module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_57_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_13_cast_fp16)[name = string("linear_57_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_57_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_13_cast_fp16)[name = string("linear_57_cast_fp16")];
             tensor<int32, [4]> var_1227 = const()[name = string("op_1227"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_37_cast_fp16 = reshape(shape = var_1227, x = linear_57_cast_fp16)[name = string("q_37_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521918]> module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90736896))), indices_nonzero_data = tensor<uint6, [521918]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90868032))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91259584))))[name = string("module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> q_37_cast_fp16 = reshape(shape = var_1227, x = linear_57_cast_fp16)[name = string("q_37_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521918]> module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87688960))), indices_nonzero_data = tensor<uint6, [521918]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87820096))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88211648))))[name = string("module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_58_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_13_cast_fp16)[name = string("linear_58_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_58_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_13_cast_fp16)[name = string("linear_58_cast_fp16")];
             tensor<int32, [4]> var_1231 = const()[name = string("op_1231"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_25_cast_fp16 = reshape(shape = var_1231, x = linear_58_cast_fp16)[name = string("k_25_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520459]> module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91276032))), indices_nonzero_data = tensor<uint6, [520459]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91407168))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91797632))))[name = string("module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> k_25_cast_fp16 = reshape(shape = var_1231, x = linear_58_cast_fp16)[name = string("k_25_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520459]> module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88228096))), indices_nonzero_data = tensor<uint6, [520459]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88359232))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88749696))))[name = string("module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_59_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_13_cast_fp16)[name = string("linear_59_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_59_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_13_cast_fp16)[name = string("linear_59_cast_fp16")];
             tensor<int32, [4]> var_1235 = const()[name = string("op_1235"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_13_cast_fp16 = reshape(shape = var_1235, x = linear_59_cast_fp16)[name = string("v_13_cast_fp16")];
+            tensor<fp16, [1, 128, 8, 128]> v_13_cast_fp16 = reshape(shape = var_1235, x = linear_59_cast_fp16)[name = string("v_13_cast_fp16")];
             tensor<int32, [4]> value_15_perm_0 = const()[name = string("value_15_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_6_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_6_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91814080)))];
-            tensor<fp16, [1, 188, 8, 128]> var_1247_cast_fp16 = add(x = q_37_cast_fp16, y = module_layers_6_self_attn_pos_bias_u_to_fp16)[name = string("op_1247_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_6_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_6_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91816192)))];
-            tensor<fp16, [1, 188, 8, 128]> var_1249_cast_fp16 = add(x = q_37_cast_fp16, y = module_layers_6_self_attn_pos_bias_v_to_fp16)[name = string("op_1249_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_6_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_6_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88766144)))];
+            tensor<fp16, [1, 128, 8, 128]> var_1247_cast_fp16 = add(x = q_37_cast_fp16, y = module_layers_6_self_attn_pos_bias_u_to_fp16)[name = string("op_1247_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_6_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_6_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88768256)))];
+            tensor<fp16, [1, 128, 8, 128]> var_1249_cast_fp16 = add(x = q_37_cast_fp16, y = module_layers_6_self_attn_pos_bias_v_to_fp16)[name = string("op_1249_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_13_perm_0 = const()[name = string("q_with_bias_v_13_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_139_transpose_x_0 = const()[name = string("x_139_transpose_x_0"), val = bool(false)];
             bool x_139_transpose_y_0 = const()[name = string("x_139_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_1251_to_fp16 = const()[name = string("op_1251_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91818304)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_13_cast_fp16 = transpose(perm = q_with_bias_v_13_perm_0, x = var_1249_cast_fp16)[name = string("transpose_270")];
-            tensor<fp16, [1, 8, 188, 375]> x_139_cast_fp16 = matmul(transpose_x = x_139_transpose_x_0, transpose_y = x_139_transpose_y_0, x = q_with_bias_v_13_cast_fp16, y = var_1251_to_fp16)[name = string("x_139_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 255]> var_1251_to_fp16 = const()[name = string("op_1251_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(310555648)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_13_cast_fp16 = transpose(perm = q_with_bias_v_13_perm_0, x = var_1249_cast_fp16)[name = string("transpose_270")];
+            tensor<fp16, [1, 8, 128, 255]> x_139_cast_fp16 = matmul(transpose_x = x_139_transpose_x_0, transpose_y = x_139_transpose_y_0, x = q_with_bias_v_13_cast_fp16, y = var_1251_to_fp16)[name = string("x_139_cast_fp16")];
             tensor<int32, [8]> x_141_pad_0 = const()[name = string("x_141_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_141_mode_0 = const()[name = string("x_141_mode_0"), val = string("constant")];
             fp16 const_74_to_fp16 = const()[name = string("const_74_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_141_cast_fp16 = pad(constant_val = const_74_to_fp16, mode = x_141_mode_0, pad = x_141_pad_0, x = x_139_cast_fp16)[name = string("x_141_cast_fp16")];
-            tensor<int32, [4]> var_1259 = const()[name = string("op_1259"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_143_cast_fp16 = reshape(shape = var_1259, x = x_141_cast_fp16)[name = string("x_143_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 256]> x_141_cast_fp16 = pad(constant_val = const_74_to_fp16, mode = x_141_mode_0, pad = x_141_pad_0, x = x_139_cast_fp16)[name = string("x_141_cast_fp16")];
+            tensor<int32, [4]> var_1259 = const()[name = string("op_1259"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_143_cast_fp16 = reshape(shape = var_1259, x = x_141_cast_fp16)[name = string("x_143_cast_fp16")];
             tensor<int32, [4]> var_1263_begin_0 = const()[name = string("op_1263_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_1263_end_0 = const()[name = string("op_1263_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_1263_end_0 = const()[name = string("op_1263_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
             tensor<bool, [4]> var_1263_end_mask_0 = const()[name = string("op_1263_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_1263_cast_fp16 = slice_by_index(begin = var_1263_begin_0, end = var_1263_end_0, end_mask = var_1263_end_mask_0, x = x_143_cast_fp16)[name = string("op_1263_cast_fp16")];
-            tensor<int32, [4]> var_1264 = const()[name = string("op_1264"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_25_cast_fp16 = reshape(shape = var_1264, x = var_1263_cast_fp16)[name = string("matrix_bd_25_cast_fp16")];
+            tensor<fp16, [1, 8, 255, 128]> var_1263_cast_fp16 = slice_by_index(begin = var_1263_begin_0, end = var_1263_end_0, end_mask = var_1263_end_mask_0, x = x_143_cast_fp16)[name = string("op_1263_cast_fp16")];
+            tensor<int32, [4]> var_1264 = const()[name = string("op_1264"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_25_cast_fp16 = reshape(shape = var_1264, x = var_1263_cast_fp16)[name = string("matrix_bd_25_cast_fp16")];
             bool matrix_ac_13_transpose_x_0 = const()[name = string("matrix_ac_13_transpose_x_0"), val = bool(false)];
             bool matrix_ac_13_transpose_y_0 = const()[name = string("matrix_ac_13_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_108_perm_0 = const()[name = string("transpose_108_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_109_perm_0 = const()[name = string("transpose_109_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_109 = transpose(perm = transpose_109_perm_0, x = k_25_cast_fp16)[name = string("transpose_268")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_108 = transpose(perm = transpose_108_perm_0, x = var_1247_cast_fp16)[name = string("transpose_269")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_13_cast_fp16 = matmul(transpose_x = matrix_ac_13_transpose_x_0, transpose_y = matrix_ac_13_transpose_y_0, x = transpose_108, y = transpose_109)[name = string("matrix_ac_13_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_109 = transpose(perm = transpose_109_perm_0, x = k_25_cast_fp16)[name = string("transpose_268")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_108 = transpose(perm = transpose_108_perm_0, x = var_1247_cast_fp16)[name = string("transpose_269")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_13_cast_fp16 = matmul(transpose_x = matrix_ac_13_transpose_x_0, transpose_y = matrix_ac_13_transpose_y_0, x = transpose_108, y = transpose_109)[name = string("matrix_ac_13_cast_fp16")];
             tensor<int32, [4]> matrix_bd_27_begin_0 = const()[name = string("matrix_bd_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_27_end_0 = const()[name = string("matrix_bd_27_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_27_end_0 = const()[name = string("matrix_bd_27_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
             tensor<bool, [4]> matrix_bd_27_end_mask_0 = const()[name = string("matrix_bd_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_27_cast_fp16 = slice_by_index(begin = matrix_bd_27_begin_0, end = matrix_bd_27_end_0, end_mask = matrix_bd_27_end_mask_0, x = matrix_bd_25_cast_fp16)[name = string("matrix_bd_27_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_1273_cast_fp16 = add(x = matrix_ac_13_cast_fp16, y = matrix_bd_27_cast_fp16)[name = string("op_1273_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_27_cast_fp16 = slice_by_index(begin = matrix_bd_27_begin_0, end = matrix_bd_27_end_0, end_mask = matrix_bd_27_end_mask_0, x = matrix_bd_25_cast_fp16)[name = string("matrix_bd_27_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_1273_cast_fp16 = add(x = matrix_ac_13_cast_fp16, y = matrix_bd_27_cast_fp16)[name = string("op_1273_cast_fp16")];
             fp16 _inversed_scores_25_y_0_to_fp16 = const()[name = string("_inversed_scores_25_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_25_cast_fp16 = mul(x = var_1273_cast_fp16, y = _inversed_scores_25_y_0_to_fp16)[name = string("_inversed_scores_25_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_27_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_25_cast_fp16, cond = mask_3)[name = string("scores_27_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_1279_cast_fp16 = softmax(axis = var_30, x = scores_27_cast_fp16)[name = string("op_1279_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_345_cast_fp16 = select(a = var_11_to_fp16, b = var_1279_cast_fp16, cond = mask_3)[name = string("input_345_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_25_cast_fp16 = mul(x = var_1273_cast_fp16, y = _inversed_scores_25_y_0_to_fp16)[name = string("_inversed_scores_25_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_27_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_25_cast_fp16, cond = mask_3)[name = string("scores_27_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_1279_cast_fp16 = softmax(axis = var_30, x = scores_27_cast_fp16)[name = string("op_1279_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_345_cast_fp16 = select(a = var_11_to_fp16, b = var_1279_cast_fp16, cond = mask_3)[name = string("input_345_cast_fp16")];
             bool x_145_transpose_x_0 = const()[name = string("x_145_transpose_x_0"), val = bool(false)];
             bool x_145_transpose_y_0 = const()[name = string("x_145_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_15_cast_fp16 = transpose(perm = value_15_perm_0, x = v_13_cast_fp16)[name = string("transpose_267")];
-            tensor<fp16, [1, 8, 188, 128]> x_145_cast_fp16 = matmul(transpose_x = x_145_transpose_x_0, transpose_y = x_145_transpose_y_0, x = input_345_cast_fp16, y = value_15_cast_fp16)[name = string("x_145_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> value_15_cast_fp16 = transpose(perm = value_15_perm_0, x = v_13_cast_fp16)[name = string("transpose_267")];
+            tensor<fp16, [1, 8, 128, 128]> x_145_cast_fp16 = matmul(transpose_x = x_145_transpose_x_0, transpose_y = x_145_transpose_y_0, x = input_345_cast_fp16, y = value_15_cast_fp16)[name = string("x_145_cast_fp16")];
             tensor<int32, [4]> var_1283_perm_0 = const()[name = string("op_1283_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_1284 = const()[name = string("op_1284"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_1283_cast_fp16 = transpose(perm = var_1283_perm_0, x = x_145_cast_fp16)[name = string("transpose_266")];
-            tensor<fp16, [1, 188, 1024]> input_347_cast_fp16 = reshape(shape = var_1284, x = var_1283_cast_fp16)[name = string("input_347_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524874]> module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92586368))), indices_nonzero_data = tensor<uint6, [524874]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92717504))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93111232))))[name = string("module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> var_1283_cast_fp16 = transpose(perm = var_1283_perm_0, x = x_145_cast_fp16)[name = string("transpose_266")];
+            tensor<fp16, [1, 128, 1024]> input_347_cast_fp16 = reshape(shape = var_1284, x = var_1283_cast_fp16)[name = string("input_347_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524874]> module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89030528))), indices_nonzero_data = tensor<uint6, [524874]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89161664))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89555392))))[name = string("module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_61_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_347_cast_fp16)[name = string("linear_61_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_351_cast_fp16 = add(x = input_343_cast_fp16, y = linear_61_cast_fp16)[name = string("input_351_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_61_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_347_cast_fp16)[name = string("linear_61_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_351_cast_fp16 = add(x = input_343_cast_fp16, y = linear_61_cast_fp16)[name = string("input_351_cast_fp16")];
             tensor<int32, [1]> x_149_axes_0 = const()[name = string("x_149_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_6_norm_conv_weight_to_fp16 = const()[name = string("module_layers_6_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93127680)))];
-            tensor<fp16, [1024]> module_layers_6_norm_conv_bias_to_fp16 = const()[name = string("module_layers_6_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93129792)))];
-            tensor<fp16, [1, 188, 1024]> x_149_cast_fp16 = layer_norm(axes = x_149_axes_0, beta = module_layers_6_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_conv_weight_to_fp16, x = input_351_cast_fp16)[name = string("x_149_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_6_norm_conv_weight_to_fp16 = const()[name = string("module_layers_6_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89571840)))];
+            tensor<fp16, [1024]> module_layers_6_norm_conv_bias_to_fp16 = const()[name = string("module_layers_6_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89573952)))];
+            tensor<fp16, [1, 128, 1024]> x_149_cast_fp16 = layer_norm(axes = x_149_axes_0, beta = module_layers_6_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_conv_weight_to_fp16, x = input_351_cast_fp16)[name = string("x_149_cast_fp16")];
             tensor<int32, [3]> input_353_perm_0 = const()[name = string("input_353_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_355_pad_type_0 = const()[name = string("input_355_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_355_strides_0 = const()[name = string("input_355_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_355_pad_0 = const()[name = string("input_355_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_355_dilations_0 = const()[name = string("input_355_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_355_groups_0 = const()[name = string("input_355_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1048536]> module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93131904))), indices_nonzero_data = tensor<uint6, [1048536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93394112))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94180608))))[name = string("module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1048536]> module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89576064))), indices_nonzero_data = tensor<uint6, [1048536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89838272))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90624768))))[name = string("module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_353_cast_fp16 = transpose(perm = input_353_perm_0, x = x_149_cast_fp16)[name = string("transpose_265")];
-            tensor<fp16, [1, 2048, 188]> input_355_cast_fp16 = conv(dilations = input_355_dilations_0, groups = input_355_groups_0, pad = input_355_pad_0, pad_type = input_355_pad_type_0, strides = input_355_strides_0, weight = module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_353_cast_fp16)[name = string("input_355_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_353_cast_fp16 = transpose(perm = input_353_perm_0, x = x_149_cast_fp16)[name = string("transpose_265")];
+            tensor<fp16, [1, 2048, 128]> input_355_cast_fp16 = conv(dilations = input_355_dilations_0, groups = input_355_groups_0, pad = input_355_pad_0, pad_type = input_355_pad_type_0, strides = input_355_strides_0, weight = module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_353_cast_fp16)[name = string("input_355_cast_fp16")];
             int32 x_151_split_num_splits_0 = const()[name = string("x_151_split_num_splits_0"), val = int32(2)];
             int32 x_151_split_axis_0 = const()[name = string("x_151_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_151_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_151_split_cast_fp16_1 = split(axis = x_151_split_axis_0, num_splits = x_151_split_num_splits_0, x = input_355_cast_fp16)[name = string("x_151_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_151_split_1_sigmoid_cast_fp16 = sigmoid(x = x_151_split_cast_fp16_1)[name = string("x_151_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_151_cast_fp16 = mul(x = x_151_split_cast_fp16_0, y = x_151_split_1_sigmoid_cast_fp16)[name = string("x_151_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_357_cast_fp16 = select(a = var_11_to_fp16, b = x_151_cast_fp16, cond = var_328)[name = string("input_357_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_151_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_151_split_cast_fp16_1 = split(axis = x_151_split_axis_0, num_splits = x_151_split_num_splits_0, x = input_355_cast_fp16)[name = string("x_151_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_151_split_1_sigmoid_cast_fp16 = sigmoid(x = x_151_split_cast_fp16_1)[name = string("x_151_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_151_cast_fp16 = mul(x = x_151_split_cast_fp16_0, y = x_151_split_1_sigmoid_cast_fp16)[name = string("x_151_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_357_cast_fp16 = select(a = var_11_to_fp16, b = x_151_cast_fp16, cond = var_328)[name = string("input_357_cast_fp16")];
             tensor<int32, [6]> input_359_pad_0 = const()[name = string("input_359_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_359_mode_0 = const()[name = string("input_359_mode_0"), val = string("constant")];
             fp16 const_77_to_fp16 = const()[name = string("const_77_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_359_cast_fp16 = pad(constant_val = const_77_to_fp16, mode = input_359_mode_0, pad = input_359_pad_0, x = input_357_cast_fp16)[name = string("input_359_cast_fp16")];
+            tensor<fp16, [1, 1024, 136]> input_359_cast_fp16 = pad(constant_val = const_77_to_fp16, mode = input_359_mode_0, pad = input_359_pad_0, x = input_357_cast_fp16)[name = string("input_359_cast_fp16")];
             string input_361_pad_type_0 = const()[name = string("input_361_pad_type_0"), val = string("valid")];
             int32 input_361_groups_0 = const()[name = string("input_361_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_361_strides_0 = const()[name = string("input_361_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_361_pad_0 = const()[name = string("input_361_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_361_dilations_0 = const()[name = string("input_361_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_260_to_fp16 = const()[name = string("const_260_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94213440)))];
-            tensor<fp16, [1024]> const_261_to_fp16 = const()[name = string("const_261_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94231936)))];
-            tensor<fp16, [1, 1024, 188]> input_363_cast_fp16 = conv(bias = const_261_to_fp16, dilations = input_361_dilations_0, groups = input_361_groups_0, pad = input_361_pad_0, pad_type = input_361_pad_type_0, strides = input_361_strides_0, weight = const_260_to_fp16, x = input_359_cast_fp16)[name = string("input_363_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_365_cast_fp16 = silu(x = input_363_cast_fp16)[name = string("input_365_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_260_to_fp16 = const()[name = string("const_260_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90657600)))];
+            tensor<fp16, [1024]> const_261_to_fp16 = const()[name = string("const_261_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90676096)))];
+            tensor<fp16, [1, 1024, 128]> input_363_cast_fp16 = conv(bias = const_261_to_fp16, dilations = input_361_dilations_0, groups = input_361_groups_0, pad = input_361_pad_0, pad_type = input_361_pad_type_0, strides = input_361_strides_0, weight = const_260_to_fp16, x = input_359_cast_fp16)[name = string("input_363_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_365_cast_fp16 = silu(x = input_363_cast_fp16)[name = string("input_365_cast_fp16")];
             string x_153_pad_type_0 = const()[name = string("x_153_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_153_strides_0 = const()[name = string("x_153_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_153_pad_0 = const()[name = string("x_153_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_153_dilations_0 = const()[name = string("x_153_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_153_groups_0 = const()[name = string("x_153_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519645]> module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94234048))), indices_nonzero_data = tensor<uint6, [519645]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94365184))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94755008))))[name = string("module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519645]> module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90678208))), indices_nonzero_data = tensor<uint6, [519645]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90809344))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91199168))))[name = string("module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_153_cast_fp16 = conv(dilations = x_153_dilations_0, groups = x_153_groups_0, pad = x_153_pad_0, pad_type = x_153_pad_type_0, strides = x_153_strides_0, weight = module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_365_cast_fp16)[name = string("x_153_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_153_cast_fp16 = conv(dilations = x_153_dilations_0, groups = x_153_groups_0, pad = x_153_pad_0, pad_type = x_153_pad_type_0, strides = x_153_strides_0, weight = module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_365_cast_fp16)[name = string("x_153_cast_fp16")];
             tensor<int32, [3]> input_367_perm_0 = const()[name = string("input_367_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_367_cast_fp16 = transpose(perm = input_367_perm_0, x = x_153_cast_fp16)[name = string("transpose_264")];
-            tensor<fp16, [1, 188, 1024]> input_369_cast_fp16 = add(x = input_351_cast_fp16, y = input_367_cast_fp16)[name = string("input_369_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_367_cast_fp16 = transpose(perm = input_367_perm_0, x = x_153_cast_fp16)[name = string("transpose_264")];
+            tensor<fp16, [1, 128, 1024]> input_369_cast_fp16 = add(x = input_351_cast_fp16, y = input_367_cast_fp16)[name = string("input_369_cast_fp16")];
             tensor<int32, [1]> input_371_axes_0 = const()[name = string("input_371_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_6_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_6_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94771456)))];
-            tensor<fp16, [1024]> module_layers_6_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_6_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94773568)))];
-            tensor<fp16, [1, 188, 1024]> input_371_cast_fp16 = layer_norm(axes = input_371_axes_0, beta = module_layers_6_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_feed_forward2_weight_to_fp16, x = input_369_cast_fp16)[name = string("input_371_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2063870]> module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94775680))), indices_nonzero_data = tensor<uint6, [2063870]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95300032))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96848000))))[name = string("module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_6_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_6_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91215616)))];
+            tensor<fp16, [1024]> module_layers_6_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_6_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91217728)))];
+            tensor<fp16, [1, 128, 1024]> input_371_cast_fp16 = layer_norm(axes = input_371_axes_0, beta = module_layers_6_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_feed_forward2_weight_to_fp16, x = input_369_cast_fp16)[name = string("input_371_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2063870]> module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91219840))), indices_nonzero_data = tensor<uint6, [2063870]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91744192))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93292160))))[name = string("module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_62_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_371_cast_fp16)[name = string("linear_62_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_375_cast_fp16 = silu(x = linear_62_cast_fp16)[name = string("input_375_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2019355]> module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96913600))), indices_nonzero_data = tensor<uint6, [2019355]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97437952))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98952576))))[name = string("module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_62_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_371_cast_fp16)[name = string("linear_62_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_375_cast_fp16 = silu(x = linear_62_cast_fp16)[name = string("input_375_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2019355]> module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93357760))), indices_nonzero_data = tensor<uint6, [2019355]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93882112))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95396736))))[name = string("module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_63_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_375_cast_fp16)[name = string("linear_63_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_63_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_375_cast_fp16)[name = string("linear_63_cast_fp16")];
             fp16 var_1344_to_fp16 = const()[name = string("op_1344_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_1345_cast_fp16 = mul(x = linear_63_cast_fp16, y = var_1344_to_fp16)[name = string("op_1345_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_381_cast_fp16 = add(x = input_369_cast_fp16, y = var_1345_cast_fp16)[name = string("input_381_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_1345_cast_fp16 = mul(x = linear_63_cast_fp16, y = var_1344_to_fp16)[name = string("op_1345_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_381_cast_fp16 = add(x = input_369_cast_fp16, y = var_1345_cast_fp16)[name = string("input_381_cast_fp16")];
             tensor<int32, [1]> input_383_axes_0 = const()[name = string("input_383_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_6_norm_out_weight_to_fp16 = const()[name = string("module_layers_6_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98969024)))];
-            tensor<fp16, [1024]> module_layers_6_norm_out_bias_to_fp16 = const()[name = string("module_layers_6_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98971136)))];
-            tensor<fp16, [1, 188, 1024]> input_383_cast_fp16 = layer_norm(axes = input_383_axes_0, beta = module_layers_6_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_out_weight_to_fp16, x = input_381_cast_fp16)[name = string("input_383_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_6_norm_out_weight_to_fp16 = const()[name = string("module_layers_6_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95413184)))];
+            tensor<fp16, [1024]> module_layers_6_norm_out_bias_to_fp16 = const()[name = string("module_layers_6_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95415296)))];
+            tensor<fp16, [1, 128, 1024]> input_383_cast_fp16 = layer_norm(axes = input_383_axes_0, beta = module_layers_6_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_out_weight_to_fp16, x = input_381_cast_fp16)[name = string("input_383_cast_fp16")];
             tensor<int32, [1]> input_385_axes_0 = const()[name = string("input_385_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_7_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_7_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98973248)))];
-            tensor<fp16, [1024]> module_layers_7_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_7_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98975360)))];
-            tensor<fp16, [1, 188, 1024]> input_385_cast_fp16 = layer_norm(axes = input_385_axes_0, beta = module_layers_7_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_feed_forward1_weight_to_fp16, x = input_383_cast_fp16)[name = string("input_385_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2072252]> module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98977472))), indices_nonzero_data = tensor<uint6, [2072252]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99501824))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101056128))))[name = string("module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_7_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_7_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95417408)))];
+            tensor<fp16, [1024]> module_layers_7_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_7_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95419520)))];
+            tensor<fp16, [1, 128, 1024]> input_385_cast_fp16 = layer_norm(axes = input_385_axes_0, beta = module_layers_7_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_feed_forward1_weight_to_fp16, x = input_383_cast_fp16)[name = string("input_385_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2072252]> module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95421632))), indices_nonzero_data = tensor<uint6, [2072252]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95945984))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97500288))))[name = string("module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_64_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_385_cast_fp16)[name = string("linear_64_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_389_cast_fp16 = silu(x = linear_64_cast_fp16)[name = string("input_389_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1978853]> module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101121728))), indices_nonzero_data = tensor<uint6, [1978853]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101646080))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103130304))))[name = string("module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_64_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_385_cast_fp16)[name = string("linear_64_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_389_cast_fp16 = silu(x = linear_64_cast_fp16)[name = string("input_389_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1978853]> module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97565888))), indices_nonzero_data = tensor<uint6, [1978853]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98090240))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99574464))))[name = string("module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_65_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_389_cast_fp16)[name = string("linear_65_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_65_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_389_cast_fp16)[name = string("linear_65_cast_fp16")];
             fp16 var_1373_to_fp16 = const()[name = string("op_1373_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_1374_cast_fp16 = mul(x = linear_65_cast_fp16, y = var_1373_to_fp16)[name = string("op_1374_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_395_cast_fp16 = add(x = input_383_cast_fp16, y = var_1374_cast_fp16)[name = string("input_395_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_1374_cast_fp16 = mul(x = linear_65_cast_fp16, y = var_1373_to_fp16)[name = string("op_1374_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_395_cast_fp16 = add(x = input_383_cast_fp16, y = var_1374_cast_fp16)[name = string("input_395_cast_fp16")];
             tensor<int32, [1]> query_15_axes_0 = const()[name = string("query_15_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_7_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_7_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103146752)))];
-            tensor<fp16, [1024]> module_layers_7_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_7_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103148864)))];
-            tensor<fp16, [1, 188, 1024]> query_15_cast_fp16 = layer_norm(axes = query_15_axes_0, beta = module_layers_7_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_self_att_weight_to_fp16, x = input_395_cast_fp16)[name = string("query_15_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522949]> module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103150976))), indices_nonzero_data = tensor<uint6, [522949]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103282112))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103674432))))[name = string("module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_7_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_7_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99590912)))];
+            tensor<fp16, [1024]> module_layers_7_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_7_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99593024)))];
+            tensor<fp16, [1, 128, 1024]> query_15_cast_fp16 = layer_norm(axes = query_15_axes_0, beta = module_layers_7_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_self_att_weight_to_fp16, x = input_395_cast_fp16)[name = string("query_15_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522949]> module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99595136))), indices_nonzero_data = tensor<uint6, [522949]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99726272))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100118592))))[name = string("module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_66_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_15_cast_fp16)[name = string("linear_66_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_66_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_15_cast_fp16)[name = string("linear_66_cast_fp16")];
             tensor<int32, [4]> var_1390 = const()[name = string("op_1390"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_43_cast_fp16 = reshape(shape = var_1390, x = linear_66_cast_fp16)[name = string("q_43_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [515713]> module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103690880))), indices_nonzero_data = tensor<uint6, [515713]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103822016))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104208896))))[name = string("module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> q_43_cast_fp16 = reshape(shape = var_1390, x = linear_66_cast_fp16)[name = string("q_43_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [515713]> module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100135040))), indices_nonzero_data = tensor<uint6, [515713]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100266176))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100653056))))[name = string("module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_67_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_15_cast_fp16)[name = string("linear_67_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_67_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_15_cast_fp16)[name = string("linear_67_cast_fp16")];
             tensor<int32, [4]> var_1394 = const()[name = string("op_1394"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_29_cast_fp16 = reshape(shape = var_1394, x = linear_67_cast_fp16)[name = string("k_29_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [507764]> module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104225344))), indices_nonzero_data = tensor<uint6, [507764]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104356480))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104737408))))[name = string("module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> k_29_cast_fp16 = reshape(shape = var_1394, x = linear_67_cast_fp16)[name = string("k_29_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [507764]> module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100669504))), indices_nonzero_data = tensor<uint6, [507764]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100800640))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101181568))))[name = string("module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_68_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_15_cast_fp16)[name = string("linear_68_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_68_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_15_cast_fp16)[name = string("linear_68_cast_fp16")];
             tensor<int32, [4]> var_1398 = const()[name = string("op_1398"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_15_cast_fp16 = reshape(shape = var_1398, x = linear_68_cast_fp16)[name = string("v_15_cast_fp16")];
+            tensor<fp16, [1, 128, 8, 128]> v_15_cast_fp16 = reshape(shape = var_1398, x = linear_68_cast_fp16)[name = string("v_15_cast_fp16")];
             tensor<int32, [4]> value_17_perm_0 = const()[name = string("value_17_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_7_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_7_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104753856)))];
-            tensor<fp16, [1, 188, 8, 128]> var_1410_cast_fp16 = add(x = q_43_cast_fp16, y = module_layers_7_self_attn_pos_bias_u_to_fp16)[name = string("op_1410_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_7_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_7_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104755968)))];
-            tensor<fp16, [1, 188, 8, 128]> var_1412_cast_fp16 = add(x = q_43_cast_fp16, y = module_layers_7_self_attn_pos_bias_v_to_fp16)[name = string("op_1412_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_7_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_7_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101198016)))];
+            tensor<fp16, [1, 128, 8, 128]> var_1410_cast_fp16 = add(x = q_43_cast_fp16, y = module_layers_7_self_attn_pos_bias_u_to_fp16)[name = string("op_1410_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_7_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_7_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101200128)))];
+            tensor<fp16, [1, 128, 8, 128]> var_1412_cast_fp16 = add(x = q_43_cast_fp16, y = module_layers_7_self_attn_pos_bias_v_to_fp16)[name = string("op_1412_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_15_perm_0 = const()[name = string("q_with_bias_v_15_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_161_transpose_x_0 = const()[name = string("x_161_transpose_x_0"), val = bool(false)];
             bool x_161_transpose_y_0 = const()[name = string("x_161_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_1414_to_fp16 = const()[name = string("op_1414_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104758080)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_15_cast_fp16 = transpose(perm = q_with_bias_v_15_perm_0, x = var_1412_cast_fp16)[name = string("transpose_263")];
-            tensor<fp16, [1, 8, 188, 375]> x_161_cast_fp16 = matmul(transpose_x = x_161_transpose_x_0, transpose_y = x_161_transpose_y_0, x = q_with_bias_v_15_cast_fp16, y = var_1414_to_fp16)[name = string("x_161_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 255]> var_1414_to_fp16 = const()[name = string("op_1414_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311077952)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_15_cast_fp16 = transpose(perm = q_with_bias_v_15_perm_0, x = var_1412_cast_fp16)[name = string("transpose_263")];
+            tensor<fp16, [1, 8, 128, 255]> x_161_cast_fp16 = matmul(transpose_x = x_161_transpose_x_0, transpose_y = x_161_transpose_y_0, x = q_with_bias_v_15_cast_fp16, y = var_1414_to_fp16)[name = string("x_161_cast_fp16")];
             tensor<int32, [8]> x_163_pad_0 = const()[name = string("x_163_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_163_mode_0 = const()[name = string("x_163_mode_0"), val = string("constant")];
             fp16 const_84_to_fp16 = const()[name = string("const_84_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_163_cast_fp16 = pad(constant_val = const_84_to_fp16, mode = x_163_mode_0, pad = x_163_pad_0, x = x_161_cast_fp16)[name = string("x_163_cast_fp16")];
-            tensor<int32, [4]> var_1422 = const()[name = string("op_1422"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_165_cast_fp16 = reshape(shape = var_1422, x = x_163_cast_fp16)[name = string("x_165_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 256]> x_163_cast_fp16 = pad(constant_val = const_84_to_fp16, mode = x_163_mode_0, pad = x_163_pad_0, x = x_161_cast_fp16)[name = string("x_163_cast_fp16")];
+            tensor<int32, [4]> var_1422 = const()[name = string("op_1422"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_165_cast_fp16 = reshape(shape = var_1422, x = x_163_cast_fp16)[name = string("x_165_cast_fp16")];
             tensor<int32, [4]> var_1426_begin_0 = const()[name = string("op_1426_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_1426_end_0 = const()[name = string("op_1426_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_1426_end_0 = const()[name = string("op_1426_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
             tensor<bool, [4]> var_1426_end_mask_0 = const()[name = string("op_1426_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_1426_cast_fp16 = slice_by_index(begin = var_1426_begin_0, end = var_1426_end_0, end_mask = var_1426_end_mask_0, x = x_165_cast_fp16)[name = string("op_1426_cast_fp16")];
-            tensor<int32, [4]> var_1427 = const()[name = string("op_1427"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_29_cast_fp16 = reshape(shape = var_1427, x = var_1426_cast_fp16)[name = string("matrix_bd_29_cast_fp16")];
+            tensor<fp16, [1, 8, 255, 128]> var_1426_cast_fp16 = slice_by_index(begin = var_1426_begin_0, end = var_1426_end_0, end_mask = var_1426_end_mask_0, x = x_165_cast_fp16)[name = string("op_1426_cast_fp16")];
+            tensor<int32, [4]> var_1427 = const()[name = string("op_1427"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_29_cast_fp16 = reshape(shape = var_1427, x = var_1426_cast_fp16)[name = string("matrix_bd_29_cast_fp16")];
             bool matrix_ac_15_transpose_x_0 = const()[name = string("matrix_ac_15_transpose_x_0"), val = bool(false)];
             bool matrix_ac_15_transpose_y_0 = const()[name = string("matrix_ac_15_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_110_perm_0 = const()[name = string("transpose_110_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_111_perm_0 = const()[name = string("transpose_111_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_111 = transpose(perm = transpose_111_perm_0, x = k_29_cast_fp16)[name = string("transpose_261")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_110 = transpose(perm = transpose_110_perm_0, x = var_1410_cast_fp16)[name = string("transpose_262")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_15_cast_fp16 = matmul(transpose_x = matrix_ac_15_transpose_x_0, transpose_y = matrix_ac_15_transpose_y_0, x = transpose_110, y = transpose_111)[name = string("matrix_ac_15_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_111 = transpose(perm = transpose_111_perm_0, x = k_29_cast_fp16)[name = string("transpose_261")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_110 = transpose(perm = transpose_110_perm_0, x = var_1410_cast_fp16)[name = string("transpose_262")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_15_cast_fp16 = matmul(transpose_x = matrix_ac_15_transpose_x_0, transpose_y = matrix_ac_15_transpose_y_0, x = transpose_110, y = transpose_111)[name = string("matrix_ac_15_cast_fp16")];
             tensor<int32, [4]> matrix_bd_31_begin_0 = const()[name = string("matrix_bd_31_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_31_end_0 = const()[name = string("matrix_bd_31_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_31_end_0 = const()[name = string("matrix_bd_31_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
             tensor<bool, [4]> matrix_bd_31_end_mask_0 = const()[name = string("matrix_bd_31_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_31_cast_fp16 = slice_by_index(begin = matrix_bd_31_begin_0, end = matrix_bd_31_end_0, end_mask = matrix_bd_31_end_mask_0, x = matrix_bd_29_cast_fp16)[name = string("matrix_bd_31_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_1436_cast_fp16 = add(x = matrix_ac_15_cast_fp16, y = matrix_bd_31_cast_fp16)[name = string("op_1436_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_31_cast_fp16 = slice_by_index(begin = matrix_bd_31_begin_0, end = matrix_bd_31_end_0, end_mask = matrix_bd_31_end_mask_0, x = matrix_bd_29_cast_fp16)[name = string("matrix_bd_31_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_1436_cast_fp16 = add(x = matrix_ac_15_cast_fp16, y = matrix_bd_31_cast_fp16)[name = string("op_1436_cast_fp16")];
             fp16 _inversed_scores_29_y_0_to_fp16 = const()[name = string("_inversed_scores_29_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_29_cast_fp16 = mul(x = var_1436_cast_fp16, y = _inversed_scores_29_y_0_to_fp16)[name = string("_inversed_scores_29_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_31_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_29_cast_fp16, cond = mask_3)[name = string("scores_31_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_1442_cast_fp16 = softmax(axis = var_30, x = scores_31_cast_fp16)[name = string("op_1442_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_397_cast_fp16 = select(a = var_11_to_fp16, b = var_1442_cast_fp16, cond = mask_3)[name = string("input_397_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_29_cast_fp16 = mul(x = var_1436_cast_fp16, y = _inversed_scores_29_y_0_to_fp16)[name = string("_inversed_scores_29_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_31_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_29_cast_fp16, cond = mask_3)[name = string("scores_31_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_1442_cast_fp16 = softmax(axis = var_30, x = scores_31_cast_fp16)[name = string("op_1442_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_397_cast_fp16 = select(a = var_11_to_fp16, b = var_1442_cast_fp16, cond = mask_3)[name = string("input_397_cast_fp16")];
             bool x_167_transpose_x_0 = const()[name = string("x_167_transpose_x_0"), val = bool(false)];
             bool x_167_transpose_y_0 = const()[name = string("x_167_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_17_cast_fp16 = transpose(perm = value_17_perm_0, x = v_15_cast_fp16)[name = string("transpose_260")];
-            tensor<fp16, [1, 8, 188, 128]> x_167_cast_fp16 = matmul(transpose_x = x_167_transpose_x_0, transpose_y = x_167_transpose_y_0, x = input_397_cast_fp16, y = value_17_cast_fp16)[name = string("x_167_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> value_17_cast_fp16 = transpose(perm = value_17_perm_0, x = v_15_cast_fp16)[name = string("transpose_260")];
+            tensor<fp16, [1, 8, 128, 128]> x_167_cast_fp16 = matmul(transpose_x = x_167_transpose_x_0, transpose_y = x_167_transpose_y_0, x = input_397_cast_fp16, y = value_17_cast_fp16)[name = string("x_167_cast_fp16")];
             tensor<int32, [4]> var_1446_perm_0 = const()[name = string("op_1446_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_1447 = const()[name = string("op_1447"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_1446_cast_fp16 = transpose(perm = var_1446_perm_0, x = x_167_cast_fp16)[name = string("transpose_259")];
-            tensor<fp16, [1, 188, 1024]> input_399_cast_fp16 = reshape(shape = var_1447, x = var_1446_cast_fp16)[name = string("input_399_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525262]> module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105526144))), indices_nonzero_data = tensor<uint6, [525262]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105657280))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106051328))))[name = string("module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> var_1446_cast_fp16 = transpose(perm = var_1446_perm_0, x = x_167_cast_fp16)[name = string("transpose_259")];
+            tensor<fp16, [1, 128, 1024]> input_399_cast_fp16 = reshape(shape = var_1447, x = var_1446_cast_fp16)[name = string("input_399_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525262]> module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101462400))), indices_nonzero_data = tensor<uint6, [525262]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101593536))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101987584))))[name = string("module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_70_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_399_cast_fp16)[name = string("linear_70_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_403_cast_fp16 = add(x = input_395_cast_fp16, y = linear_70_cast_fp16)[name = string("input_403_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_70_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_399_cast_fp16)[name = string("linear_70_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_403_cast_fp16 = add(x = input_395_cast_fp16, y = linear_70_cast_fp16)[name = string("input_403_cast_fp16")];
             tensor<int32, [1]> x_171_axes_0 = const()[name = string("x_171_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_7_norm_conv_weight_to_fp16 = const()[name = string("module_layers_7_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106067776)))];
-            tensor<fp16, [1024]> module_layers_7_norm_conv_bias_to_fp16 = const()[name = string("module_layers_7_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106069888)))];
-            tensor<fp16, [1, 188, 1024]> x_171_cast_fp16 = layer_norm(axes = x_171_axes_0, beta = module_layers_7_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_conv_weight_to_fp16, x = input_403_cast_fp16)[name = string("x_171_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_7_norm_conv_weight_to_fp16 = const()[name = string("module_layers_7_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102004032)))];
+            tensor<fp16, [1024]> module_layers_7_norm_conv_bias_to_fp16 = const()[name = string("module_layers_7_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102006144)))];
+            tensor<fp16, [1, 128, 1024]> x_171_cast_fp16 = layer_norm(axes = x_171_axes_0, beta = module_layers_7_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_conv_weight_to_fp16, x = input_403_cast_fp16)[name = string("x_171_cast_fp16")];
             tensor<int32, [3]> input_405_perm_0 = const()[name = string("input_405_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_407_pad_type_0 = const()[name = string("input_407_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_407_strides_0 = const()[name = string("input_407_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_407_pad_0 = const()[name = string("input_407_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_407_dilations_0 = const()[name = string("input_407_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_407_groups_0 = const()[name = string("input_407_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047562]> module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106072000))), indices_nonzero_data = tensor<uint6, [1047562]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106334208))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107120000))))[name = string("module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047562]> module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102008256))), indices_nonzero_data = tensor<uint6, [1047562]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102270464))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103056256))))[name = string("module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_405_cast_fp16 = transpose(perm = input_405_perm_0, x = x_171_cast_fp16)[name = string("transpose_258")];
-            tensor<fp16, [1, 2048, 188]> input_407_cast_fp16 = conv(dilations = input_407_dilations_0, groups = input_407_groups_0, pad = input_407_pad_0, pad_type = input_407_pad_type_0, strides = input_407_strides_0, weight = module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_405_cast_fp16)[name = string("input_407_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_405_cast_fp16 = transpose(perm = input_405_perm_0, x = x_171_cast_fp16)[name = string("transpose_258")];
+            tensor<fp16, [1, 2048, 128]> input_407_cast_fp16 = conv(dilations = input_407_dilations_0, groups = input_407_groups_0, pad = input_407_pad_0, pad_type = input_407_pad_type_0, strides = input_407_strides_0, weight = module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_405_cast_fp16)[name = string("input_407_cast_fp16")];
             int32 x_173_split_num_splits_0 = const()[name = string("x_173_split_num_splits_0"), val = int32(2)];
             int32 x_173_split_axis_0 = const()[name = string("x_173_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_173_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_173_split_cast_fp16_1 = split(axis = x_173_split_axis_0, num_splits = x_173_split_num_splits_0, x = input_407_cast_fp16)[name = string("x_173_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_173_split_1_sigmoid_cast_fp16 = sigmoid(x = x_173_split_cast_fp16_1)[name = string("x_173_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_173_cast_fp16 = mul(x = x_173_split_cast_fp16_0, y = x_173_split_1_sigmoid_cast_fp16)[name = string("x_173_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_409_cast_fp16 = select(a = var_11_to_fp16, b = x_173_cast_fp16, cond = var_328)[name = string("input_409_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_173_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_173_split_cast_fp16_1 = split(axis = x_173_split_axis_0, num_splits = x_173_split_num_splits_0, x = input_407_cast_fp16)[name = string("x_173_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_173_split_1_sigmoid_cast_fp16 = sigmoid(x = x_173_split_cast_fp16_1)[name = string("x_173_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_173_cast_fp16 = mul(x = x_173_split_cast_fp16_0, y = x_173_split_1_sigmoid_cast_fp16)[name = string("x_173_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_409_cast_fp16 = select(a = var_11_to_fp16, b = x_173_cast_fp16, cond = var_328)[name = string("input_409_cast_fp16")];
             tensor<int32, [6]> input_411_pad_0 = const()[name = string("input_411_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_411_mode_0 = const()[name = string("input_411_mode_0"), val = string("constant")];
             fp16 const_87_to_fp16 = const()[name = string("const_87_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_411_cast_fp16 = pad(constant_val = const_87_to_fp16, mode = input_411_mode_0, pad = input_411_pad_0, x = input_409_cast_fp16)[name = string("input_411_cast_fp16")];
+            tensor<fp16, [1, 1024, 136]> input_411_cast_fp16 = pad(constant_val = const_87_to_fp16, mode = input_411_mode_0, pad = input_411_pad_0, x = input_409_cast_fp16)[name = string("input_411_cast_fp16")];
             string input_413_pad_type_0 = const()[name = string("input_413_pad_type_0"), val = string("valid")];
             int32 input_413_groups_0 = const()[name = string("input_413_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_413_strides_0 = const()[name = string("input_413_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_413_pad_0 = const()[name = string("input_413_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_413_dilations_0 = const()[name = string("input_413_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_262_to_fp16 = const()[name = string("const_262_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107152832)))];
-            tensor<fp16, [1024]> const_263_to_fp16 = const()[name = string("const_263_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107171328)))];
-            tensor<fp16, [1, 1024, 188]> input_415_cast_fp16 = conv(bias = const_263_to_fp16, dilations = input_413_dilations_0, groups = input_413_groups_0, pad = input_413_pad_0, pad_type = input_413_pad_type_0, strides = input_413_strides_0, weight = const_262_to_fp16, x = input_411_cast_fp16)[name = string("input_415_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_417_cast_fp16 = silu(x = input_415_cast_fp16)[name = string("input_417_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_262_to_fp16 = const()[name = string("const_262_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103089088)))];
+            tensor<fp16, [1024]> const_263_to_fp16 = const()[name = string("const_263_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103107584)))];
+            tensor<fp16, [1, 1024, 128]> input_415_cast_fp16 = conv(bias = const_263_to_fp16, dilations = input_413_dilations_0, groups = input_413_groups_0, pad = input_413_pad_0, pad_type = input_413_pad_type_0, strides = input_413_strides_0, weight = const_262_to_fp16, x = input_411_cast_fp16)[name = string("input_415_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_417_cast_fp16 = silu(x = input_415_cast_fp16)[name = string("input_417_cast_fp16")];
             string x_175_pad_type_0 = const()[name = string("x_175_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_175_strides_0 = const()[name = string("x_175_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_175_pad_0 = const()[name = string("x_175_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_175_dilations_0 = const()[name = string("x_175_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_175_groups_0 = const()[name = string("x_175_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519522]> module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107173440))), indices_nonzero_data = tensor<uint6, [519522]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107304576))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107694336))))[name = string("module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519522]> module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103109696))), indices_nonzero_data = tensor<uint6, [519522]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103240832))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103630592))))[name = string("module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_175_cast_fp16 = conv(dilations = x_175_dilations_0, groups = x_175_groups_0, pad = x_175_pad_0, pad_type = x_175_pad_type_0, strides = x_175_strides_0, weight = module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_417_cast_fp16)[name = string("x_175_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_175_cast_fp16 = conv(dilations = x_175_dilations_0, groups = x_175_groups_0, pad = x_175_pad_0, pad_type = x_175_pad_type_0, strides = x_175_strides_0, weight = module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_417_cast_fp16)[name = string("x_175_cast_fp16")];
             tensor<int32, [3]> input_419_perm_0 = const()[name = string("input_419_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_419_cast_fp16 = transpose(perm = input_419_perm_0, x = x_175_cast_fp16)[name = string("transpose_257")];
-            tensor<fp16, [1, 188, 1024]> input_421_cast_fp16 = add(x = input_403_cast_fp16, y = input_419_cast_fp16)[name = string("input_421_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_419_cast_fp16 = transpose(perm = input_419_perm_0, x = x_175_cast_fp16)[name = string("transpose_257")];
+            tensor<fp16, [1, 128, 1024]> input_421_cast_fp16 = add(x = input_403_cast_fp16, y = input_419_cast_fp16)[name = string("input_421_cast_fp16")];
             tensor<int32, [1]> input_423_axes_0 = const()[name = string("input_423_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_7_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_7_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107710784)))];
-            tensor<fp16, [1024]> module_layers_7_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_7_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107712896)))];
-            tensor<fp16, [1, 188, 1024]> input_423_cast_fp16 = layer_norm(axes = input_423_axes_0, beta = module_layers_7_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_feed_forward2_weight_to_fp16, x = input_421_cast_fp16)[name = string("input_423_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2062964]> module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107715008))), indices_nonzero_data = tensor<uint6, [2062964]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108239360))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109786688))))[name = string("module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_7_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_7_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103647040)))];
+            tensor<fp16, [1024]> module_layers_7_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_7_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103649152)))];
+            tensor<fp16, [1, 128, 1024]> input_423_cast_fp16 = layer_norm(axes = input_423_axes_0, beta = module_layers_7_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_feed_forward2_weight_to_fp16, x = input_421_cast_fp16)[name = string("input_423_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2062964]> module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103651264))), indices_nonzero_data = tensor<uint6, [2062964]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104175616))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105722944))))[name = string("module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_71_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_423_cast_fp16)[name = string("linear_71_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_427_cast_fp16 = silu(x = linear_71_cast_fp16)[name = string("input_427_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2021791]> module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109852288))), indices_nonzero_data = tensor<uint6, [2021791]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110376640))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111893056))))[name = string("module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_71_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_423_cast_fp16)[name = string("linear_71_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_427_cast_fp16 = silu(x = linear_71_cast_fp16)[name = string("input_427_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2021791]> module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105788544))), indices_nonzero_data = tensor<uint6, [2021791]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106312896))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107829312))))[name = string("module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_72_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_427_cast_fp16)[name = string("linear_72_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_72_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_427_cast_fp16)[name = string("linear_72_cast_fp16")];
             fp16 var_1507_to_fp16 = const()[name = string("op_1507_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_1508_cast_fp16 = mul(x = linear_72_cast_fp16, y = var_1507_to_fp16)[name = string("op_1508_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_433_cast_fp16 = add(x = input_421_cast_fp16, y = var_1508_cast_fp16)[name = string("input_433_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_1508_cast_fp16 = mul(x = linear_72_cast_fp16, y = var_1507_to_fp16)[name = string("op_1508_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_433_cast_fp16 = add(x = input_421_cast_fp16, y = var_1508_cast_fp16)[name = string("input_433_cast_fp16")];
             tensor<int32, [1]> input_435_axes_0 = const()[name = string("input_435_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_7_norm_out_weight_to_fp16 = const()[name = string("module_layers_7_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111909504)))];
-            tensor<fp16, [1024]> module_layers_7_norm_out_bias_to_fp16 = const()[name = string("module_layers_7_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111911616)))];
-            tensor<fp16, [1, 188, 1024]> input_435_cast_fp16 = layer_norm(axes = input_435_axes_0, beta = module_layers_7_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_out_weight_to_fp16, x = input_433_cast_fp16)[name = string("input_435_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_7_norm_out_weight_to_fp16 = const()[name = string("module_layers_7_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107845760)))];
+            tensor<fp16, [1024]> module_layers_7_norm_out_bias_to_fp16 = const()[name = string("module_layers_7_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107847872)))];
+            tensor<fp16, [1, 128, 1024]> input_435_cast_fp16 = layer_norm(axes = input_435_axes_0, beta = module_layers_7_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_out_weight_to_fp16, x = input_433_cast_fp16)[name = string("input_435_cast_fp16")];
             tensor<int32, [1]> input_437_axes_0 = const()[name = string("input_437_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_8_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_8_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111913728)))];
-            tensor<fp16, [1024]> module_layers_8_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_8_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111915840)))];
-            tensor<fp16, [1, 188, 1024]> input_437_cast_fp16 = layer_norm(axes = input_437_axes_0, beta = module_layers_8_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_feed_forward1_weight_to_fp16, x = input_435_cast_fp16)[name = string("input_437_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2070011]> module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111917952))), indices_nonzero_data = tensor<uint6, [2070011]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112442304))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113994880))))[name = string("module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_8_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_8_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107849984)))];
+            tensor<fp16, [1024]> module_layers_8_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_8_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107852096)))];
+            tensor<fp16, [1, 128, 1024]> input_437_cast_fp16 = layer_norm(axes = input_437_axes_0, beta = module_layers_8_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_feed_forward1_weight_to_fp16, x = input_435_cast_fp16)[name = string("input_437_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2070011]> module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107854208))), indices_nonzero_data = tensor<uint6, [2070011]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108378560))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109931136))))[name = string("module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_73_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_437_cast_fp16)[name = string("linear_73_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_441_cast_fp16 = silu(x = linear_73_cast_fp16)[name = string("input_441_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1995292]> module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114060480))), indices_nonzero_data = tensor<uint6, [1995292]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114584832))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116081408))))[name = string("module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_73_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_437_cast_fp16)[name = string("linear_73_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_441_cast_fp16 = silu(x = linear_73_cast_fp16)[name = string("input_441_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1995292]> module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109996736))), indices_nonzero_data = tensor<uint6, [1995292]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110521088))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112017664))))[name = string("module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_74_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_441_cast_fp16)[name = string("linear_74_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_74_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_441_cast_fp16)[name = string("linear_74_cast_fp16")];
             fp16 var_1536_to_fp16 = const()[name = string("op_1536_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_1537_cast_fp16 = mul(x = linear_74_cast_fp16, y = var_1536_to_fp16)[name = string("op_1537_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_447_cast_fp16 = add(x = input_435_cast_fp16, y = var_1537_cast_fp16)[name = string("input_447_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_1537_cast_fp16 = mul(x = linear_74_cast_fp16, y = var_1536_to_fp16)[name = string("op_1537_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_447_cast_fp16 = add(x = input_435_cast_fp16, y = var_1537_cast_fp16)[name = string("input_447_cast_fp16")];
             tensor<int32, [1]> query_17_axes_0 = const()[name = string("query_17_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_8_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_8_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116097856)))];
-            tensor<fp16, [1024]> module_layers_8_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_8_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116099968)))];
-            tensor<fp16, [1, 188, 1024]> query_17_cast_fp16 = layer_norm(axes = query_17_axes_0, beta = module_layers_8_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_self_att_weight_to_fp16, x = input_447_cast_fp16)[name = string("query_17_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522805]> module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116102080))), indices_nonzero_data = tensor<uint6, [522805]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116233216))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116625408))))[name = string("module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_8_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_8_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112034112)))];
+            tensor<fp16, [1024]> module_layers_8_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_8_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112036224)))];
+            tensor<fp16, [1, 128, 1024]> query_17_cast_fp16 = layer_norm(axes = query_17_axes_0, beta = module_layers_8_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_self_att_weight_to_fp16, x = input_447_cast_fp16)[name = string("query_17_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522805]> module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112038336))), indices_nonzero_data = tensor<uint6, [522805]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112169472))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112561664))))[name = string("module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_75_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_17_cast_fp16)[name = string("linear_75_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_75_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_17_cast_fp16)[name = string("linear_75_cast_fp16")];
             tensor<int32, [4]> var_1553 = const()[name = string("op_1553"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_49_cast_fp16 = reshape(shape = var_1553, x = linear_75_cast_fp16)[name = string("q_49_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521269]> module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116641856))), indices_nonzero_data = tensor<uint6, [521269]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116772992))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117164032))))[name = string("module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> q_49_cast_fp16 = reshape(shape = var_1553, x = linear_75_cast_fp16)[name = string("q_49_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521269]> module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112578112))), indices_nonzero_data = tensor<uint6, [521269]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112709248))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113100288))))[name = string("module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_76_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_17_cast_fp16)[name = string("linear_76_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_76_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_17_cast_fp16)[name = string("linear_76_cast_fp16")];
             tensor<int32, [4]> var_1557 = const()[name = string("op_1557"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_33_cast_fp16 = reshape(shape = var_1557, x = linear_76_cast_fp16)[name = string("k_33_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [517626]> module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117180480))), indices_nonzero_data = tensor<uint6, [517626]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117311616))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117699904))))[name = string("module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> k_33_cast_fp16 = reshape(shape = var_1557, x = linear_76_cast_fp16)[name = string("k_33_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [517626]> module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113116736))), indices_nonzero_data = tensor<uint6, [517626]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113247872))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113636160))))[name = string("module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_77_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_17_cast_fp16)[name = string("linear_77_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_77_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_17_cast_fp16)[name = string("linear_77_cast_fp16")];
             tensor<int32, [4]> var_1561 = const()[name = string("op_1561"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_17_cast_fp16 = reshape(shape = var_1561, x = linear_77_cast_fp16)[name = string("v_17_cast_fp16")];
+            tensor<fp16, [1, 128, 8, 128]> v_17_cast_fp16 = reshape(shape = var_1561, x = linear_77_cast_fp16)[name = string("v_17_cast_fp16")];
             tensor<int32, [4]> value_19_perm_0 = const()[name = string("value_19_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_8_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_8_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117716352)))];
-            tensor<fp16, [1, 188, 8, 128]> var_1573_cast_fp16 = add(x = q_49_cast_fp16, y = module_layers_8_self_attn_pos_bias_u_to_fp16)[name = string("op_1573_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_8_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_8_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117718464)))];
-            tensor<fp16, [1, 188, 8, 128]> var_1575_cast_fp16 = add(x = q_49_cast_fp16, y = module_layers_8_self_attn_pos_bias_v_to_fp16)[name = string("op_1575_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_8_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_8_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113652608)))];
+            tensor<fp16, [1, 128, 8, 128]> var_1573_cast_fp16 = add(x = q_49_cast_fp16, y = module_layers_8_self_attn_pos_bias_u_to_fp16)[name = string("op_1573_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_8_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_8_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113654720)))];
+            tensor<fp16, [1, 128, 8, 128]> var_1575_cast_fp16 = add(x = q_49_cast_fp16, y = module_layers_8_self_attn_pos_bias_v_to_fp16)[name = string("op_1575_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_17_perm_0 = const()[name = string("q_with_bias_v_17_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_183_transpose_x_0 = const()[name = string("x_183_transpose_x_0"), val = bool(false)];
             bool x_183_transpose_y_0 = const()[name = string("x_183_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_1577_to_fp16 = const()[name = string("op_1577_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117720576)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_17_cast_fp16 = transpose(perm = q_with_bias_v_17_perm_0, x = var_1575_cast_fp16)[name = string("transpose_256")];
-            tensor<fp16, [1, 8, 188, 375]> x_183_cast_fp16 = matmul(transpose_x = x_183_transpose_x_0, transpose_y = x_183_transpose_y_0, x = q_with_bias_v_17_cast_fp16, y = var_1577_to_fp16)[name = string("x_183_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 255]> var_1577_to_fp16 = const()[name = string("op_1577_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311600256)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_17_cast_fp16 = transpose(perm = q_with_bias_v_17_perm_0, x = var_1575_cast_fp16)[name = string("transpose_256")];
+            tensor<fp16, [1, 8, 128, 255]> x_183_cast_fp16 = matmul(transpose_x = x_183_transpose_x_0, transpose_y = x_183_transpose_y_0, x = q_with_bias_v_17_cast_fp16, y = var_1577_to_fp16)[name = string("x_183_cast_fp16")];
             tensor<int32, [8]> x_185_pad_0 = const()[name = string("x_185_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_185_mode_0 = const()[name = string("x_185_mode_0"), val = string("constant")];
             fp16 const_94_to_fp16 = const()[name = string("const_94_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_185_cast_fp16 = pad(constant_val = const_94_to_fp16, mode = x_185_mode_0, pad = x_185_pad_0, x = x_183_cast_fp16)[name = string("x_185_cast_fp16")];
-            tensor<int32, [4]> var_1585 = const()[name = string("op_1585"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_187_cast_fp16 = reshape(shape = var_1585, x = x_185_cast_fp16)[name = string("x_187_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 256]> x_185_cast_fp16 = pad(constant_val = const_94_to_fp16, mode = x_185_mode_0, pad = x_185_pad_0, x = x_183_cast_fp16)[name = string("x_185_cast_fp16")];
+            tensor<int32, [4]> var_1585 = const()[name = string("op_1585"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_187_cast_fp16 = reshape(shape = var_1585, x = x_185_cast_fp16)[name = string("x_187_cast_fp16")];
             tensor<int32, [4]> var_1589_begin_0 = const()[name = string("op_1589_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_1589_end_0 = const()[name = string("op_1589_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_1589_end_0 = const()[name = string("op_1589_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
             tensor<bool, [4]> var_1589_end_mask_0 = const()[name = string("op_1589_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_1589_cast_fp16 = slice_by_index(begin = var_1589_begin_0, end = var_1589_end_0, end_mask = var_1589_end_mask_0, x = x_187_cast_fp16)[name = string("op_1589_cast_fp16")];
-            tensor<int32, [4]> var_1590 = const()[name = string("op_1590"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_33_cast_fp16 = reshape(shape = var_1590, x = var_1589_cast_fp16)[name = string("matrix_bd_33_cast_fp16")];
+            tensor<fp16, [1, 8, 255, 128]> var_1589_cast_fp16 = slice_by_index(begin = var_1589_begin_0, end = var_1589_end_0, end_mask = var_1589_end_mask_0, x = x_187_cast_fp16)[name = string("op_1589_cast_fp16")];
+            tensor<int32, [4]> var_1590 = const()[name = string("op_1590"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_33_cast_fp16 = reshape(shape = var_1590, x = var_1589_cast_fp16)[name = string("matrix_bd_33_cast_fp16")];
             bool matrix_ac_17_transpose_x_0 = const()[name = string("matrix_ac_17_transpose_x_0"), val = bool(false)];
             bool matrix_ac_17_transpose_y_0 = const()[name = string("matrix_ac_17_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_112_perm_0 = const()[name = string("transpose_112_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_113_perm_0 = const()[name = string("transpose_113_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_113 = transpose(perm = transpose_113_perm_0, x = k_33_cast_fp16)[name = string("transpose_254")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_112 = transpose(perm = transpose_112_perm_0, x = var_1573_cast_fp16)[name = string("transpose_255")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_17_cast_fp16 = matmul(transpose_x = matrix_ac_17_transpose_x_0, transpose_y = matrix_ac_17_transpose_y_0, x = transpose_112, y = transpose_113)[name = string("matrix_ac_17_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_113 = transpose(perm = transpose_113_perm_0, x = k_33_cast_fp16)[name = string("transpose_254")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_112 = transpose(perm = transpose_112_perm_0, x = var_1573_cast_fp16)[name = string("transpose_255")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_17_cast_fp16 = matmul(transpose_x = matrix_ac_17_transpose_x_0, transpose_y = matrix_ac_17_transpose_y_0, x = transpose_112, y = transpose_113)[name = string("matrix_ac_17_cast_fp16")];
             tensor<int32, [4]> matrix_bd_35_begin_0 = const()[name = string("matrix_bd_35_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_35_end_0 = const()[name = string("matrix_bd_35_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_35_end_0 = const()[name = string("matrix_bd_35_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
             tensor<bool, [4]> matrix_bd_35_end_mask_0 = const()[name = string("matrix_bd_35_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_35_cast_fp16 = slice_by_index(begin = matrix_bd_35_begin_0, end = matrix_bd_35_end_0, end_mask = matrix_bd_35_end_mask_0, x = matrix_bd_33_cast_fp16)[name = string("matrix_bd_35_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_1599_cast_fp16 = add(x = matrix_ac_17_cast_fp16, y = matrix_bd_35_cast_fp16)[name = string("op_1599_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_35_cast_fp16 = slice_by_index(begin = matrix_bd_35_begin_0, end = matrix_bd_35_end_0, end_mask = matrix_bd_35_end_mask_0, x = matrix_bd_33_cast_fp16)[name = string("matrix_bd_35_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_1599_cast_fp16 = add(x = matrix_ac_17_cast_fp16, y = matrix_bd_35_cast_fp16)[name = string("op_1599_cast_fp16")];
             fp16 _inversed_scores_33_y_0_to_fp16 = const()[name = string("_inversed_scores_33_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_33_cast_fp16 = mul(x = var_1599_cast_fp16, y = _inversed_scores_33_y_0_to_fp16)[name = string("_inversed_scores_33_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_35_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_33_cast_fp16, cond = mask_3)[name = string("scores_35_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_1605_cast_fp16 = softmax(axis = var_30, x = scores_35_cast_fp16)[name = string("op_1605_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_449_cast_fp16 = select(a = var_11_to_fp16, b = var_1605_cast_fp16, cond = mask_3)[name = string("input_449_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_33_cast_fp16 = mul(x = var_1599_cast_fp16, y = _inversed_scores_33_y_0_to_fp16)[name = string("_inversed_scores_33_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_35_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_33_cast_fp16, cond = mask_3)[name = string("scores_35_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_1605_cast_fp16 = softmax(axis = var_30, x = scores_35_cast_fp16)[name = string("op_1605_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_449_cast_fp16 = select(a = var_11_to_fp16, b = var_1605_cast_fp16, cond = mask_3)[name = string("input_449_cast_fp16")];
             bool x_189_transpose_x_0 = const()[name = string("x_189_transpose_x_0"), val = bool(false)];
             bool x_189_transpose_y_0 = const()[name = string("x_189_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_19_cast_fp16 = transpose(perm = value_19_perm_0, x = v_17_cast_fp16)[name = string("transpose_253")];
-            tensor<fp16, [1, 8, 188, 128]> x_189_cast_fp16 = matmul(transpose_x = x_189_transpose_x_0, transpose_y = x_189_transpose_y_0, x = input_449_cast_fp16, y = value_19_cast_fp16)[name = string("x_189_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> value_19_cast_fp16 = transpose(perm = value_19_perm_0, x = v_17_cast_fp16)[name = string("transpose_253")];
+            tensor<fp16, [1, 8, 128, 128]> x_189_cast_fp16 = matmul(transpose_x = x_189_transpose_x_0, transpose_y = x_189_transpose_y_0, x = input_449_cast_fp16, y = value_19_cast_fp16)[name = string("x_189_cast_fp16")];
             tensor<int32, [4]> var_1609_perm_0 = const()[name = string("op_1609_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_1610 = const()[name = string("op_1610"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_1609_cast_fp16 = transpose(perm = var_1609_perm_0, x = x_189_cast_fp16)[name = string("transpose_252")];
-            tensor<fp16, [1, 188, 1024]> input_451_cast_fp16 = reshape(shape = var_1610, x = var_1609_cast_fp16)[name = string("input_451_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522953]> module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118488640))), indices_nonzero_data = tensor<uint6, [522953]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118619776))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119012096))))[name = string("module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> var_1609_cast_fp16 = transpose(perm = var_1609_perm_0, x = x_189_cast_fp16)[name = string("transpose_252")];
+            tensor<fp16, [1, 128, 1024]> input_451_cast_fp16 = reshape(shape = var_1610, x = var_1609_cast_fp16)[name = string("input_451_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522953]> module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113916992))), indices_nonzero_data = tensor<uint6, [522953]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114048128))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114440448))))[name = string("module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_79_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_451_cast_fp16)[name = string("linear_79_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_455_cast_fp16 = add(x = input_447_cast_fp16, y = linear_79_cast_fp16)[name = string("input_455_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_79_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_451_cast_fp16)[name = string("linear_79_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_455_cast_fp16 = add(x = input_447_cast_fp16, y = linear_79_cast_fp16)[name = string("input_455_cast_fp16")];
             tensor<int32, [1]> x_193_axes_0 = const()[name = string("x_193_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_8_norm_conv_weight_to_fp16 = const()[name = string("module_layers_8_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119028544)))];
-            tensor<fp16, [1024]> module_layers_8_norm_conv_bias_to_fp16 = const()[name = string("module_layers_8_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119030656)))];
-            tensor<fp16, [1, 188, 1024]> x_193_cast_fp16 = layer_norm(axes = x_193_axes_0, beta = module_layers_8_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_conv_weight_to_fp16, x = input_455_cast_fp16)[name = string("x_193_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_8_norm_conv_weight_to_fp16 = const()[name = string("module_layers_8_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114456896)))];
+            tensor<fp16, [1024]> module_layers_8_norm_conv_bias_to_fp16 = const()[name = string("module_layers_8_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114459008)))];
+            tensor<fp16, [1, 128, 1024]> x_193_cast_fp16 = layer_norm(axes = x_193_axes_0, beta = module_layers_8_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_conv_weight_to_fp16, x = input_455_cast_fp16)[name = string("x_193_cast_fp16")];
             tensor<int32, [3]> input_457_perm_0 = const()[name = string("input_457_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_459_pad_type_0 = const()[name = string("input_459_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_459_strides_0 = const()[name = string("input_459_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_459_pad_0 = const()[name = string("input_459_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_459_dilations_0 = const()[name = string("input_459_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_459_groups_0 = const()[name = string("input_459_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1050224]> module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119032768))), indices_nonzero_data = tensor<uint6, [1050224]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119294976))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120082752))))[name = string("module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1050224]> module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114461120))), indices_nonzero_data = tensor<uint6, [1050224]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114723328))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115511104))))[name = string("module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_457_cast_fp16 = transpose(perm = input_457_perm_0, x = x_193_cast_fp16)[name = string("transpose_251")];
-            tensor<fp16, [1, 2048, 188]> input_459_cast_fp16 = conv(dilations = input_459_dilations_0, groups = input_459_groups_0, pad = input_459_pad_0, pad_type = input_459_pad_type_0, strides = input_459_strides_0, weight = module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_457_cast_fp16)[name = string("input_459_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_457_cast_fp16 = transpose(perm = input_457_perm_0, x = x_193_cast_fp16)[name = string("transpose_251")];
+            tensor<fp16, [1, 2048, 128]> input_459_cast_fp16 = conv(dilations = input_459_dilations_0, groups = input_459_groups_0, pad = input_459_pad_0, pad_type = input_459_pad_type_0, strides = input_459_strides_0, weight = module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_457_cast_fp16)[name = string("input_459_cast_fp16")];
             int32 x_195_split_num_splits_0 = const()[name = string("x_195_split_num_splits_0"), val = int32(2)];
             int32 x_195_split_axis_0 = const()[name = string("x_195_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_195_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_195_split_cast_fp16_1 = split(axis = x_195_split_axis_0, num_splits = x_195_split_num_splits_0, x = input_459_cast_fp16)[name = string("x_195_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_195_split_1_sigmoid_cast_fp16 = sigmoid(x = x_195_split_cast_fp16_1)[name = string("x_195_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_195_cast_fp16 = mul(x = x_195_split_cast_fp16_0, y = x_195_split_1_sigmoid_cast_fp16)[name = string("x_195_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_461_cast_fp16 = select(a = var_11_to_fp16, b = x_195_cast_fp16, cond = var_328)[name = string("input_461_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_195_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_195_split_cast_fp16_1 = split(axis = x_195_split_axis_0, num_splits = x_195_split_num_splits_0, x = input_459_cast_fp16)[name = string("x_195_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_195_split_1_sigmoid_cast_fp16 = sigmoid(x = x_195_split_cast_fp16_1)[name = string("x_195_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_195_cast_fp16 = mul(x = x_195_split_cast_fp16_0, y = x_195_split_1_sigmoid_cast_fp16)[name = string("x_195_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_461_cast_fp16 = select(a = var_11_to_fp16, b = x_195_cast_fp16, cond = var_328)[name = string("input_461_cast_fp16")];
             tensor<int32, [6]> input_463_pad_0 = const()[name = string("input_463_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_463_mode_0 = const()[name = string("input_463_mode_0"), val = string("constant")];
             fp16 const_97_to_fp16 = const()[name = string("const_97_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_463_cast_fp16 = pad(constant_val = const_97_to_fp16, mode = input_463_mode_0, pad = input_463_pad_0, x = input_461_cast_fp16)[name = string("input_463_cast_fp16")];
+            tensor<fp16, [1, 1024, 136]> input_463_cast_fp16 = pad(constant_val = const_97_to_fp16, mode = input_463_mode_0, pad = input_463_pad_0, x = input_461_cast_fp16)[name = string("input_463_cast_fp16")];
             string input_465_pad_type_0 = const()[name = string("input_465_pad_type_0"), val = string("valid")];
             int32 input_465_groups_0 = const()[name = string("input_465_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_465_strides_0 = const()[name = string("input_465_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_465_pad_0 = const()[name = string("input_465_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_465_dilations_0 = const()[name = string("input_465_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_264_to_fp16 = const()[name = string("const_264_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120115584)))];
-            tensor<fp16, [1024]> const_265_to_fp16 = const()[name = string("const_265_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120134080)))];
-            tensor<fp16, [1, 1024, 188]> input_467_cast_fp16 = conv(bias = const_265_to_fp16, dilations = input_465_dilations_0, groups = input_465_groups_0, pad = input_465_pad_0, pad_type = input_465_pad_type_0, strides = input_465_strides_0, weight = const_264_to_fp16, x = input_463_cast_fp16)[name = string("input_467_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_469_cast_fp16 = silu(x = input_467_cast_fp16)[name = string("input_469_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_264_to_fp16 = const()[name = string("const_264_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115543936)))];
+            tensor<fp16, [1024]> const_265_to_fp16 = const()[name = string("const_265_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115562432)))];
+            tensor<fp16, [1, 1024, 128]> input_467_cast_fp16 = conv(bias = const_265_to_fp16, dilations = input_465_dilations_0, groups = input_465_groups_0, pad = input_465_pad_0, pad_type = input_465_pad_type_0, strides = input_465_strides_0, weight = const_264_to_fp16, x = input_463_cast_fp16)[name = string("input_467_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_469_cast_fp16 = silu(x = input_467_cast_fp16)[name = string("input_469_cast_fp16")];
             string x_197_pad_type_0 = const()[name = string("x_197_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_197_strides_0 = const()[name = string("x_197_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_197_pad_0 = const()[name = string("x_197_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_197_dilations_0 = const()[name = string("x_197_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_197_groups_0 = const()[name = string("x_197_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521521]> module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120136192))), indices_nonzero_data = tensor<uint6, [521521]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120267328))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120658560))))[name = string("module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521521]> module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115564544))), indices_nonzero_data = tensor<uint6, [521521]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115695680))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116086912))))[name = string("module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_197_cast_fp16 = conv(dilations = x_197_dilations_0, groups = x_197_groups_0, pad = x_197_pad_0, pad_type = x_197_pad_type_0, strides = x_197_strides_0, weight = module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_469_cast_fp16)[name = string("x_197_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_197_cast_fp16 = conv(dilations = x_197_dilations_0, groups = x_197_groups_0, pad = x_197_pad_0, pad_type = x_197_pad_type_0, strides = x_197_strides_0, weight = module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_469_cast_fp16)[name = string("x_197_cast_fp16")];
             tensor<int32, [3]> input_471_perm_0 = const()[name = string("input_471_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_471_cast_fp16 = transpose(perm = input_471_perm_0, x = x_197_cast_fp16)[name = string("transpose_250")];
-            tensor<fp16, [1, 188, 1024]> input_473_cast_fp16 = add(x = input_455_cast_fp16, y = input_471_cast_fp16)[name = string("input_473_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_471_cast_fp16 = transpose(perm = input_471_perm_0, x = x_197_cast_fp16)[name = string("transpose_250")];
+            tensor<fp16, [1, 128, 1024]> input_473_cast_fp16 = add(x = input_455_cast_fp16, y = input_471_cast_fp16)[name = string("input_473_cast_fp16")];
             tensor<int32, [1]> input_475_axes_0 = const()[name = string("input_475_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_8_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_8_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120675008)))];
-            tensor<fp16, [1024]> module_layers_8_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_8_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120677120)))];
-            tensor<fp16, [1, 188, 1024]> input_475_cast_fp16 = layer_norm(axes = input_475_axes_0, beta = module_layers_8_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_feed_forward2_weight_to_fp16, x = input_473_cast_fp16)[name = string("input_475_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2065024]> module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120679232))), indices_nonzero_data = tensor<uint6, [2065024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121203584))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122752448))))[name = string("module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_8_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_8_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116103360)))];
+            tensor<fp16, [1024]> module_layers_8_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_8_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116105472)))];
+            tensor<fp16, [1, 128, 1024]> input_475_cast_fp16 = layer_norm(axes = input_475_axes_0, beta = module_layers_8_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_feed_forward2_weight_to_fp16, x = input_473_cast_fp16)[name = string("input_475_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2065024]> module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116107584))), indices_nonzero_data = tensor<uint6, [2065024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116631936))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118180800))))[name = string("module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_80_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_475_cast_fp16)[name = string("linear_80_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_479_cast_fp16 = silu(x = linear_80_cast_fp16)[name = string("input_479_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2041094]> module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122818048))), indices_nonzero_data = tensor<uint6, [2041094]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123342400))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124873344))))[name = string("module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_80_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_475_cast_fp16)[name = string("linear_80_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_479_cast_fp16 = silu(x = linear_80_cast_fp16)[name = string("input_479_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2041094]> module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118246400))), indices_nonzero_data = tensor<uint6, [2041094]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118770752))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120301696))))[name = string("module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_81_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_479_cast_fp16)[name = string("linear_81_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_81_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_479_cast_fp16)[name = string("linear_81_cast_fp16")];
             fp16 var_1670_to_fp16 = const()[name = string("op_1670_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_1671_cast_fp16 = mul(x = linear_81_cast_fp16, y = var_1670_to_fp16)[name = string("op_1671_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_485_cast_fp16 = add(x = input_473_cast_fp16, y = var_1671_cast_fp16)[name = string("input_485_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_1671_cast_fp16 = mul(x = linear_81_cast_fp16, y = var_1670_to_fp16)[name = string("op_1671_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_485_cast_fp16 = add(x = input_473_cast_fp16, y = var_1671_cast_fp16)[name = string("input_485_cast_fp16")];
             tensor<int32, [1]> input_487_axes_0 = const()[name = string("input_487_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_8_norm_out_weight_to_fp16 = const()[name = string("module_layers_8_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124889792)))];
-            tensor<fp16, [1024]> module_layers_8_norm_out_bias_to_fp16 = const()[name = string("module_layers_8_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124891904)))];
-            tensor<fp16, [1, 188, 1024]> input_487_cast_fp16 = layer_norm(axes = input_487_axes_0, beta = module_layers_8_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_out_weight_to_fp16, x = input_485_cast_fp16)[name = string("input_487_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_8_norm_out_weight_to_fp16 = const()[name = string("module_layers_8_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120318144)))];
+            tensor<fp16, [1024]> module_layers_8_norm_out_bias_to_fp16 = const()[name = string("module_layers_8_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120320256)))];
+            tensor<fp16, [1, 128, 1024]> input_487_cast_fp16 = layer_norm(axes = input_487_axes_0, beta = module_layers_8_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_out_weight_to_fp16, x = input_485_cast_fp16)[name = string("input_487_cast_fp16")];
             tensor<int32, [1]> input_489_axes_0 = const()[name = string("input_489_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_9_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_9_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124894016)))];
-            tensor<fp16, [1024]> module_layers_9_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_9_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124896128)))];
-            tensor<fp16, [1, 188, 1024]> input_489_cast_fp16 = layer_norm(axes = input_489_axes_0, beta = module_layers_9_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_feed_forward1_weight_to_fp16, x = input_487_cast_fp16)[name = string("input_489_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2056248]> module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124898240))), indices_nonzero_data = tensor<uint6, [2056248]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125422592))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126964864))))[name = string("module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_9_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_9_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120322368)))];
+            tensor<fp16, [1024]> module_layers_9_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_9_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120324480)))];
+            tensor<fp16, [1, 128, 1024]> input_489_cast_fp16 = layer_norm(axes = input_489_axes_0, beta = module_layers_9_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_feed_forward1_weight_to_fp16, x = input_487_cast_fp16)[name = string("input_489_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2056248]> module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120326592))), indices_nonzero_data = tensor<uint6, [2056248]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120850944))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122393216))))[name = string("module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_82_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_489_cast_fp16)[name = string("linear_82_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_493_cast_fp16 = silu(x = linear_82_cast_fp16)[name = string("input_493_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1988988]> module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127030464))), indices_nonzero_data = tensor<uint6, [1988988]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127554816))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129046656))))[name = string("module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_82_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_489_cast_fp16)[name = string("linear_82_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_493_cast_fp16 = silu(x = linear_82_cast_fp16)[name = string("input_493_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1988988]> module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122458816))), indices_nonzero_data = tensor<uint6, [1988988]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122983168))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124475008))))[name = string("module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_83_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_493_cast_fp16)[name = string("linear_83_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_83_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_493_cast_fp16)[name = string("linear_83_cast_fp16")];
             fp16 var_1699_to_fp16 = const()[name = string("op_1699_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_1700_cast_fp16 = mul(x = linear_83_cast_fp16, y = var_1699_to_fp16)[name = string("op_1700_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_499_cast_fp16 = add(x = input_487_cast_fp16, y = var_1700_cast_fp16)[name = string("input_499_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_1700_cast_fp16 = mul(x = linear_83_cast_fp16, y = var_1699_to_fp16)[name = string("op_1700_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_499_cast_fp16 = add(x = input_487_cast_fp16, y = var_1700_cast_fp16)[name = string("input_499_cast_fp16")];
             tensor<int32, [1]> query_19_axes_0 = const()[name = string("query_19_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_9_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_9_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129063104)))];
-            tensor<fp16, [1024]> module_layers_9_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_9_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129065216)))];
-            tensor<fp16, [1, 188, 1024]> query_19_cast_fp16 = layer_norm(axes = query_19_axes_0, beta = module_layers_9_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_self_att_weight_to_fp16, x = input_499_cast_fp16)[name = string("query_19_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524317]> module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129067328))), indices_nonzero_data = tensor<uint6, [524317]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129198464))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129591808))))[name = string("module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_9_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_9_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124491456)))];
+            tensor<fp16, [1024]> module_layers_9_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_9_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124493568)))];
+            tensor<fp16, [1, 128, 1024]> query_19_cast_fp16 = layer_norm(axes = query_19_axes_0, beta = module_layers_9_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_self_att_weight_to_fp16, x = input_499_cast_fp16)[name = string("query_19_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524317]> module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124495680))), indices_nonzero_data = tensor<uint6, [524317]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124626816))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125020160))))[name = string("module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_84_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_19_cast_fp16)[name = string("linear_84_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_84_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_19_cast_fp16)[name = string("linear_84_cast_fp16")];
             tensor<int32, [4]> var_1716 = const()[name = string("op_1716"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_55_cast_fp16 = reshape(shape = var_1716, x = linear_84_cast_fp16)[name = string("q_55_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524008]> module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129608256))), indices_nonzero_data = tensor<uint6, [524008]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129739392))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130132480))))[name = string("module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> q_55_cast_fp16 = reshape(shape = var_1716, x = linear_84_cast_fp16)[name = string("q_55_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524008]> module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125036608))), indices_nonzero_data = tensor<uint6, [524008]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125167744))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125560832))))[name = string("module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_85_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_19_cast_fp16)[name = string("linear_85_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_85_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_19_cast_fp16)[name = string("linear_85_cast_fp16")];
             tensor<int32, [4]> var_1720 = const()[name = string("op_1720"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_37_cast_fp16 = reshape(shape = var_1720, x = linear_85_cast_fp16)[name = string("k_37_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524728]> module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130148928))), indices_nonzero_data = tensor<uint6, [524728]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130280064))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130673728))))[name = string("module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> k_37_cast_fp16 = reshape(shape = var_1720, x = linear_85_cast_fp16)[name = string("k_37_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524728]> module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125577280))), indices_nonzero_data = tensor<uint6, [524728]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125708416))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126102080))))[name = string("module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_86_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_19_cast_fp16)[name = string("linear_86_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_86_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_19_cast_fp16)[name = string("linear_86_cast_fp16")];
             tensor<int32, [4]> var_1724 = const()[name = string("op_1724"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_19_cast_fp16 = reshape(shape = var_1724, x = linear_86_cast_fp16)[name = string("v_19_cast_fp16")];
+            tensor<fp16, [1, 128, 8, 128]> v_19_cast_fp16 = reshape(shape = var_1724, x = linear_86_cast_fp16)[name = string("v_19_cast_fp16")];
             tensor<int32, [4]> value_21_perm_0 = const()[name = string("value_21_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_9_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_9_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130690176)))];
-            tensor<fp16, [1, 188, 8, 128]> var_1736_cast_fp16 = add(x = q_55_cast_fp16, y = module_layers_9_self_attn_pos_bias_u_to_fp16)[name = string("op_1736_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_9_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_9_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130692288)))];
-            tensor<fp16, [1, 188, 8, 128]> var_1738_cast_fp16 = add(x = q_55_cast_fp16, y = module_layers_9_self_attn_pos_bias_v_to_fp16)[name = string("op_1738_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_9_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_9_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126118528)))];
+            tensor<fp16, [1, 128, 8, 128]> var_1736_cast_fp16 = add(x = q_55_cast_fp16, y = module_layers_9_self_attn_pos_bias_u_to_fp16)[name = string("op_1736_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_9_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_9_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126120640)))];
+            tensor<fp16, [1, 128, 8, 128]> var_1738_cast_fp16 = add(x = q_55_cast_fp16, y = module_layers_9_self_attn_pos_bias_v_to_fp16)[name = string("op_1738_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_19_perm_0 = const()[name = string("q_with_bias_v_19_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_205_transpose_x_0 = const()[name = string("x_205_transpose_x_0"), val = bool(false)];
             bool x_205_transpose_y_0 = const()[name = string("x_205_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_1740_to_fp16 = const()[name = string("op_1740_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130694400)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_19_cast_fp16 = transpose(perm = q_with_bias_v_19_perm_0, x = var_1738_cast_fp16)[name = string("transpose_249")];
-            tensor<fp16, [1, 8, 188, 375]> x_205_cast_fp16 = matmul(transpose_x = x_205_transpose_x_0, transpose_y = x_205_transpose_y_0, x = q_with_bias_v_19_cast_fp16, y = var_1740_to_fp16)[name = string("x_205_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 255]> var_1740_to_fp16 = const()[name = string("op_1740_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312122560)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_19_cast_fp16 = transpose(perm = q_with_bias_v_19_perm_0, x = var_1738_cast_fp16)[name = string("transpose_249")];
+            tensor<fp16, [1, 8, 128, 255]> x_205_cast_fp16 = matmul(transpose_x = x_205_transpose_x_0, transpose_y = x_205_transpose_y_0, x = q_with_bias_v_19_cast_fp16, y = var_1740_to_fp16)[name = string("x_205_cast_fp16")];
             tensor<int32, [8]> x_207_pad_0 = const()[name = string("x_207_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_207_mode_0 = const()[name = string("x_207_mode_0"), val = string("constant")];
             fp16 const_104_to_fp16 = const()[name = string("const_104_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_207_cast_fp16 = pad(constant_val = const_104_to_fp16, mode = x_207_mode_0, pad = x_207_pad_0, x = x_205_cast_fp16)[name = string("x_207_cast_fp16")];
-            tensor<int32, [4]> var_1748 = const()[name = string("op_1748"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_209_cast_fp16 = reshape(shape = var_1748, x = x_207_cast_fp16)[name = string("x_209_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 256]> x_207_cast_fp16 = pad(constant_val = const_104_to_fp16, mode = x_207_mode_0, pad = x_207_pad_0, x = x_205_cast_fp16)[name = string("x_207_cast_fp16")];
+            tensor<int32, [4]> var_1748 = const()[name = string("op_1748"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_209_cast_fp16 = reshape(shape = var_1748, x = x_207_cast_fp16)[name = string("x_209_cast_fp16")];
             tensor<int32, [4]> var_1752_begin_0 = const()[name = string("op_1752_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_1752_end_0 = const()[name = string("op_1752_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_1752_end_0 = const()[name = string("op_1752_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
             tensor<bool, [4]> var_1752_end_mask_0 = const()[name = string("op_1752_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_1752_cast_fp16 = slice_by_index(begin = var_1752_begin_0, end = var_1752_end_0, end_mask = var_1752_end_mask_0, x = x_209_cast_fp16)[name = string("op_1752_cast_fp16")];
-            tensor<int32, [4]> var_1753 = const()[name = string("op_1753"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_37_cast_fp16 = reshape(shape = var_1753, x = var_1752_cast_fp16)[name = string("matrix_bd_37_cast_fp16")];
+            tensor<fp16, [1, 8, 255, 128]> var_1752_cast_fp16 = slice_by_index(begin = var_1752_begin_0, end = var_1752_end_0, end_mask = var_1752_end_mask_0, x = x_209_cast_fp16)[name = string("op_1752_cast_fp16")];
+            tensor<int32, [4]> var_1753 = const()[name = string("op_1753"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_37_cast_fp16 = reshape(shape = var_1753, x = var_1752_cast_fp16)[name = string("matrix_bd_37_cast_fp16")];
             bool matrix_ac_19_transpose_x_0 = const()[name = string("matrix_ac_19_transpose_x_0"), val = bool(false)];
             bool matrix_ac_19_transpose_y_0 = const()[name = string("matrix_ac_19_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_114_perm_0 = const()[name = string("transpose_114_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_115_perm_0 = const()[name = string("transpose_115_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_115 = transpose(perm = transpose_115_perm_0, x = k_37_cast_fp16)[name = string("transpose_247")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_114 = transpose(perm = transpose_114_perm_0, x = var_1736_cast_fp16)[name = string("transpose_248")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_19_cast_fp16 = matmul(transpose_x = matrix_ac_19_transpose_x_0, transpose_y = matrix_ac_19_transpose_y_0, x = transpose_114, y = transpose_115)[name = string("matrix_ac_19_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_115 = transpose(perm = transpose_115_perm_0, x = k_37_cast_fp16)[name = string("transpose_247")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_114 = transpose(perm = transpose_114_perm_0, x = var_1736_cast_fp16)[name = string("transpose_248")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_19_cast_fp16 = matmul(transpose_x = matrix_ac_19_transpose_x_0, transpose_y = matrix_ac_19_transpose_y_0, x = transpose_114, y = transpose_115)[name = string("matrix_ac_19_cast_fp16")];
             tensor<int32, [4]> matrix_bd_39_begin_0 = const()[name = string("matrix_bd_39_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_39_end_0 = const()[name = string("matrix_bd_39_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_39_end_0 = const()[name = string("matrix_bd_39_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
             tensor<bool, [4]> matrix_bd_39_end_mask_0 = const()[name = string("matrix_bd_39_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_39_cast_fp16 = slice_by_index(begin = matrix_bd_39_begin_0, end = matrix_bd_39_end_0, end_mask = matrix_bd_39_end_mask_0, x = matrix_bd_37_cast_fp16)[name = string("matrix_bd_39_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_1762_cast_fp16 = add(x = matrix_ac_19_cast_fp16, y = matrix_bd_39_cast_fp16)[name = string("op_1762_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_39_cast_fp16 = slice_by_index(begin = matrix_bd_39_begin_0, end = matrix_bd_39_end_0, end_mask = matrix_bd_39_end_mask_0, x = matrix_bd_37_cast_fp16)[name = string("matrix_bd_39_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_1762_cast_fp16 = add(x = matrix_ac_19_cast_fp16, y = matrix_bd_39_cast_fp16)[name = string("op_1762_cast_fp16")];
             fp16 _inversed_scores_37_y_0_to_fp16 = const()[name = string("_inversed_scores_37_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_37_cast_fp16 = mul(x = var_1762_cast_fp16, y = _inversed_scores_37_y_0_to_fp16)[name = string("_inversed_scores_37_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_39_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_37_cast_fp16, cond = mask_3)[name = string("scores_39_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_1768_cast_fp16 = softmax(axis = var_30, x = scores_39_cast_fp16)[name = string("op_1768_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_501_cast_fp16 = select(a = var_11_to_fp16, b = var_1768_cast_fp16, cond = mask_3)[name = string("input_501_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_37_cast_fp16 = mul(x = var_1762_cast_fp16, y = _inversed_scores_37_y_0_to_fp16)[name = string("_inversed_scores_37_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_39_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_37_cast_fp16, cond = mask_3)[name = string("scores_39_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_1768_cast_fp16 = softmax(axis = var_30, x = scores_39_cast_fp16)[name = string("op_1768_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_501_cast_fp16 = select(a = var_11_to_fp16, b = var_1768_cast_fp16, cond = mask_3)[name = string("input_501_cast_fp16")];
             bool x_211_transpose_x_0 = const()[name = string("x_211_transpose_x_0"), val = bool(false)];
             bool x_211_transpose_y_0 = const()[name = string("x_211_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_21_cast_fp16 = transpose(perm = value_21_perm_0, x = v_19_cast_fp16)[name = string("transpose_246")];
-            tensor<fp16, [1, 8, 188, 128]> x_211_cast_fp16 = matmul(transpose_x = x_211_transpose_x_0, transpose_y = x_211_transpose_y_0, x = input_501_cast_fp16, y = value_21_cast_fp16)[name = string("x_211_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> value_21_cast_fp16 = transpose(perm = value_21_perm_0, x = v_19_cast_fp16)[name = string("transpose_246")];
+            tensor<fp16, [1, 8, 128, 128]> x_211_cast_fp16 = matmul(transpose_x = x_211_transpose_x_0, transpose_y = x_211_transpose_y_0, x = input_501_cast_fp16, y = value_21_cast_fp16)[name = string("x_211_cast_fp16")];
             tensor<int32, [4]> var_1772_perm_0 = const()[name = string("op_1772_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_1773 = const()[name = string("op_1773"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_1772_cast_fp16 = transpose(perm = var_1772_perm_0, x = x_211_cast_fp16)[name = string("transpose_245")];
-            tensor<fp16, [1, 188, 1024]> input_503_cast_fp16 = reshape(shape = var_1773, x = var_1772_cast_fp16)[name = string("input_503_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524470]> module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131462464))), indices_nonzero_data = tensor<uint6, [524470]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131593600))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131987072))))[name = string("module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> var_1772_cast_fp16 = transpose(perm = var_1772_perm_0, x = x_211_cast_fp16)[name = string("transpose_245")];
+            tensor<fp16, [1, 128, 1024]> input_503_cast_fp16 = reshape(shape = var_1773, x = var_1772_cast_fp16)[name = string("input_503_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524470]> module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126382912))), indices_nonzero_data = tensor<uint6, [524470]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126514048))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126907520))))[name = string("module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_88_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_503_cast_fp16)[name = string("linear_88_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_507_cast_fp16 = add(x = input_499_cast_fp16, y = linear_88_cast_fp16)[name = string("input_507_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_88_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_503_cast_fp16)[name = string("linear_88_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_507_cast_fp16 = add(x = input_499_cast_fp16, y = linear_88_cast_fp16)[name = string("input_507_cast_fp16")];
             tensor<int32, [1]> x_215_axes_0 = const()[name = string("x_215_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_9_norm_conv_weight_to_fp16 = const()[name = string("module_layers_9_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132003520)))];
-            tensor<fp16, [1024]> module_layers_9_norm_conv_bias_to_fp16 = const()[name = string("module_layers_9_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132005632)))];
-            tensor<fp16, [1, 188, 1024]> x_215_cast_fp16 = layer_norm(axes = x_215_axes_0, beta = module_layers_9_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_conv_weight_to_fp16, x = input_507_cast_fp16)[name = string("x_215_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_9_norm_conv_weight_to_fp16 = const()[name = string("module_layers_9_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126923968)))];
+            tensor<fp16, [1024]> module_layers_9_norm_conv_bias_to_fp16 = const()[name = string("module_layers_9_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126926080)))];
+            tensor<fp16, [1, 128, 1024]> x_215_cast_fp16 = layer_norm(axes = x_215_axes_0, beta = module_layers_9_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_conv_weight_to_fp16, x = input_507_cast_fp16)[name = string("x_215_cast_fp16")];
             tensor<int32, [3]> input_509_perm_0 = const()[name = string("input_509_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_511_pad_type_0 = const()[name = string("input_511_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_511_strides_0 = const()[name = string("input_511_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_511_pad_0 = const()[name = string("input_511_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_511_dilations_0 = const()[name = string("input_511_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_511_groups_0 = const()[name = string("input_511_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1050959]> module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132007744))), indices_nonzero_data = tensor<uint6, [1050959]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132269952))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133058240))))[name = string("module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1050959]> module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126928192))), indices_nonzero_data = tensor<uint6, [1050959]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127190400))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127978688))))[name = string("module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_509_cast_fp16 = transpose(perm = input_509_perm_0, x = x_215_cast_fp16)[name = string("transpose_244")];
-            tensor<fp16, [1, 2048, 188]> input_511_cast_fp16 = conv(dilations = input_511_dilations_0, groups = input_511_groups_0, pad = input_511_pad_0, pad_type = input_511_pad_type_0, strides = input_511_strides_0, weight = module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_509_cast_fp16)[name = string("input_511_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_509_cast_fp16 = transpose(perm = input_509_perm_0, x = x_215_cast_fp16)[name = string("transpose_244")];
+            tensor<fp16, [1, 2048, 128]> input_511_cast_fp16 = conv(dilations = input_511_dilations_0, groups = input_511_groups_0, pad = input_511_pad_0, pad_type = input_511_pad_type_0, strides = input_511_strides_0, weight = module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_509_cast_fp16)[name = string("input_511_cast_fp16")];
             int32 x_217_split_num_splits_0 = const()[name = string("x_217_split_num_splits_0"), val = int32(2)];
             int32 x_217_split_axis_0 = const()[name = string("x_217_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_217_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_217_split_cast_fp16_1 = split(axis = x_217_split_axis_0, num_splits = x_217_split_num_splits_0, x = input_511_cast_fp16)[name = string("x_217_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_217_split_1_sigmoid_cast_fp16 = sigmoid(x = x_217_split_cast_fp16_1)[name = string("x_217_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_217_cast_fp16 = mul(x = x_217_split_cast_fp16_0, y = x_217_split_1_sigmoid_cast_fp16)[name = string("x_217_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_513_cast_fp16 = select(a = var_11_to_fp16, b = x_217_cast_fp16, cond = var_328)[name = string("input_513_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_217_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_217_split_cast_fp16_1 = split(axis = x_217_split_axis_0, num_splits = x_217_split_num_splits_0, x = input_511_cast_fp16)[name = string("x_217_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_217_split_1_sigmoid_cast_fp16 = sigmoid(x = x_217_split_cast_fp16_1)[name = string("x_217_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_217_cast_fp16 = mul(x = x_217_split_cast_fp16_0, y = x_217_split_1_sigmoid_cast_fp16)[name = string("x_217_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_513_cast_fp16 = select(a = var_11_to_fp16, b = x_217_cast_fp16, cond = var_328)[name = string("input_513_cast_fp16")];
             tensor<int32, [6]> input_515_pad_0 = const()[name = string("input_515_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_515_mode_0 = const()[name = string("input_515_mode_0"), val = string("constant")];
             fp16 const_107_to_fp16 = const()[name = string("const_107_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_515_cast_fp16 = pad(constant_val = const_107_to_fp16, mode = input_515_mode_0, pad = input_515_pad_0, x = input_513_cast_fp16)[name = string("input_515_cast_fp16")];
+            tensor<fp16, [1, 1024, 136]> input_515_cast_fp16 = pad(constant_val = const_107_to_fp16, mode = input_515_mode_0, pad = input_515_pad_0, x = input_513_cast_fp16)[name = string("input_515_cast_fp16")];
             string input_517_pad_type_0 = const()[name = string("input_517_pad_type_0"), val = string("valid")];
             int32 input_517_groups_0 = const()[name = string("input_517_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_517_strides_0 = const()[name = string("input_517_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_517_pad_0 = const()[name = string("input_517_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_517_dilations_0 = const()[name = string("input_517_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_266_to_fp16 = const()[name = string("const_266_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133091072)))];
-            tensor<fp16, [1024]> const_267_to_fp16 = const()[name = string("const_267_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133109568)))];
-            tensor<fp16, [1, 1024, 188]> input_519_cast_fp16 = conv(bias = const_267_to_fp16, dilations = input_517_dilations_0, groups = input_517_groups_0, pad = input_517_pad_0, pad_type = input_517_pad_type_0, strides = input_517_strides_0, weight = const_266_to_fp16, x = input_515_cast_fp16)[name = string("input_519_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_521_cast_fp16 = silu(x = input_519_cast_fp16)[name = string("input_521_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_266_to_fp16 = const()[name = string("const_266_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128011520)))];
+            tensor<fp16, [1024]> const_267_to_fp16 = const()[name = string("const_267_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128030016)))];
+            tensor<fp16, [1, 1024, 128]> input_519_cast_fp16 = conv(bias = const_267_to_fp16, dilations = input_517_dilations_0, groups = input_517_groups_0, pad = input_517_pad_0, pad_type = input_517_pad_type_0, strides = input_517_strides_0, weight = const_266_to_fp16, x = input_515_cast_fp16)[name = string("input_519_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_521_cast_fp16 = silu(x = input_519_cast_fp16)[name = string("input_521_cast_fp16")];
             string x_219_pad_type_0 = const()[name = string("x_219_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_219_strides_0 = const()[name = string("x_219_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_219_pad_0 = const()[name = string("x_219_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_219_dilations_0 = const()[name = string("x_219_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_219_groups_0 = const()[name = string("x_219_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519248]> module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133111680))), indices_nonzero_data = tensor<uint6, [519248]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133242816))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133632320))))[name = string("module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519248]> module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128032128))), indices_nonzero_data = tensor<uint6, [519248]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128163264))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128552768))))[name = string("module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_219_cast_fp16 = conv(dilations = x_219_dilations_0, groups = x_219_groups_0, pad = x_219_pad_0, pad_type = x_219_pad_type_0, strides = x_219_strides_0, weight = module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_521_cast_fp16)[name = string("x_219_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_219_cast_fp16 = conv(dilations = x_219_dilations_0, groups = x_219_groups_0, pad = x_219_pad_0, pad_type = x_219_pad_type_0, strides = x_219_strides_0, weight = module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_521_cast_fp16)[name = string("x_219_cast_fp16")];
             tensor<int32, [3]> input_523_perm_0 = const()[name = string("input_523_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_523_cast_fp16 = transpose(perm = input_523_perm_0, x = x_219_cast_fp16)[name = string("transpose_243")];
-            tensor<fp16, [1, 188, 1024]> input_525_cast_fp16 = add(x = input_507_cast_fp16, y = input_523_cast_fp16)[name = string("input_525_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_523_cast_fp16 = transpose(perm = input_523_perm_0, x = x_219_cast_fp16)[name = string("transpose_243")];
+            tensor<fp16, [1, 128, 1024]> input_525_cast_fp16 = add(x = input_507_cast_fp16, y = input_523_cast_fp16)[name = string("input_525_cast_fp16")];
             tensor<int32, [1]> input_527_axes_0 = const()[name = string("input_527_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_9_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_9_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133648768)))];
-            tensor<fp16, [1024]> module_layers_9_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_9_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133650880)))];
-            tensor<fp16, [1, 188, 1024]> input_527_cast_fp16 = layer_norm(axes = input_527_axes_0, beta = module_layers_9_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_feed_forward2_weight_to_fp16, x = input_525_cast_fp16)[name = string("input_527_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2059311]> module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133652992))), indices_nonzero_data = tensor<uint6, [2059311]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134177344))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135721920))))[name = string("module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_9_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_9_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128569216)))];
+            tensor<fp16, [1024]> module_layers_9_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_9_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128571328)))];
+            tensor<fp16, [1, 128, 1024]> input_527_cast_fp16 = layer_norm(axes = input_527_axes_0, beta = module_layers_9_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_feed_forward2_weight_to_fp16, x = input_525_cast_fp16)[name = string("input_527_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2059311]> module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128573440))), indices_nonzero_data = tensor<uint6, [2059311]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129097792))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130642368))))[name = string("module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_89_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_527_cast_fp16)[name = string("linear_89_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_531_cast_fp16 = silu(x = linear_89_cast_fp16)[name = string("input_531_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2055301]> module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135787520))), indices_nonzero_data = tensor<uint6, [2055301]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136311872))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137853440))))[name = string("module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_89_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_527_cast_fp16)[name = string("linear_89_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_531_cast_fp16 = silu(x = linear_89_cast_fp16)[name = string("input_531_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2055301]> module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130707968))), indices_nonzero_data = tensor<uint6, [2055301]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131232320))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132773888))))[name = string("module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_90_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_531_cast_fp16)[name = string("linear_90_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_90_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_531_cast_fp16)[name = string("linear_90_cast_fp16")];
             fp16 var_1833_to_fp16 = const()[name = string("op_1833_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_1834_cast_fp16 = mul(x = linear_90_cast_fp16, y = var_1833_to_fp16)[name = string("op_1834_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_537_cast_fp16 = add(x = input_525_cast_fp16, y = var_1834_cast_fp16)[name = string("input_537_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_1834_cast_fp16 = mul(x = linear_90_cast_fp16, y = var_1833_to_fp16)[name = string("op_1834_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_537_cast_fp16 = add(x = input_525_cast_fp16, y = var_1834_cast_fp16)[name = string("input_537_cast_fp16")];
             tensor<int32, [1]> input_539_axes_0 = const()[name = string("input_539_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_9_norm_out_weight_to_fp16 = const()[name = string("module_layers_9_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137869888)))];
-            tensor<fp16, [1024]> module_layers_9_norm_out_bias_to_fp16 = const()[name = string("module_layers_9_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137872000)))];
-            tensor<fp16, [1, 188, 1024]> input_539_cast_fp16 = layer_norm(axes = input_539_axes_0, beta = module_layers_9_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_out_weight_to_fp16, x = input_537_cast_fp16)[name = string("input_539_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_9_norm_out_weight_to_fp16 = const()[name = string("module_layers_9_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132790336)))];
+            tensor<fp16, [1024]> module_layers_9_norm_out_bias_to_fp16 = const()[name = string("module_layers_9_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132792448)))];
+            tensor<fp16, [1, 128, 1024]> input_539_cast_fp16 = layer_norm(axes = input_539_axes_0, beta = module_layers_9_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_out_weight_to_fp16, x = input_537_cast_fp16)[name = string("input_539_cast_fp16")];
             tensor<int32, [1]> input_541_axes_0 = const()[name = string("input_541_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_10_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_10_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137874112)))];
-            tensor<fp16, [1024]> module_layers_10_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_10_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137876224)))];
-            tensor<fp16, [1, 188, 1024]> input_541_cast_fp16 = layer_norm(axes = input_541_axes_0, beta = module_layers_10_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_feed_forward1_weight_to_fp16, x = input_539_cast_fp16)[name = string("input_541_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2042962]> module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137878336))), indices_nonzero_data = tensor<uint6, [2042962]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138402688))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139934976))))[name = string("module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_10_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_10_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132794560)))];
+            tensor<fp16, [1024]> module_layers_10_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_10_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132796672)))];
+            tensor<fp16, [1, 128, 1024]> input_541_cast_fp16 = layer_norm(axes = input_541_axes_0, beta = module_layers_10_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_feed_forward1_weight_to_fp16, x = input_539_cast_fp16)[name = string("input_541_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2042962]> module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132798784))), indices_nonzero_data = tensor<uint6, [2042962]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133323136))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134855424))))[name = string("module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_91_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_541_cast_fp16)[name = string("linear_91_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_545_cast_fp16 = silu(x = linear_91_cast_fp16)[name = string("input_545_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1990388]> module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140000576))), indices_nonzero_data = tensor<uint6, [1990388]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140524928))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142017792))))[name = string("module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_91_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_541_cast_fp16)[name = string("linear_91_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_545_cast_fp16 = silu(x = linear_91_cast_fp16)[name = string("input_545_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1990388]> module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134921024))), indices_nonzero_data = tensor<uint6, [1990388]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135445376))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136938240))))[name = string("module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_92_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_545_cast_fp16)[name = string("linear_92_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_92_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_545_cast_fp16)[name = string("linear_92_cast_fp16")];
             fp16 var_1862_to_fp16 = const()[name = string("op_1862_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_1863_cast_fp16 = mul(x = linear_92_cast_fp16, y = var_1862_to_fp16)[name = string("op_1863_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_551_cast_fp16 = add(x = input_539_cast_fp16, y = var_1863_cast_fp16)[name = string("input_551_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_1863_cast_fp16 = mul(x = linear_92_cast_fp16, y = var_1862_to_fp16)[name = string("op_1863_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_551_cast_fp16 = add(x = input_539_cast_fp16, y = var_1863_cast_fp16)[name = string("input_551_cast_fp16")];
             tensor<int32, [1]> query_21_axes_0 = const()[name = string("query_21_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_10_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_10_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142034240)))];
-            tensor<fp16, [1024]> module_layers_10_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_10_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142036352)))];
-            tensor<fp16, [1, 188, 1024]> query_21_cast_fp16 = layer_norm(axes = query_21_axes_0, beta = module_layers_10_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_self_att_weight_to_fp16, x = input_551_cast_fp16)[name = string("query_21_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525519]> module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142038464))), indices_nonzero_data = tensor<uint6, [525519]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142169600))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142563840))))[name = string("module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_10_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_10_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136954688)))];
+            tensor<fp16, [1024]> module_layers_10_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_10_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136956800)))];
+            tensor<fp16, [1, 128, 1024]> query_21_cast_fp16 = layer_norm(axes = query_21_axes_0, beta = module_layers_10_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_self_att_weight_to_fp16, x = input_551_cast_fp16)[name = string("query_21_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525519]> module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136958912))), indices_nonzero_data = tensor<uint6, [525519]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137090048))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137484288))))[name = string("module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_93_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_21_cast_fp16)[name = string("linear_93_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_93_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_21_cast_fp16)[name = string("linear_93_cast_fp16")];
             tensor<int32, [4]> var_1879 = const()[name = string("op_1879"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_61_cast_fp16 = reshape(shape = var_1879, x = linear_93_cast_fp16)[name = string("q_61_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523272]> module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142580288))), indices_nonzero_data = tensor<uint6, [523272]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142711424))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143104000))))[name = string("module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> q_61_cast_fp16 = reshape(shape = var_1879, x = linear_93_cast_fp16)[name = string("q_61_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523272]> module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137500736))), indices_nonzero_data = tensor<uint6, [523272]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137631872))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138024448))))[name = string("module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_94_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_21_cast_fp16)[name = string("linear_94_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_94_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_21_cast_fp16)[name = string("linear_94_cast_fp16")];
             tensor<int32, [4]> var_1883 = const()[name = string("op_1883"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_41_cast_fp16 = reshape(shape = var_1883, x = linear_94_cast_fp16)[name = string("k_41_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523359]> module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143120448))), indices_nonzero_data = tensor<uint6, [523359]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143251584))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143644224))))[name = string("module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> k_41_cast_fp16 = reshape(shape = var_1883, x = linear_94_cast_fp16)[name = string("k_41_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523359]> module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138040896))), indices_nonzero_data = tensor<uint6, [523359]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138172032))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138564672))))[name = string("module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_95_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_21_cast_fp16)[name = string("linear_95_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_95_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_21_cast_fp16)[name = string("linear_95_cast_fp16")];
             tensor<int32, [4]> var_1887 = const()[name = string("op_1887"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_21_cast_fp16 = reshape(shape = var_1887, x = linear_95_cast_fp16)[name = string("v_21_cast_fp16")];
+            tensor<fp16, [1, 128, 8, 128]> v_21_cast_fp16 = reshape(shape = var_1887, x = linear_95_cast_fp16)[name = string("v_21_cast_fp16")];
             tensor<int32, [4]> value_23_perm_0 = const()[name = string("value_23_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_10_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_10_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143660672)))];
-            tensor<fp16, [1, 188, 8, 128]> var_1899_cast_fp16 = add(x = q_61_cast_fp16, y = module_layers_10_self_attn_pos_bias_u_to_fp16)[name = string("op_1899_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_10_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_10_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143662784)))];
-            tensor<fp16, [1, 188, 8, 128]> var_1901_cast_fp16 = add(x = q_61_cast_fp16, y = module_layers_10_self_attn_pos_bias_v_to_fp16)[name = string("op_1901_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_10_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_10_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138581120)))];
+            tensor<fp16, [1, 128, 8, 128]> var_1899_cast_fp16 = add(x = q_61_cast_fp16, y = module_layers_10_self_attn_pos_bias_u_to_fp16)[name = string("op_1899_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_10_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_10_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138583232)))];
+            tensor<fp16, [1, 128, 8, 128]> var_1901_cast_fp16 = add(x = q_61_cast_fp16, y = module_layers_10_self_attn_pos_bias_v_to_fp16)[name = string("op_1901_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_21_perm_0 = const()[name = string("q_with_bias_v_21_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_227_transpose_x_0 = const()[name = string("x_227_transpose_x_0"), val = bool(false)];
             bool x_227_transpose_y_0 = const()[name = string("x_227_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_1903_to_fp16 = const()[name = string("op_1903_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143664896)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_21_cast_fp16 = transpose(perm = q_with_bias_v_21_perm_0, x = var_1901_cast_fp16)[name = string("transpose_242")];
-            tensor<fp16, [1, 8, 188, 375]> x_227_cast_fp16 = matmul(transpose_x = x_227_transpose_x_0, transpose_y = x_227_transpose_y_0, x = q_with_bias_v_21_cast_fp16, y = var_1903_to_fp16)[name = string("x_227_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 255]> var_1903_to_fp16 = const()[name = string("op_1903_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312644864)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_21_cast_fp16 = transpose(perm = q_with_bias_v_21_perm_0, x = var_1901_cast_fp16)[name = string("transpose_242")];
+            tensor<fp16, [1, 8, 128, 255]> x_227_cast_fp16 = matmul(transpose_x = x_227_transpose_x_0, transpose_y = x_227_transpose_y_0, x = q_with_bias_v_21_cast_fp16, y = var_1903_to_fp16)[name = string("x_227_cast_fp16")];
             tensor<int32, [8]> x_229_pad_0 = const()[name = string("x_229_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_229_mode_0 = const()[name = string("x_229_mode_0"), val = string("constant")];
             fp16 const_114_to_fp16 = const()[name = string("const_114_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_229_cast_fp16 = pad(constant_val = const_114_to_fp16, mode = x_229_mode_0, pad = x_229_pad_0, x = x_227_cast_fp16)[name = string("x_229_cast_fp16")];
-            tensor<int32, [4]> var_1911 = const()[name = string("op_1911"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_231_cast_fp16 = reshape(shape = var_1911, x = x_229_cast_fp16)[name = string("x_231_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 256]> x_229_cast_fp16 = pad(constant_val = const_114_to_fp16, mode = x_229_mode_0, pad = x_229_pad_0, x = x_227_cast_fp16)[name = string("x_229_cast_fp16")];
+            tensor<int32, [4]> var_1911 = const()[name = string("op_1911"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_231_cast_fp16 = reshape(shape = var_1911, x = x_229_cast_fp16)[name = string("x_231_cast_fp16")];
             tensor<int32, [4]> var_1915_begin_0 = const()[name = string("op_1915_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_1915_end_0 = const()[name = string("op_1915_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_1915_end_0 = const()[name = string("op_1915_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
             tensor<bool, [4]> var_1915_end_mask_0 = const()[name = string("op_1915_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_1915_cast_fp16 = slice_by_index(begin = var_1915_begin_0, end = var_1915_end_0, end_mask = var_1915_end_mask_0, x = x_231_cast_fp16)[name = string("op_1915_cast_fp16")];
-            tensor<int32, [4]> var_1916 = const()[name = string("op_1916"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_41_cast_fp16 = reshape(shape = var_1916, x = var_1915_cast_fp16)[name = string("matrix_bd_41_cast_fp16")];
+            tensor<fp16, [1, 8, 255, 128]> var_1915_cast_fp16 = slice_by_index(begin = var_1915_begin_0, end = var_1915_end_0, end_mask = var_1915_end_mask_0, x = x_231_cast_fp16)[name = string("op_1915_cast_fp16")];
+            tensor<int32, [4]> var_1916 = const()[name = string("op_1916"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_41_cast_fp16 = reshape(shape = var_1916, x = var_1915_cast_fp16)[name = string("matrix_bd_41_cast_fp16")];
             bool matrix_ac_21_transpose_x_0 = const()[name = string("matrix_ac_21_transpose_x_0"), val = bool(false)];
             bool matrix_ac_21_transpose_y_0 = const()[name = string("matrix_ac_21_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_116_perm_0 = const()[name = string("transpose_116_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_117_perm_0 = const()[name = string("transpose_117_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_117 = transpose(perm = transpose_117_perm_0, x = k_41_cast_fp16)[name = string("transpose_240")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_116 = transpose(perm = transpose_116_perm_0, x = var_1899_cast_fp16)[name = string("transpose_241")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_21_cast_fp16 = matmul(transpose_x = matrix_ac_21_transpose_x_0, transpose_y = matrix_ac_21_transpose_y_0, x = transpose_116, y = transpose_117)[name = string("matrix_ac_21_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_117 = transpose(perm = transpose_117_perm_0, x = k_41_cast_fp16)[name = string("transpose_240")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_116 = transpose(perm = transpose_116_perm_0, x = var_1899_cast_fp16)[name = string("transpose_241")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_21_cast_fp16 = matmul(transpose_x = matrix_ac_21_transpose_x_0, transpose_y = matrix_ac_21_transpose_y_0, x = transpose_116, y = transpose_117)[name = string("matrix_ac_21_cast_fp16")];
             tensor<int32, [4]> matrix_bd_43_begin_0 = const()[name = string("matrix_bd_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_43_end_0 = const()[name = string("matrix_bd_43_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_43_end_0 = const()[name = string("matrix_bd_43_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
             tensor<bool, [4]> matrix_bd_43_end_mask_0 = const()[name = string("matrix_bd_43_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_43_cast_fp16 = slice_by_index(begin = matrix_bd_43_begin_0, end = matrix_bd_43_end_0, end_mask = matrix_bd_43_end_mask_0, x = matrix_bd_41_cast_fp16)[name = string("matrix_bd_43_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_1925_cast_fp16 = add(x = matrix_ac_21_cast_fp16, y = matrix_bd_43_cast_fp16)[name = string("op_1925_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_43_cast_fp16 = slice_by_index(begin = matrix_bd_43_begin_0, end = matrix_bd_43_end_0, end_mask = matrix_bd_43_end_mask_0, x = matrix_bd_41_cast_fp16)[name = string("matrix_bd_43_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_1925_cast_fp16 = add(x = matrix_ac_21_cast_fp16, y = matrix_bd_43_cast_fp16)[name = string("op_1925_cast_fp16")];
             fp16 _inversed_scores_41_y_0_to_fp16 = const()[name = string("_inversed_scores_41_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_41_cast_fp16 = mul(x = var_1925_cast_fp16, y = _inversed_scores_41_y_0_to_fp16)[name = string("_inversed_scores_41_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_43_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_41_cast_fp16, cond = mask_3)[name = string("scores_43_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_1931_cast_fp16 = softmax(axis = var_30, x = scores_43_cast_fp16)[name = string("op_1931_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_553_cast_fp16 = select(a = var_11_to_fp16, b = var_1931_cast_fp16, cond = mask_3)[name = string("input_553_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_41_cast_fp16 = mul(x = var_1925_cast_fp16, y = _inversed_scores_41_y_0_to_fp16)[name = string("_inversed_scores_41_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_43_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_41_cast_fp16, cond = mask_3)[name = string("scores_43_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_1931_cast_fp16 = softmax(axis = var_30, x = scores_43_cast_fp16)[name = string("op_1931_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_553_cast_fp16 = select(a = var_11_to_fp16, b = var_1931_cast_fp16, cond = mask_3)[name = string("input_553_cast_fp16")];
             bool x_233_transpose_x_0 = const()[name = string("x_233_transpose_x_0"), val = bool(false)];
             bool x_233_transpose_y_0 = const()[name = string("x_233_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_23_cast_fp16 = transpose(perm = value_23_perm_0, x = v_21_cast_fp16)[name = string("transpose_239")];
-            tensor<fp16, [1, 8, 188, 128]> x_233_cast_fp16 = matmul(transpose_x = x_233_transpose_x_0, transpose_y = x_233_transpose_y_0, x = input_553_cast_fp16, y = value_23_cast_fp16)[name = string("x_233_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> value_23_cast_fp16 = transpose(perm = value_23_perm_0, x = v_21_cast_fp16)[name = string("transpose_239")];
+            tensor<fp16, [1, 8, 128, 128]> x_233_cast_fp16 = matmul(transpose_x = x_233_transpose_x_0, transpose_y = x_233_transpose_y_0, x = input_553_cast_fp16, y = value_23_cast_fp16)[name = string("x_233_cast_fp16")];
             tensor<int32, [4]> var_1935_perm_0 = const()[name = string("op_1935_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_1936 = const()[name = string("op_1936"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_1935_cast_fp16 = transpose(perm = var_1935_perm_0, x = x_233_cast_fp16)[name = string("transpose_238")];
-            tensor<fp16, [1, 188, 1024]> input_555_cast_fp16 = reshape(shape = var_1936, x = var_1935_cast_fp16)[name = string("input_555_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524561]> module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144432960))), indices_nonzero_data = tensor<uint6, [524561]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144564096))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144957632))))[name = string("module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> var_1935_cast_fp16 = transpose(perm = var_1935_perm_0, x = x_233_cast_fp16)[name = string("transpose_238")];
+            tensor<fp16, [1, 128, 1024]> input_555_cast_fp16 = reshape(shape = var_1936, x = var_1935_cast_fp16)[name = string("input_555_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524561]> module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138845504))), indices_nonzero_data = tensor<uint6, [524561]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138976640))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139370176))))[name = string("module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_97_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_555_cast_fp16)[name = string("linear_97_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_559_cast_fp16 = add(x = input_551_cast_fp16, y = linear_97_cast_fp16)[name = string("input_559_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_97_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_555_cast_fp16)[name = string("linear_97_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_559_cast_fp16 = add(x = input_551_cast_fp16, y = linear_97_cast_fp16)[name = string("input_559_cast_fp16")];
             tensor<int32, [1]> x_237_axes_0 = const()[name = string("x_237_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_10_norm_conv_weight_to_fp16 = const()[name = string("module_layers_10_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144974080)))];
-            tensor<fp16, [1024]> module_layers_10_norm_conv_bias_to_fp16 = const()[name = string("module_layers_10_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144976192)))];
-            tensor<fp16, [1, 188, 1024]> x_237_cast_fp16 = layer_norm(axes = x_237_axes_0, beta = module_layers_10_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_conv_weight_to_fp16, x = input_559_cast_fp16)[name = string("x_237_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_10_norm_conv_weight_to_fp16 = const()[name = string("module_layers_10_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139386624)))];
+            tensor<fp16, [1024]> module_layers_10_norm_conv_bias_to_fp16 = const()[name = string("module_layers_10_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139388736)))];
+            tensor<fp16, [1, 128, 1024]> x_237_cast_fp16 = layer_norm(axes = x_237_axes_0, beta = module_layers_10_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_conv_weight_to_fp16, x = input_559_cast_fp16)[name = string("x_237_cast_fp16")];
             tensor<int32, [3]> input_561_perm_0 = const()[name = string("input_561_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_563_pad_type_0 = const()[name = string("input_563_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_563_strides_0 = const()[name = string("input_563_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_563_pad_0 = const()[name = string("input_563_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_563_dilations_0 = const()[name = string("input_563_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_563_groups_0 = const()[name = string("input_563_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1050237]> module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144978304))), indices_nonzero_data = tensor<uint6, [1050237]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145240512))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146028288))))[name = string("module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1050237]> module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139390848))), indices_nonzero_data = tensor<uint6, [1050237]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139653056))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140440832))))[name = string("module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_561_cast_fp16 = transpose(perm = input_561_perm_0, x = x_237_cast_fp16)[name = string("transpose_237")];
-            tensor<fp16, [1, 2048, 188]> input_563_cast_fp16 = conv(dilations = input_563_dilations_0, groups = input_563_groups_0, pad = input_563_pad_0, pad_type = input_563_pad_type_0, strides = input_563_strides_0, weight = module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_561_cast_fp16)[name = string("input_563_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_561_cast_fp16 = transpose(perm = input_561_perm_0, x = x_237_cast_fp16)[name = string("transpose_237")];
+            tensor<fp16, [1, 2048, 128]> input_563_cast_fp16 = conv(dilations = input_563_dilations_0, groups = input_563_groups_0, pad = input_563_pad_0, pad_type = input_563_pad_type_0, strides = input_563_strides_0, weight = module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_561_cast_fp16)[name = string("input_563_cast_fp16")];
             int32 x_239_split_num_splits_0 = const()[name = string("x_239_split_num_splits_0"), val = int32(2)];
             int32 x_239_split_axis_0 = const()[name = string("x_239_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_239_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_239_split_cast_fp16_1 = split(axis = x_239_split_axis_0, num_splits = x_239_split_num_splits_0, x = input_563_cast_fp16)[name = string("x_239_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_239_split_1_sigmoid_cast_fp16 = sigmoid(x = x_239_split_cast_fp16_1)[name = string("x_239_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_239_cast_fp16 = mul(x = x_239_split_cast_fp16_0, y = x_239_split_1_sigmoid_cast_fp16)[name = string("x_239_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_565_cast_fp16 = select(a = var_11_to_fp16, b = x_239_cast_fp16, cond = var_328)[name = string("input_565_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_239_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_239_split_cast_fp16_1 = split(axis = x_239_split_axis_0, num_splits = x_239_split_num_splits_0, x = input_563_cast_fp16)[name = string("x_239_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_239_split_1_sigmoid_cast_fp16 = sigmoid(x = x_239_split_cast_fp16_1)[name = string("x_239_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_239_cast_fp16 = mul(x = x_239_split_cast_fp16_0, y = x_239_split_1_sigmoid_cast_fp16)[name = string("x_239_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_565_cast_fp16 = select(a = var_11_to_fp16, b = x_239_cast_fp16, cond = var_328)[name = string("input_565_cast_fp16")];
             tensor<int32, [6]> input_567_pad_0 = const()[name = string("input_567_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_567_mode_0 = const()[name = string("input_567_mode_0"), val = string("constant")];
             fp16 const_117_to_fp16 = const()[name = string("const_117_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_567_cast_fp16 = pad(constant_val = const_117_to_fp16, mode = input_567_mode_0, pad = input_567_pad_0, x = input_565_cast_fp16)[name = string("input_567_cast_fp16")];
+            tensor<fp16, [1, 1024, 136]> input_567_cast_fp16 = pad(constant_val = const_117_to_fp16, mode = input_567_mode_0, pad = input_567_pad_0, x = input_565_cast_fp16)[name = string("input_567_cast_fp16")];
             string input_569_pad_type_0 = const()[name = string("input_569_pad_type_0"), val = string("valid")];
             int32 input_569_groups_0 = const()[name = string("input_569_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_569_strides_0 = const()[name = string("input_569_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_569_pad_0 = const()[name = string("input_569_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_569_dilations_0 = const()[name = string("input_569_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_268_to_fp16 = const()[name = string("const_268_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146061120)))];
-            tensor<fp16, [1024]> const_269_to_fp16 = const()[name = string("const_269_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146079616)))];
-            tensor<fp16, [1, 1024, 188]> input_571_cast_fp16 = conv(bias = const_269_to_fp16, dilations = input_569_dilations_0, groups = input_569_groups_0, pad = input_569_pad_0, pad_type = input_569_pad_type_0, strides = input_569_strides_0, weight = const_268_to_fp16, x = input_567_cast_fp16)[name = string("input_571_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_573_cast_fp16 = silu(x = input_571_cast_fp16)[name = string("input_573_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_268_to_fp16 = const()[name = string("const_268_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140473664)))];
+            tensor<fp16, [1024]> const_269_to_fp16 = const()[name = string("const_269_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140492160)))];
+            tensor<fp16, [1, 1024, 128]> input_571_cast_fp16 = conv(bias = const_269_to_fp16, dilations = input_569_dilations_0, groups = input_569_groups_0, pad = input_569_pad_0, pad_type = input_569_pad_type_0, strides = input_569_strides_0, weight = const_268_to_fp16, x = input_567_cast_fp16)[name = string("input_571_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_573_cast_fp16 = silu(x = input_571_cast_fp16)[name = string("input_573_cast_fp16")];
             string x_241_pad_type_0 = const()[name = string("x_241_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_241_strides_0 = const()[name = string("x_241_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_241_pad_0 = const()[name = string("x_241_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_241_dilations_0 = const()[name = string("x_241_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_241_groups_0 = const()[name = string("x_241_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520978]> module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146081728))), indices_nonzero_data = tensor<uint6, [520978]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146212864))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146603712))))[name = string("module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520978]> module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140494272))), indices_nonzero_data = tensor<uint6, [520978]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140625408))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141016256))))[name = string("module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_241_cast_fp16 = conv(dilations = x_241_dilations_0, groups = x_241_groups_0, pad = x_241_pad_0, pad_type = x_241_pad_type_0, strides = x_241_strides_0, weight = module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_573_cast_fp16)[name = string("x_241_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_241_cast_fp16 = conv(dilations = x_241_dilations_0, groups = x_241_groups_0, pad = x_241_pad_0, pad_type = x_241_pad_type_0, strides = x_241_strides_0, weight = module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_573_cast_fp16)[name = string("x_241_cast_fp16")];
             tensor<int32, [3]> input_575_perm_0 = const()[name = string("input_575_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_575_cast_fp16 = transpose(perm = input_575_perm_0, x = x_241_cast_fp16)[name = string("transpose_236")];
-            tensor<fp16, [1, 188, 1024]> input_577_cast_fp16 = add(x = input_559_cast_fp16, y = input_575_cast_fp16)[name = string("input_577_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_575_cast_fp16 = transpose(perm = input_575_perm_0, x = x_241_cast_fp16)[name = string("transpose_236")];
+            tensor<fp16, [1, 128, 1024]> input_577_cast_fp16 = add(x = input_559_cast_fp16, y = input_575_cast_fp16)[name = string("input_577_cast_fp16")];
             tensor<int32, [1]> input_579_axes_0 = const()[name = string("input_579_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_10_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_10_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146620160)))];
-            tensor<fp16, [1024]> module_layers_10_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_10_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146622272)))];
-            tensor<fp16, [1, 188, 1024]> input_579_cast_fp16 = layer_norm(axes = input_579_axes_0, beta = module_layers_10_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_feed_forward2_weight_to_fp16, x = input_577_cast_fp16)[name = string("input_579_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2057233]> module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146624384))), indices_nonzero_data = tensor<uint6, [2057233]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147148736))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148691776))))[name = string("module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_10_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_10_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141032704)))];
+            tensor<fp16, [1024]> module_layers_10_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_10_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141034816)))];
+            tensor<fp16, [1, 128, 1024]> input_579_cast_fp16 = layer_norm(axes = input_579_axes_0, beta = module_layers_10_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_feed_forward2_weight_to_fp16, x = input_577_cast_fp16)[name = string("input_579_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2057233]> module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141036928))), indices_nonzero_data = tensor<uint6, [2057233]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141561280))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143104320))))[name = string("module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_98_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_579_cast_fp16)[name = string("linear_98_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_583_cast_fp16 = silu(x = linear_98_cast_fp16)[name = string("input_583_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2031712]> module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148757376))), indices_nonzero_data = tensor<uint6, [2031712]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149281728))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150805632))))[name = string("module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_98_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_579_cast_fp16)[name = string("linear_98_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_583_cast_fp16 = silu(x = linear_98_cast_fp16)[name = string("input_583_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2031712]> module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143169920))), indices_nonzero_data = tensor<uint6, [2031712]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143694272))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145218176))))[name = string("module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_99_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_583_cast_fp16)[name = string("linear_99_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_99_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_583_cast_fp16)[name = string("linear_99_cast_fp16")];
             fp16 var_1996_to_fp16 = const()[name = string("op_1996_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_1997_cast_fp16 = mul(x = linear_99_cast_fp16, y = var_1996_to_fp16)[name = string("op_1997_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_589_cast_fp16 = add(x = input_577_cast_fp16, y = var_1997_cast_fp16)[name = string("input_589_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_1997_cast_fp16 = mul(x = linear_99_cast_fp16, y = var_1996_to_fp16)[name = string("op_1997_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_589_cast_fp16 = add(x = input_577_cast_fp16, y = var_1997_cast_fp16)[name = string("input_589_cast_fp16")];
             tensor<int32, [1]> input_591_axes_0 = const()[name = string("input_591_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_10_norm_out_weight_to_fp16 = const()[name = string("module_layers_10_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150822080)))];
-            tensor<fp16, [1024]> module_layers_10_norm_out_bias_to_fp16 = const()[name = string("module_layers_10_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150824192)))];
-            tensor<fp16, [1, 188, 1024]> input_591_cast_fp16 = layer_norm(axes = input_591_axes_0, beta = module_layers_10_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_out_weight_to_fp16, x = input_589_cast_fp16)[name = string("input_591_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_10_norm_out_weight_to_fp16 = const()[name = string("module_layers_10_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145234624)))];
+            tensor<fp16, [1024]> module_layers_10_norm_out_bias_to_fp16 = const()[name = string("module_layers_10_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145236736)))];
+            tensor<fp16, [1, 128, 1024]> input_591_cast_fp16 = layer_norm(axes = input_591_axes_0, beta = module_layers_10_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_out_weight_to_fp16, x = input_589_cast_fp16)[name = string("input_591_cast_fp16")];
             tensor<int32, [1]> input_593_axes_0 = const()[name = string("input_593_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_11_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_11_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150826304)))];
-            tensor<fp16, [1024]> module_layers_11_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_11_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150828416)))];
-            tensor<fp16, [1, 188, 1024]> input_593_cast_fp16 = layer_norm(axes = input_593_axes_0, beta = module_layers_11_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_feed_forward1_weight_to_fp16, x = input_591_cast_fp16)[name = string("input_593_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2039284]> module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150830528))), indices_nonzero_data = tensor<uint6, [2039284]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151354880))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152884416))))[name = string("module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_11_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_11_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145238848)))];
+            tensor<fp16, [1024]> module_layers_11_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_11_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145240960)))];
+            tensor<fp16, [1, 128, 1024]> input_593_cast_fp16 = layer_norm(axes = input_593_axes_0, beta = module_layers_11_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_feed_forward1_weight_to_fp16, x = input_591_cast_fp16)[name = string("input_593_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2039284]> module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145243072))), indices_nonzero_data = tensor<uint6, [2039284]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145767424))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147296960))))[name = string("module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_100_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_593_cast_fp16)[name = string("linear_100_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_597_cast_fp16 = silu(x = linear_100_cast_fp16)[name = string("input_597_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2009766]> module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152950016))), indices_nonzero_data = tensor<uint6, [2009766]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153474368))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154981760))))[name = string("module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_100_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_593_cast_fp16)[name = string("linear_100_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_597_cast_fp16 = silu(x = linear_100_cast_fp16)[name = string("input_597_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2009766]> module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147362560))), indices_nonzero_data = tensor<uint6, [2009766]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147886912))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149394304))))[name = string("module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_101_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_597_cast_fp16)[name = string("linear_101_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_101_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_597_cast_fp16)[name = string("linear_101_cast_fp16")];
             fp16 var_2025_to_fp16 = const()[name = string("op_2025_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_2026_cast_fp16 = mul(x = linear_101_cast_fp16, y = var_2025_to_fp16)[name = string("op_2026_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_603_cast_fp16 = add(x = input_591_cast_fp16, y = var_2026_cast_fp16)[name = string("input_603_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_2026_cast_fp16 = mul(x = linear_101_cast_fp16, y = var_2025_to_fp16)[name = string("op_2026_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_603_cast_fp16 = add(x = input_591_cast_fp16, y = var_2026_cast_fp16)[name = string("input_603_cast_fp16")];
             tensor<int32, [1]> query_23_axes_0 = const()[name = string("query_23_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_11_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_11_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154998208)))];
-            tensor<fp16, [1024]> module_layers_11_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_11_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155000320)))];
-            tensor<fp16, [1, 188, 1024]> query_23_cast_fp16 = layer_norm(axes = query_23_axes_0, beta = module_layers_11_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_self_att_weight_to_fp16, x = input_603_cast_fp16)[name = string("query_23_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525654]> module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155002432))), indices_nonzero_data = tensor<uint6, [525654]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155133568))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155527936))))[name = string("module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_11_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_11_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149410752)))];
+            tensor<fp16, [1024]> module_layers_11_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_11_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149412864)))];
+            tensor<fp16, [1, 128, 1024]> query_23_cast_fp16 = layer_norm(axes = query_23_axes_0, beta = module_layers_11_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_self_att_weight_to_fp16, x = input_603_cast_fp16)[name = string("query_23_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525654]> module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149414976))), indices_nonzero_data = tensor<uint6, [525654]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149546112))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149940480))))[name = string("module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_102_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_23_cast_fp16)[name = string("linear_102_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_102_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_23_cast_fp16)[name = string("linear_102_cast_fp16")];
             tensor<int32, [4]> var_2042 = const()[name = string("op_2042"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_67_cast_fp16 = reshape(shape = var_2042, x = linear_102_cast_fp16)[name = string("q_67_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522850]> module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155544384))), indices_nonzero_data = tensor<uint6, [522850]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155675520))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156067776))))[name = string("module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> q_67_cast_fp16 = reshape(shape = var_2042, x = linear_102_cast_fp16)[name = string("q_67_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522850]> module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149956928))), indices_nonzero_data = tensor<uint6, [522850]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150088064))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150480320))))[name = string("module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_103_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_23_cast_fp16)[name = string("linear_103_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_103_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_23_cast_fp16)[name = string("linear_103_cast_fp16")];
             tensor<int32, [4]> var_2046 = const()[name = string("op_2046"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_45_cast_fp16 = reshape(shape = var_2046, x = linear_103_cast_fp16)[name = string("k_45_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522138]> module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156084224))), indices_nonzero_data = tensor<uint6, [522138]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156215360))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156607040))))[name = string("module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> k_45_cast_fp16 = reshape(shape = var_2046, x = linear_103_cast_fp16)[name = string("k_45_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522138]> module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150496768))), indices_nonzero_data = tensor<uint6, [522138]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150627904))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151019584))))[name = string("module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_104_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_23_cast_fp16)[name = string("linear_104_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_104_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_23_cast_fp16)[name = string("linear_104_cast_fp16")];
             tensor<int32, [4]> var_2050 = const()[name = string("op_2050"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_23_cast_fp16 = reshape(shape = var_2050, x = linear_104_cast_fp16)[name = string("v_23_cast_fp16")];
+            tensor<fp16, [1, 128, 8, 128]> v_23_cast_fp16 = reshape(shape = var_2050, x = linear_104_cast_fp16)[name = string("v_23_cast_fp16")];
             tensor<int32, [4]> value_25_perm_0 = const()[name = string("value_25_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_11_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_11_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156623488)))];
-            tensor<fp16, [1, 188, 8, 128]> var_2062_cast_fp16 = add(x = q_67_cast_fp16, y = module_layers_11_self_attn_pos_bias_u_to_fp16)[name = string("op_2062_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_11_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_11_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156625600)))];
-            tensor<fp16, [1, 188, 8, 128]> var_2064_cast_fp16 = add(x = q_67_cast_fp16, y = module_layers_11_self_attn_pos_bias_v_to_fp16)[name = string("op_2064_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_11_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_11_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151036032)))];
+            tensor<fp16, [1, 128, 8, 128]> var_2062_cast_fp16 = add(x = q_67_cast_fp16, y = module_layers_11_self_attn_pos_bias_u_to_fp16)[name = string("op_2062_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_11_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_11_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151038144)))];
+            tensor<fp16, [1, 128, 8, 128]> var_2064_cast_fp16 = add(x = q_67_cast_fp16, y = module_layers_11_self_attn_pos_bias_v_to_fp16)[name = string("op_2064_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_23_perm_0 = const()[name = string("q_with_bias_v_23_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_249_transpose_x_0 = const()[name = string("x_249_transpose_x_0"), val = bool(false)];
             bool x_249_transpose_y_0 = const()[name = string("x_249_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_2066_to_fp16 = const()[name = string("op_2066_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156627712)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_23_cast_fp16 = transpose(perm = q_with_bias_v_23_perm_0, x = var_2064_cast_fp16)[name = string("transpose_235")];
-            tensor<fp16, [1, 8, 188, 375]> x_249_cast_fp16 = matmul(transpose_x = x_249_transpose_x_0, transpose_y = x_249_transpose_y_0, x = q_with_bias_v_23_cast_fp16, y = var_2066_to_fp16)[name = string("x_249_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 255]> var_2066_to_fp16 = const()[name = string("op_2066_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313167168)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_23_cast_fp16 = transpose(perm = q_with_bias_v_23_perm_0, x = var_2064_cast_fp16)[name = string("transpose_235")];
+            tensor<fp16, [1, 8, 128, 255]> x_249_cast_fp16 = matmul(transpose_x = x_249_transpose_x_0, transpose_y = x_249_transpose_y_0, x = q_with_bias_v_23_cast_fp16, y = var_2066_to_fp16)[name = string("x_249_cast_fp16")];
             tensor<int32, [8]> x_251_pad_0 = const()[name = string("x_251_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_251_mode_0 = const()[name = string("x_251_mode_0"), val = string("constant")];
             fp16 const_124_to_fp16 = const()[name = string("const_124_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_251_cast_fp16 = pad(constant_val = const_124_to_fp16, mode = x_251_mode_0, pad = x_251_pad_0, x = x_249_cast_fp16)[name = string("x_251_cast_fp16")];
-            tensor<int32, [4]> var_2074 = const()[name = string("op_2074"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_253_cast_fp16 = reshape(shape = var_2074, x = x_251_cast_fp16)[name = string("x_253_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 256]> x_251_cast_fp16 = pad(constant_val = const_124_to_fp16, mode = x_251_mode_0, pad = x_251_pad_0, x = x_249_cast_fp16)[name = string("x_251_cast_fp16")];
+            tensor<int32, [4]> var_2074 = const()[name = string("op_2074"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_253_cast_fp16 = reshape(shape = var_2074, x = x_251_cast_fp16)[name = string("x_253_cast_fp16")];
             tensor<int32, [4]> var_2078_begin_0 = const()[name = string("op_2078_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_2078_end_0 = const()[name = string("op_2078_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_2078_end_0 = const()[name = string("op_2078_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
             tensor<bool, [4]> var_2078_end_mask_0 = const()[name = string("op_2078_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_2078_cast_fp16 = slice_by_index(begin = var_2078_begin_0, end = var_2078_end_0, end_mask = var_2078_end_mask_0, x = x_253_cast_fp16)[name = string("op_2078_cast_fp16")];
-            tensor<int32, [4]> var_2079 = const()[name = string("op_2079"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_45_cast_fp16 = reshape(shape = var_2079, x = var_2078_cast_fp16)[name = string("matrix_bd_45_cast_fp16")];
+            tensor<fp16, [1, 8, 255, 128]> var_2078_cast_fp16 = slice_by_index(begin = var_2078_begin_0, end = var_2078_end_0, end_mask = var_2078_end_mask_0, x = x_253_cast_fp16)[name = string("op_2078_cast_fp16")];
+            tensor<int32, [4]> var_2079 = const()[name = string("op_2079"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_45_cast_fp16 = reshape(shape = var_2079, x = var_2078_cast_fp16)[name = string("matrix_bd_45_cast_fp16")];
             bool matrix_ac_23_transpose_x_0 = const()[name = string("matrix_ac_23_transpose_x_0"), val = bool(false)];
             bool matrix_ac_23_transpose_y_0 = const()[name = string("matrix_ac_23_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_118_perm_0 = const()[name = string("transpose_118_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_119_perm_0 = const()[name = string("transpose_119_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_119 = transpose(perm = transpose_119_perm_0, x = k_45_cast_fp16)[name = string("transpose_233")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_118 = transpose(perm = transpose_118_perm_0, x = var_2062_cast_fp16)[name = string("transpose_234")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_23_cast_fp16 = matmul(transpose_x = matrix_ac_23_transpose_x_0, transpose_y = matrix_ac_23_transpose_y_0, x = transpose_118, y = transpose_119)[name = string("matrix_ac_23_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_119 = transpose(perm = transpose_119_perm_0, x = k_45_cast_fp16)[name = string("transpose_233")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_118 = transpose(perm = transpose_118_perm_0, x = var_2062_cast_fp16)[name = string("transpose_234")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_23_cast_fp16 = matmul(transpose_x = matrix_ac_23_transpose_x_0, transpose_y = matrix_ac_23_transpose_y_0, x = transpose_118, y = transpose_119)[name = string("matrix_ac_23_cast_fp16")];
             tensor<int32, [4]> matrix_bd_47_begin_0 = const()[name = string("matrix_bd_47_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_47_end_0 = const()[name = string("matrix_bd_47_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_47_end_0 = const()[name = string("matrix_bd_47_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
             tensor<bool, [4]> matrix_bd_47_end_mask_0 = const()[name = string("matrix_bd_47_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_47_cast_fp16 = slice_by_index(begin = matrix_bd_47_begin_0, end = matrix_bd_47_end_0, end_mask = matrix_bd_47_end_mask_0, x = matrix_bd_45_cast_fp16)[name = string("matrix_bd_47_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_2088_cast_fp16 = add(x = matrix_ac_23_cast_fp16, y = matrix_bd_47_cast_fp16)[name = string("op_2088_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_47_cast_fp16 = slice_by_index(begin = matrix_bd_47_begin_0, end = matrix_bd_47_end_0, end_mask = matrix_bd_47_end_mask_0, x = matrix_bd_45_cast_fp16)[name = string("matrix_bd_47_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_2088_cast_fp16 = add(x = matrix_ac_23_cast_fp16, y = matrix_bd_47_cast_fp16)[name = string("op_2088_cast_fp16")];
             fp16 _inversed_scores_45_y_0_to_fp16 = const()[name = string("_inversed_scores_45_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_45_cast_fp16 = mul(x = var_2088_cast_fp16, y = _inversed_scores_45_y_0_to_fp16)[name = string("_inversed_scores_45_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_47_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_45_cast_fp16, cond = mask_3)[name = string("scores_47_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_2094_cast_fp16 = softmax(axis = var_30, x = scores_47_cast_fp16)[name = string("op_2094_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_605_cast_fp16 = select(a = var_11_to_fp16, b = var_2094_cast_fp16, cond = mask_3)[name = string("input_605_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_45_cast_fp16 = mul(x = var_2088_cast_fp16, y = _inversed_scores_45_y_0_to_fp16)[name = string("_inversed_scores_45_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_47_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_45_cast_fp16, cond = mask_3)[name = string("scores_47_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_2094_cast_fp16 = softmax(axis = var_30, x = scores_47_cast_fp16)[name = string("op_2094_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_605_cast_fp16 = select(a = var_11_to_fp16, b = var_2094_cast_fp16, cond = mask_3)[name = string("input_605_cast_fp16")];
             bool x_255_transpose_x_0 = const()[name = string("x_255_transpose_x_0"), val = bool(false)];
             bool x_255_transpose_y_0 = const()[name = string("x_255_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_25_cast_fp16 = transpose(perm = value_25_perm_0, x = v_23_cast_fp16)[name = string("transpose_232")];
-            tensor<fp16, [1, 8, 188, 128]> x_255_cast_fp16 = matmul(transpose_x = x_255_transpose_x_0, transpose_y = x_255_transpose_y_0, x = input_605_cast_fp16, y = value_25_cast_fp16)[name = string("x_255_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> value_25_cast_fp16 = transpose(perm = value_25_perm_0, x = v_23_cast_fp16)[name = string("transpose_232")];
+            tensor<fp16, [1, 8, 128, 128]> x_255_cast_fp16 = matmul(transpose_x = x_255_transpose_x_0, transpose_y = x_255_transpose_y_0, x = input_605_cast_fp16, y = value_25_cast_fp16)[name = string("x_255_cast_fp16")];
             tensor<int32, [4]> var_2098_perm_0 = const()[name = string("op_2098_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_2099 = const()[name = string("op_2099"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_2098_cast_fp16 = transpose(perm = var_2098_perm_0, x = x_255_cast_fp16)[name = string("transpose_231")];
-            tensor<fp16, [1, 188, 1024]> input_607_cast_fp16 = reshape(shape = var_2099, x = var_2098_cast_fp16)[name = string("input_607_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524206]> module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157395776))), indices_nonzero_data = tensor<uint6, [524206]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157526912))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157920192))))[name = string("module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> var_2098_cast_fp16 = transpose(perm = var_2098_perm_0, x = x_255_cast_fp16)[name = string("transpose_231")];
+            tensor<fp16, [1, 128, 1024]> input_607_cast_fp16 = reshape(shape = var_2099, x = var_2098_cast_fp16)[name = string("input_607_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524206]> module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151300416))), indices_nonzero_data = tensor<uint6, [524206]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151431552))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151824832))))[name = string("module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_106_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_607_cast_fp16)[name = string("linear_106_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_611_cast_fp16 = add(x = input_603_cast_fp16, y = linear_106_cast_fp16)[name = string("input_611_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_106_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_607_cast_fp16)[name = string("linear_106_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_611_cast_fp16 = add(x = input_603_cast_fp16, y = linear_106_cast_fp16)[name = string("input_611_cast_fp16")];
             tensor<int32, [1]> x_259_axes_0 = const()[name = string("x_259_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_11_norm_conv_weight_to_fp16 = const()[name = string("module_layers_11_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157936640)))];
-            tensor<fp16, [1024]> module_layers_11_norm_conv_bias_to_fp16 = const()[name = string("module_layers_11_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157938752)))];
-            tensor<fp16, [1, 188, 1024]> x_259_cast_fp16 = layer_norm(axes = x_259_axes_0, beta = module_layers_11_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_conv_weight_to_fp16, x = input_611_cast_fp16)[name = string("x_259_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_11_norm_conv_weight_to_fp16 = const()[name = string("module_layers_11_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151841280)))];
+            tensor<fp16, [1024]> module_layers_11_norm_conv_bias_to_fp16 = const()[name = string("module_layers_11_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151843392)))];
+            tensor<fp16, [1, 128, 1024]> x_259_cast_fp16 = layer_norm(axes = x_259_axes_0, beta = module_layers_11_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_conv_weight_to_fp16, x = input_611_cast_fp16)[name = string("x_259_cast_fp16")];
             tensor<int32, [3]> input_613_perm_0 = const()[name = string("input_613_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_615_pad_type_0 = const()[name = string("input_615_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_615_strides_0 = const()[name = string("input_615_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_615_pad_0 = const()[name = string("input_615_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_615_dilations_0 = const()[name = string("input_615_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_615_groups_0 = const()[name = string("input_615_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1049241]> module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157940864))), indices_nonzero_data = tensor<uint6, [1049241]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158203072))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158990080))))[name = string("module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1049241]> module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151845504))), indices_nonzero_data = tensor<uint6, [1049241]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152107712))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152894720))))[name = string("module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_613_cast_fp16 = transpose(perm = input_613_perm_0, x = x_259_cast_fp16)[name = string("transpose_230")];
-            tensor<fp16, [1, 2048, 188]> input_615_cast_fp16 = conv(dilations = input_615_dilations_0, groups = input_615_groups_0, pad = input_615_pad_0, pad_type = input_615_pad_type_0, strides = input_615_strides_0, weight = module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_613_cast_fp16)[name = string("input_615_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_613_cast_fp16 = transpose(perm = input_613_perm_0, x = x_259_cast_fp16)[name = string("transpose_230")];
+            tensor<fp16, [1, 2048, 128]> input_615_cast_fp16 = conv(dilations = input_615_dilations_0, groups = input_615_groups_0, pad = input_615_pad_0, pad_type = input_615_pad_type_0, strides = input_615_strides_0, weight = module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_613_cast_fp16)[name = string("input_615_cast_fp16")];
             int32 x_261_split_num_splits_0 = const()[name = string("x_261_split_num_splits_0"), val = int32(2)];
             int32 x_261_split_axis_0 = const()[name = string("x_261_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_261_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_261_split_cast_fp16_1 = split(axis = x_261_split_axis_0, num_splits = x_261_split_num_splits_0, x = input_615_cast_fp16)[name = string("x_261_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_261_split_1_sigmoid_cast_fp16 = sigmoid(x = x_261_split_cast_fp16_1)[name = string("x_261_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_261_cast_fp16 = mul(x = x_261_split_cast_fp16_0, y = x_261_split_1_sigmoid_cast_fp16)[name = string("x_261_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_617_cast_fp16 = select(a = var_11_to_fp16, b = x_261_cast_fp16, cond = var_328)[name = string("input_617_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_261_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_261_split_cast_fp16_1 = split(axis = x_261_split_axis_0, num_splits = x_261_split_num_splits_0, x = input_615_cast_fp16)[name = string("x_261_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_261_split_1_sigmoid_cast_fp16 = sigmoid(x = x_261_split_cast_fp16_1)[name = string("x_261_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_261_cast_fp16 = mul(x = x_261_split_cast_fp16_0, y = x_261_split_1_sigmoid_cast_fp16)[name = string("x_261_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_617_cast_fp16 = select(a = var_11_to_fp16, b = x_261_cast_fp16, cond = var_328)[name = string("input_617_cast_fp16")];
             tensor<int32, [6]> input_619_pad_0 = const()[name = string("input_619_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_619_mode_0 = const()[name = string("input_619_mode_0"), val = string("constant")];
             fp16 const_127_to_fp16 = const()[name = string("const_127_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_619_cast_fp16 = pad(constant_val = const_127_to_fp16, mode = input_619_mode_0, pad = input_619_pad_0, x = input_617_cast_fp16)[name = string("input_619_cast_fp16")];
+            tensor<fp16, [1, 1024, 136]> input_619_cast_fp16 = pad(constant_val = const_127_to_fp16, mode = input_619_mode_0, pad = input_619_pad_0, x = input_617_cast_fp16)[name = string("input_619_cast_fp16")];
             string input_621_pad_type_0 = const()[name = string("input_621_pad_type_0"), val = string("valid")];
             int32 input_621_groups_0 = const()[name = string("input_621_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_621_strides_0 = const()[name = string("input_621_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_621_pad_0 = const()[name = string("input_621_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_621_dilations_0 = const()[name = string("input_621_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_270_to_fp16 = const()[name = string("const_270_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159022912)))];
-            tensor<fp16, [1024]> const_271_to_fp16 = const()[name = string("const_271_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159041408)))];
-            tensor<fp16, [1, 1024, 188]> input_623_cast_fp16 = conv(bias = const_271_to_fp16, dilations = input_621_dilations_0, groups = input_621_groups_0, pad = input_621_pad_0, pad_type = input_621_pad_type_0, strides = input_621_strides_0, weight = const_270_to_fp16, x = input_619_cast_fp16)[name = string("input_623_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_625_cast_fp16 = silu(x = input_623_cast_fp16)[name = string("input_625_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_270_to_fp16 = const()[name = string("const_270_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152927552)))];
+            tensor<fp16, [1024]> const_271_to_fp16 = const()[name = string("const_271_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152946048)))];
+            tensor<fp16, [1, 1024, 128]> input_623_cast_fp16 = conv(bias = const_271_to_fp16, dilations = input_621_dilations_0, groups = input_621_groups_0, pad = input_621_pad_0, pad_type = input_621_pad_type_0, strides = input_621_strides_0, weight = const_270_to_fp16, x = input_619_cast_fp16)[name = string("input_623_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_625_cast_fp16 = silu(x = input_623_cast_fp16)[name = string("input_625_cast_fp16")];
             string x_263_pad_type_0 = const()[name = string("x_263_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_263_strides_0 = const()[name = string("x_263_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_263_pad_0 = const()[name = string("x_263_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_263_dilations_0 = const()[name = string("x_263_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_263_groups_0 = const()[name = string("x_263_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522671]> module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159043520))), indices_nonzero_data = tensor<uint6, [522671]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159174656))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159566784))))[name = string("module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522671]> module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152948160))), indices_nonzero_data = tensor<uint6, [522671]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153079296))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153471424))))[name = string("module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_263_cast_fp16 = conv(dilations = x_263_dilations_0, groups = x_263_groups_0, pad = x_263_pad_0, pad_type = x_263_pad_type_0, strides = x_263_strides_0, weight = module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_625_cast_fp16)[name = string("x_263_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_263_cast_fp16 = conv(dilations = x_263_dilations_0, groups = x_263_groups_0, pad = x_263_pad_0, pad_type = x_263_pad_type_0, strides = x_263_strides_0, weight = module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_625_cast_fp16)[name = string("x_263_cast_fp16")];
             tensor<int32, [3]> input_627_perm_0 = const()[name = string("input_627_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_627_cast_fp16 = transpose(perm = input_627_perm_0, x = x_263_cast_fp16)[name = string("transpose_229")];
-            tensor<fp16, [1, 188, 1024]> input_629_cast_fp16 = add(x = input_611_cast_fp16, y = input_627_cast_fp16)[name = string("input_629_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_627_cast_fp16 = transpose(perm = input_627_perm_0, x = x_263_cast_fp16)[name = string("transpose_229")];
+            tensor<fp16, [1, 128, 1024]> input_629_cast_fp16 = add(x = input_611_cast_fp16, y = input_627_cast_fp16)[name = string("input_629_cast_fp16")];
             tensor<int32, [1]> input_631_axes_0 = const()[name = string("input_631_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_11_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_11_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159583232)))];
-            tensor<fp16, [1024]> module_layers_11_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_11_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159585344)))];
-            tensor<fp16, [1, 188, 1024]> input_631_cast_fp16 = layer_norm(axes = input_631_axes_0, beta = module_layers_11_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_feed_forward2_weight_to_fp16, x = input_629_cast_fp16)[name = string("input_631_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2065473]> module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159587456))), indices_nonzero_data = tensor<uint6, [2065473]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160111808))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161660992))))[name = string("module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_11_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_11_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153487872)))];
+            tensor<fp16, [1024]> module_layers_11_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_11_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153489984)))];
+            tensor<fp16, [1, 128, 1024]> input_631_cast_fp16 = layer_norm(axes = input_631_axes_0, beta = module_layers_11_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_feed_forward2_weight_to_fp16, x = input_629_cast_fp16)[name = string("input_631_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2065473]> module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153492096))), indices_nonzero_data = tensor<uint6, [2065473]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154016448))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155565632))))[name = string("module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_107_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_631_cast_fp16)[name = string("linear_107_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_635_cast_fp16 = silu(x = linear_107_cast_fp16)[name = string("input_635_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1993236]> module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161726592))), indices_nonzero_data = tensor<uint6, [1993236]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162250944))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163745984))))[name = string("module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_107_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_631_cast_fp16)[name = string("linear_107_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_635_cast_fp16 = silu(x = linear_107_cast_fp16)[name = string("input_635_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1993236]> module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155631232))), indices_nonzero_data = tensor<uint6, [1993236]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156155584))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157650624))))[name = string("module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_108_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_635_cast_fp16)[name = string("linear_108_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_108_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_635_cast_fp16)[name = string("linear_108_cast_fp16")];
             fp16 var_2159_to_fp16 = const()[name = string("op_2159_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_2160_cast_fp16 = mul(x = linear_108_cast_fp16, y = var_2159_to_fp16)[name = string("op_2160_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_641_cast_fp16 = add(x = input_629_cast_fp16, y = var_2160_cast_fp16)[name = string("input_641_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_2160_cast_fp16 = mul(x = linear_108_cast_fp16, y = var_2159_to_fp16)[name = string("op_2160_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_641_cast_fp16 = add(x = input_629_cast_fp16, y = var_2160_cast_fp16)[name = string("input_641_cast_fp16")];
             tensor<int32, [1]> input_643_axes_0 = const()[name = string("input_643_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_11_norm_out_weight_to_fp16 = const()[name = string("module_layers_11_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163762432)))];
-            tensor<fp16, [1024]> module_layers_11_norm_out_bias_to_fp16 = const()[name = string("module_layers_11_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163764544)))];
-            tensor<fp16, [1, 188, 1024]> input_643_cast_fp16 = layer_norm(axes = input_643_axes_0, beta = module_layers_11_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_out_weight_to_fp16, x = input_641_cast_fp16)[name = string("input_643_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_11_norm_out_weight_to_fp16 = const()[name = string("module_layers_11_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157667072)))];
+            tensor<fp16, [1024]> module_layers_11_norm_out_bias_to_fp16 = const()[name = string("module_layers_11_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157669184)))];
+            tensor<fp16, [1, 128, 1024]> input_643_cast_fp16 = layer_norm(axes = input_643_axes_0, beta = module_layers_11_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_out_weight_to_fp16, x = input_641_cast_fp16)[name = string("input_643_cast_fp16")];
             tensor<int32, [1]> input_645_axes_0 = const()[name = string("input_645_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_12_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_12_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163766656)))];
-            tensor<fp16, [1024]> module_layers_12_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_12_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163768768)))];
-            tensor<fp16, [1, 188, 1024]> input_645_cast_fp16 = layer_norm(axes = input_645_axes_0, beta = module_layers_12_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_feed_forward1_weight_to_fp16, x = input_643_cast_fp16)[name = string("input_645_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2047312]> module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163770880))), indices_nonzero_data = tensor<uint6, [2047312]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164295232))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165830784))))[name = string("module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_12_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_12_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157671296)))];
+            tensor<fp16, [1024]> module_layers_12_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_12_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157673408)))];
+            tensor<fp16, [1, 128, 1024]> input_645_cast_fp16 = layer_norm(axes = input_645_axes_0, beta = module_layers_12_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_feed_forward1_weight_to_fp16, x = input_643_cast_fp16)[name = string("input_645_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2047312]> module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157675520))), indices_nonzero_data = tensor<uint6, [2047312]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158199872))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159735424))))[name = string("module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_109_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_645_cast_fp16)[name = string("linear_109_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_649_cast_fp16 = silu(x = linear_109_cast_fp16)[name = string("input_649_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1984634]> module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165896384))), indices_nonzero_data = tensor<uint6, [1984634]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166420736))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167909312))))[name = string("module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_109_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_645_cast_fp16)[name = string("linear_109_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_649_cast_fp16 = silu(x = linear_109_cast_fp16)[name = string("input_649_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1984634]> module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159801024))), indices_nonzero_data = tensor<uint6, [1984634]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160325376))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161813952))))[name = string("module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_110_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_649_cast_fp16)[name = string("linear_110_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_110_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_649_cast_fp16)[name = string("linear_110_cast_fp16")];
             fp16 var_2188_to_fp16 = const()[name = string("op_2188_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_2189_cast_fp16 = mul(x = linear_110_cast_fp16, y = var_2188_to_fp16)[name = string("op_2189_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_655_cast_fp16 = add(x = input_643_cast_fp16, y = var_2189_cast_fp16)[name = string("input_655_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_2189_cast_fp16 = mul(x = linear_110_cast_fp16, y = var_2188_to_fp16)[name = string("op_2189_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_655_cast_fp16 = add(x = input_643_cast_fp16, y = var_2189_cast_fp16)[name = string("input_655_cast_fp16")];
             tensor<int32, [1]> query_25_axes_0 = const()[name = string("query_25_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_12_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_12_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167925760)))];
-            tensor<fp16, [1024]> module_layers_12_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_12_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167927872)))];
-            tensor<fp16, [1, 188, 1024]> query_25_cast_fp16 = layer_norm(axes = query_25_axes_0, beta = module_layers_12_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_self_att_weight_to_fp16, x = input_655_cast_fp16)[name = string("query_25_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525969]> module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167929984))), indices_nonzero_data = tensor<uint6, [525969]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168061120))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168455680))))[name = string("module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_12_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_12_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161830400)))];
+            tensor<fp16, [1024]> module_layers_12_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_12_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161832512)))];
+            tensor<fp16, [1, 128, 1024]> query_25_cast_fp16 = layer_norm(axes = query_25_axes_0, beta = module_layers_12_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_self_att_weight_to_fp16, x = input_655_cast_fp16)[name = string("query_25_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525969]> module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161834624))), indices_nonzero_data = tensor<uint6, [525969]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161965760))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162360320))))[name = string("module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_111_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_25_cast_fp16)[name = string("linear_111_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_111_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_25_cast_fp16)[name = string("linear_111_cast_fp16")];
             tensor<int32, [4]> var_2205 = const()[name = string("op_2205"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_73_cast_fp16 = reshape(shape = var_2205, x = linear_111_cast_fp16)[name = string("q_73_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522730]> module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168472128))), indices_nonzero_data = tensor<uint6, [522730]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168603264))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168995392))))[name = string("module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> q_73_cast_fp16 = reshape(shape = var_2205, x = linear_111_cast_fp16)[name = string("q_73_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522730]> module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162376768))), indices_nonzero_data = tensor<uint6, [522730]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162507904))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162900032))))[name = string("module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_112_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_25_cast_fp16)[name = string("linear_112_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_112_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_25_cast_fp16)[name = string("linear_112_cast_fp16")];
             tensor<int32, [4]> var_2209 = const()[name = string("op_2209"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_49_cast_fp16 = reshape(shape = var_2209, x = linear_112_cast_fp16)[name = string("k_49_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521113]> module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169011840))), indices_nonzero_data = tensor<uint6, [521113]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169142976))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169533888))))[name = string("module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> k_49_cast_fp16 = reshape(shape = var_2209, x = linear_112_cast_fp16)[name = string("k_49_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521113]> module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162916480))), indices_nonzero_data = tensor<uint6, [521113]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163047616))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163438528))))[name = string("module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_113_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_25_cast_fp16)[name = string("linear_113_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_113_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_25_cast_fp16)[name = string("linear_113_cast_fp16")];
             tensor<int32, [4]> var_2213 = const()[name = string("op_2213"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_25_cast_fp16 = reshape(shape = var_2213, x = linear_113_cast_fp16)[name = string("v_25_cast_fp16")];
+            tensor<fp16, [1, 128, 8, 128]> v_25_cast_fp16 = reshape(shape = var_2213, x = linear_113_cast_fp16)[name = string("v_25_cast_fp16")];
             tensor<int32, [4]> value_27_perm_0 = const()[name = string("value_27_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_12_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_12_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169550336)))];
-            tensor<fp16, [1, 188, 8, 128]> var_2225_cast_fp16 = add(x = q_73_cast_fp16, y = module_layers_12_self_attn_pos_bias_u_to_fp16)[name = string("op_2225_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_12_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_12_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169552448)))];
-            tensor<fp16, [1, 188, 8, 128]> var_2227_cast_fp16 = add(x = q_73_cast_fp16, y = module_layers_12_self_attn_pos_bias_v_to_fp16)[name = string("op_2227_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_12_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_12_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163454976)))];
+            tensor<fp16, [1, 128, 8, 128]> var_2225_cast_fp16 = add(x = q_73_cast_fp16, y = module_layers_12_self_attn_pos_bias_u_to_fp16)[name = string("op_2225_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_12_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_12_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163457088)))];
+            tensor<fp16, [1, 128, 8, 128]> var_2227_cast_fp16 = add(x = q_73_cast_fp16, y = module_layers_12_self_attn_pos_bias_v_to_fp16)[name = string("op_2227_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_25_perm_0 = const()[name = string("q_with_bias_v_25_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_271_transpose_x_0 = const()[name = string("x_271_transpose_x_0"), val = bool(false)];
             bool x_271_transpose_y_0 = const()[name = string("x_271_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_2229_to_fp16 = const()[name = string("op_2229_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169554560)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_25_cast_fp16 = transpose(perm = q_with_bias_v_25_perm_0, x = var_2227_cast_fp16)[name = string("transpose_228")];
-            tensor<fp16, [1, 8, 188, 375]> x_271_cast_fp16 = matmul(transpose_x = x_271_transpose_x_0, transpose_y = x_271_transpose_y_0, x = q_with_bias_v_25_cast_fp16, y = var_2229_to_fp16)[name = string("x_271_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 255]> var_2229_to_fp16 = const()[name = string("op_2229_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313689472)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_25_cast_fp16 = transpose(perm = q_with_bias_v_25_perm_0, x = var_2227_cast_fp16)[name = string("transpose_228")];
+            tensor<fp16, [1, 8, 128, 255]> x_271_cast_fp16 = matmul(transpose_x = x_271_transpose_x_0, transpose_y = x_271_transpose_y_0, x = q_with_bias_v_25_cast_fp16, y = var_2229_to_fp16)[name = string("x_271_cast_fp16")];
             tensor<int32, [8]> x_273_pad_0 = const()[name = string("x_273_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_273_mode_0 = const()[name = string("x_273_mode_0"), val = string("constant")];
             fp16 const_134_to_fp16 = const()[name = string("const_134_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_273_cast_fp16 = pad(constant_val = const_134_to_fp16, mode = x_273_mode_0, pad = x_273_pad_0, x = x_271_cast_fp16)[name = string("x_273_cast_fp16")];
-            tensor<int32, [4]> var_2237 = const()[name = string("op_2237"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_275_cast_fp16 = reshape(shape = var_2237, x = x_273_cast_fp16)[name = string("x_275_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 256]> x_273_cast_fp16 = pad(constant_val = const_134_to_fp16, mode = x_273_mode_0, pad = x_273_pad_0, x = x_271_cast_fp16)[name = string("x_273_cast_fp16")];
+            tensor<int32, [4]> var_2237 = const()[name = string("op_2237"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_275_cast_fp16 = reshape(shape = var_2237, x = x_273_cast_fp16)[name = string("x_275_cast_fp16")];
             tensor<int32, [4]> var_2241_begin_0 = const()[name = string("op_2241_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_2241_end_0 = const()[name = string("op_2241_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_2241_end_0 = const()[name = string("op_2241_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
             tensor<bool, [4]> var_2241_end_mask_0 = const()[name = string("op_2241_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_2241_cast_fp16 = slice_by_index(begin = var_2241_begin_0, end = var_2241_end_0, end_mask = var_2241_end_mask_0, x = x_275_cast_fp16)[name = string("op_2241_cast_fp16")];
-            tensor<int32, [4]> var_2242 = const()[name = string("op_2242"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_49_cast_fp16 = reshape(shape = var_2242, x = var_2241_cast_fp16)[name = string("matrix_bd_49_cast_fp16")];
+            tensor<fp16, [1, 8, 255, 128]> var_2241_cast_fp16 = slice_by_index(begin = var_2241_begin_0, end = var_2241_end_0, end_mask = var_2241_end_mask_0, x = x_275_cast_fp16)[name = string("op_2241_cast_fp16")];
+            tensor<int32, [4]> var_2242 = const()[name = string("op_2242"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_49_cast_fp16 = reshape(shape = var_2242, x = var_2241_cast_fp16)[name = string("matrix_bd_49_cast_fp16")];
             bool matrix_ac_25_transpose_x_0 = const()[name = string("matrix_ac_25_transpose_x_0"), val = bool(false)];
             bool matrix_ac_25_transpose_y_0 = const()[name = string("matrix_ac_25_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_120_perm_0 = const()[name = string("transpose_120_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_121_perm_0 = const()[name = string("transpose_121_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_121 = transpose(perm = transpose_121_perm_0, x = k_49_cast_fp16)[name = string("transpose_226")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_120 = transpose(perm = transpose_120_perm_0, x = var_2225_cast_fp16)[name = string("transpose_227")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_25_cast_fp16 = matmul(transpose_x = matrix_ac_25_transpose_x_0, transpose_y = matrix_ac_25_transpose_y_0, x = transpose_120, y = transpose_121)[name = string("matrix_ac_25_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_121 = transpose(perm = transpose_121_perm_0, x = k_49_cast_fp16)[name = string("transpose_226")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_120 = transpose(perm = transpose_120_perm_0, x = var_2225_cast_fp16)[name = string("transpose_227")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_25_cast_fp16 = matmul(transpose_x = matrix_ac_25_transpose_x_0, transpose_y = matrix_ac_25_transpose_y_0, x = transpose_120, y = transpose_121)[name = string("matrix_ac_25_cast_fp16")];
             tensor<int32, [4]> matrix_bd_51_begin_0 = const()[name = string("matrix_bd_51_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_51_end_0 = const()[name = string("matrix_bd_51_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_51_end_0 = const()[name = string("matrix_bd_51_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
             tensor<bool, [4]> matrix_bd_51_end_mask_0 = const()[name = string("matrix_bd_51_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_51_cast_fp16 = slice_by_index(begin = matrix_bd_51_begin_0, end = matrix_bd_51_end_0, end_mask = matrix_bd_51_end_mask_0, x = matrix_bd_49_cast_fp16)[name = string("matrix_bd_51_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_2251_cast_fp16 = add(x = matrix_ac_25_cast_fp16, y = matrix_bd_51_cast_fp16)[name = string("op_2251_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_51_cast_fp16 = slice_by_index(begin = matrix_bd_51_begin_0, end = matrix_bd_51_end_0, end_mask = matrix_bd_51_end_mask_0, x = matrix_bd_49_cast_fp16)[name = string("matrix_bd_51_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_2251_cast_fp16 = add(x = matrix_ac_25_cast_fp16, y = matrix_bd_51_cast_fp16)[name = string("op_2251_cast_fp16")];
             fp16 _inversed_scores_49_y_0_to_fp16 = const()[name = string("_inversed_scores_49_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_49_cast_fp16 = mul(x = var_2251_cast_fp16, y = _inversed_scores_49_y_0_to_fp16)[name = string("_inversed_scores_49_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_51_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_49_cast_fp16, cond = mask_3)[name = string("scores_51_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_2257_cast_fp16 = softmax(axis = var_30, x = scores_51_cast_fp16)[name = string("op_2257_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_657_cast_fp16 = select(a = var_11_to_fp16, b = var_2257_cast_fp16, cond = mask_3)[name = string("input_657_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_49_cast_fp16 = mul(x = var_2251_cast_fp16, y = _inversed_scores_49_y_0_to_fp16)[name = string("_inversed_scores_49_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_51_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_49_cast_fp16, cond = mask_3)[name = string("scores_51_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_2257_cast_fp16 = softmax(axis = var_30, x = scores_51_cast_fp16)[name = string("op_2257_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_657_cast_fp16 = select(a = var_11_to_fp16, b = var_2257_cast_fp16, cond = mask_3)[name = string("input_657_cast_fp16")];
             bool x_277_transpose_x_0 = const()[name = string("x_277_transpose_x_0"), val = bool(false)];
             bool x_277_transpose_y_0 = const()[name = string("x_277_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_27_cast_fp16 = transpose(perm = value_27_perm_0, x = v_25_cast_fp16)[name = string("transpose_225")];
-            tensor<fp16, [1, 8, 188, 128]> x_277_cast_fp16 = matmul(transpose_x = x_277_transpose_x_0, transpose_y = x_277_transpose_y_0, x = input_657_cast_fp16, y = value_27_cast_fp16)[name = string("x_277_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> value_27_cast_fp16 = transpose(perm = value_27_perm_0, x = v_25_cast_fp16)[name = string("transpose_225")];
+            tensor<fp16, [1, 8, 128, 128]> x_277_cast_fp16 = matmul(transpose_x = x_277_transpose_x_0, transpose_y = x_277_transpose_y_0, x = input_657_cast_fp16, y = value_27_cast_fp16)[name = string("x_277_cast_fp16")];
             tensor<int32, [4]> var_2261_perm_0 = const()[name = string("op_2261_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_2262 = const()[name = string("op_2262"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_2261_cast_fp16 = transpose(perm = var_2261_perm_0, x = x_277_cast_fp16)[name = string("transpose_224")];
-            tensor<fp16, [1, 188, 1024]> input_659_cast_fp16 = reshape(shape = var_2262, x = var_2261_cast_fp16)[name = string("input_659_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [527263]> module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170322624))), indices_nonzero_data = tensor<uint6, [527263]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170453760))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170849280))))[name = string("module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> var_2261_cast_fp16 = transpose(perm = var_2261_perm_0, x = x_277_cast_fp16)[name = string("transpose_224")];
+            tensor<fp16, [1, 128, 1024]> input_659_cast_fp16 = reshape(shape = var_2262, x = var_2261_cast_fp16)[name = string("input_659_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [527263]> module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163719360))), indices_nonzero_data = tensor<uint6, [527263]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163850496))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164246016))))[name = string("module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_115_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_659_cast_fp16)[name = string("linear_115_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_663_cast_fp16 = add(x = input_655_cast_fp16, y = linear_115_cast_fp16)[name = string("input_663_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_115_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_659_cast_fp16)[name = string("linear_115_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_663_cast_fp16 = add(x = input_655_cast_fp16, y = linear_115_cast_fp16)[name = string("input_663_cast_fp16")];
             tensor<int32, [1]> x_281_axes_0 = const()[name = string("x_281_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_12_norm_conv_weight_to_fp16 = const()[name = string("module_layers_12_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170865728)))];
-            tensor<fp16, [1024]> module_layers_12_norm_conv_bias_to_fp16 = const()[name = string("module_layers_12_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170867840)))];
-            tensor<fp16, [1, 188, 1024]> x_281_cast_fp16 = layer_norm(axes = x_281_axes_0, beta = module_layers_12_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_conv_weight_to_fp16, x = input_663_cast_fp16)[name = string("x_281_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_12_norm_conv_weight_to_fp16 = const()[name = string("module_layers_12_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164262464)))];
+            tensor<fp16, [1024]> module_layers_12_norm_conv_bias_to_fp16 = const()[name = string("module_layers_12_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164264576)))];
+            tensor<fp16, [1, 128, 1024]> x_281_cast_fp16 = layer_norm(axes = x_281_axes_0, beta = module_layers_12_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_conv_weight_to_fp16, x = input_663_cast_fp16)[name = string("x_281_cast_fp16")];
             tensor<int32, [3]> input_665_perm_0 = const()[name = string("input_665_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_667_pad_type_0 = const()[name = string("input_667_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_667_strides_0 = const()[name = string("input_667_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_667_pad_0 = const()[name = string("input_667_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_667_dilations_0 = const()[name = string("input_667_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_667_groups_0 = const()[name = string("input_667_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1044574]> module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170869952))), indices_nonzero_data = tensor<uint6, [1044574]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171132160))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171915712))))[name = string("module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1044574]> module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164266688))), indices_nonzero_data = tensor<uint6, [1044574]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164528896))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165312448))))[name = string("module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_665_cast_fp16 = transpose(perm = input_665_perm_0, x = x_281_cast_fp16)[name = string("transpose_223")];
-            tensor<fp16, [1, 2048, 188]> input_667_cast_fp16 = conv(dilations = input_667_dilations_0, groups = input_667_groups_0, pad = input_667_pad_0, pad_type = input_667_pad_type_0, strides = input_667_strides_0, weight = module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_665_cast_fp16)[name = string("input_667_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_665_cast_fp16 = transpose(perm = input_665_perm_0, x = x_281_cast_fp16)[name = string("transpose_223")];
+            tensor<fp16, [1, 2048, 128]> input_667_cast_fp16 = conv(dilations = input_667_dilations_0, groups = input_667_groups_0, pad = input_667_pad_0, pad_type = input_667_pad_type_0, strides = input_667_strides_0, weight = module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_665_cast_fp16)[name = string("input_667_cast_fp16")];
             int32 x_283_split_num_splits_0 = const()[name = string("x_283_split_num_splits_0"), val = int32(2)];
             int32 x_283_split_axis_0 = const()[name = string("x_283_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_283_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_283_split_cast_fp16_1 = split(axis = x_283_split_axis_0, num_splits = x_283_split_num_splits_0, x = input_667_cast_fp16)[name = string("x_283_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_283_split_1_sigmoid_cast_fp16 = sigmoid(x = x_283_split_cast_fp16_1)[name = string("x_283_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_283_cast_fp16 = mul(x = x_283_split_cast_fp16_0, y = x_283_split_1_sigmoid_cast_fp16)[name = string("x_283_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_669_cast_fp16 = select(a = var_11_to_fp16, b = x_283_cast_fp16, cond = var_328)[name = string("input_669_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_283_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_283_split_cast_fp16_1 = split(axis = x_283_split_axis_0, num_splits = x_283_split_num_splits_0, x = input_667_cast_fp16)[name = string("x_283_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_283_split_1_sigmoid_cast_fp16 = sigmoid(x = x_283_split_cast_fp16_1)[name = string("x_283_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_283_cast_fp16 = mul(x = x_283_split_cast_fp16_0, y = x_283_split_1_sigmoid_cast_fp16)[name = string("x_283_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_669_cast_fp16 = select(a = var_11_to_fp16, b = x_283_cast_fp16, cond = var_328)[name = string("input_669_cast_fp16")];
             tensor<int32, [6]> input_671_pad_0 = const()[name = string("input_671_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_671_mode_0 = const()[name = string("input_671_mode_0"), val = string("constant")];
             fp16 const_137_to_fp16 = const()[name = string("const_137_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_671_cast_fp16 = pad(constant_val = const_137_to_fp16, mode = input_671_mode_0, pad = input_671_pad_0, x = input_669_cast_fp16)[name = string("input_671_cast_fp16")];
+            tensor<fp16, [1, 1024, 136]> input_671_cast_fp16 = pad(constant_val = const_137_to_fp16, mode = input_671_mode_0, pad = input_671_pad_0, x = input_669_cast_fp16)[name = string("input_671_cast_fp16")];
             string input_673_pad_type_0 = const()[name = string("input_673_pad_type_0"), val = string("valid")];
             int32 input_673_groups_0 = const()[name = string("input_673_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_673_strides_0 = const()[name = string("input_673_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_673_pad_0 = const()[name = string("input_673_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_673_dilations_0 = const()[name = string("input_673_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_272_to_fp16 = const()[name = string("const_272_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171948544)))];
-            tensor<fp16, [1024]> const_273_to_fp16 = const()[name = string("const_273_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171967040)))];
-            tensor<fp16, [1, 1024, 188]> input_675_cast_fp16 = conv(bias = const_273_to_fp16, dilations = input_673_dilations_0, groups = input_673_groups_0, pad = input_673_pad_0, pad_type = input_673_pad_type_0, strides = input_673_strides_0, weight = const_272_to_fp16, x = input_671_cast_fp16)[name = string("input_675_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_677_cast_fp16 = silu(x = input_675_cast_fp16)[name = string("input_677_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_272_to_fp16 = const()[name = string("const_272_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165345280)))];
+            tensor<fp16, [1024]> const_273_to_fp16 = const()[name = string("const_273_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165363776)))];
+            tensor<fp16, [1, 1024, 128]> input_675_cast_fp16 = conv(bias = const_273_to_fp16, dilations = input_673_dilations_0, groups = input_673_groups_0, pad = input_673_pad_0, pad_type = input_673_pad_type_0, strides = input_673_strides_0, weight = const_272_to_fp16, x = input_671_cast_fp16)[name = string("input_675_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_677_cast_fp16 = silu(x = input_675_cast_fp16)[name = string("input_677_cast_fp16")];
             string x_285_pad_type_0 = const()[name = string("x_285_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_285_strides_0 = const()[name = string("x_285_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_285_pad_0 = const()[name = string("x_285_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_285_dilations_0 = const()[name = string("x_285_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_285_groups_0 = const()[name = string("x_285_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522908]> module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171969152))), indices_nonzero_data = tensor<uint6, [522908]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172100288))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172492544))))[name = string("module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522908]> module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165365888))), indices_nonzero_data = tensor<uint6, [522908]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165497024))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165889280))))[name = string("module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_285_cast_fp16 = conv(dilations = x_285_dilations_0, groups = x_285_groups_0, pad = x_285_pad_0, pad_type = x_285_pad_type_0, strides = x_285_strides_0, weight = module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_677_cast_fp16)[name = string("x_285_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_285_cast_fp16 = conv(dilations = x_285_dilations_0, groups = x_285_groups_0, pad = x_285_pad_0, pad_type = x_285_pad_type_0, strides = x_285_strides_0, weight = module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_677_cast_fp16)[name = string("x_285_cast_fp16")];
             tensor<int32, [3]> input_679_perm_0 = const()[name = string("input_679_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_679_cast_fp16 = transpose(perm = input_679_perm_0, x = x_285_cast_fp16)[name = string("transpose_222")];
-            tensor<fp16, [1, 188, 1024]> input_681_cast_fp16 = add(x = input_663_cast_fp16, y = input_679_cast_fp16)[name = string("input_681_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_679_cast_fp16 = transpose(perm = input_679_perm_0, x = x_285_cast_fp16)[name = string("transpose_222")];
+            tensor<fp16, [1, 128, 1024]> input_681_cast_fp16 = add(x = input_663_cast_fp16, y = input_679_cast_fp16)[name = string("input_681_cast_fp16")];
             tensor<int32, [1]> input_683_axes_0 = const()[name = string("input_683_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_12_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_12_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172508992)))];
-            tensor<fp16, [1024]> module_layers_12_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_12_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172511104)))];
-            tensor<fp16, [1, 188, 1024]> input_683_cast_fp16 = layer_norm(axes = input_683_axes_0, beta = module_layers_12_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_feed_forward2_weight_to_fp16, x = input_681_cast_fp16)[name = string("input_683_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2072965]> module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172513216))), indices_nonzero_data = tensor<uint6, [2072965]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173037568))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174592384))))[name = string("module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_12_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_12_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165905728)))];
+            tensor<fp16, [1024]> module_layers_12_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_12_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165907840)))];
+            tensor<fp16, [1, 128, 1024]> input_683_cast_fp16 = layer_norm(axes = input_683_axes_0, beta = module_layers_12_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_feed_forward2_weight_to_fp16, x = input_681_cast_fp16)[name = string("input_683_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2072965]> module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165909952))), indices_nonzero_data = tensor<uint6, [2072965]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166434304))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167989120))))[name = string("module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_116_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_683_cast_fp16)[name = string("linear_116_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_687_cast_fp16 = silu(x = linear_116_cast_fp16)[name = string("input_687_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1985707]> module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174657984))), indices_nonzero_data = tensor<uint6, [1985707]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175182336))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176671744))))[name = string("module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_116_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_683_cast_fp16)[name = string("linear_116_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_687_cast_fp16 = silu(x = linear_116_cast_fp16)[name = string("input_687_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1985707]> module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168054720))), indices_nonzero_data = tensor<uint6, [1985707]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168579072))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170068480))))[name = string("module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_117_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_687_cast_fp16)[name = string("linear_117_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_117_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_687_cast_fp16)[name = string("linear_117_cast_fp16")];
             fp16 var_2322_to_fp16 = const()[name = string("op_2322_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_2323_cast_fp16 = mul(x = linear_117_cast_fp16, y = var_2322_to_fp16)[name = string("op_2323_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_693_cast_fp16 = add(x = input_681_cast_fp16, y = var_2323_cast_fp16)[name = string("input_693_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_2323_cast_fp16 = mul(x = linear_117_cast_fp16, y = var_2322_to_fp16)[name = string("op_2323_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_693_cast_fp16 = add(x = input_681_cast_fp16, y = var_2323_cast_fp16)[name = string("input_693_cast_fp16")];
             tensor<int32, [1]> input_695_axes_0 = const()[name = string("input_695_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_12_norm_out_weight_to_fp16 = const()[name = string("module_layers_12_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176688192)))];
-            tensor<fp16, [1024]> module_layers_12_norm_out_bias_to_fp16 = const()[name = string("module_layers_12_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176690304)))];
-            tensor<fp16, [1, 188, 1024]> input_695_cast_fp16 = layer_norm(axes = input_695_axes_0, beta = module_layers_12_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_out_weight_to_fp16, x = input_693_cast_fp16)[name = string("input_695_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_12_norm_out_weight_to_fp16 = const()[name = string("module_layers_12_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170084928)))];
+            tensor<fp16, [1024]> module_layers_12_norm_out_bias_to_fp16 = const()[name = string("module_layers_12_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170087040)))];
+            tensor<fp16, [1, 128, 1024]> input_695_cast_fp16 = layer_norm(axes = input_695_axes_0, beta = module_layers_12_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_out_weight_to_fp16, x = input_693_cast_fp16)[name = string("input_695_cast_fp16")];
             tensor<int32, [1]> input_697_axes_0 = const()[name = string("input_697_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_13_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_13_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176692416)))];
-            tensor<fp16, [1024]> module_layers_13_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_13_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176694528)))];
-            tensor<fp16, [1, 188, 1024]> input_697_cast_fp16 = layer_norm(axes = input_697_axes_0, beta = module_layers_13_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_feed_forward1_weight_to_fp16, x = input_695_cast_fp16)[name = string("input_697_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2052653]> module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176696640))), indices_nonzero_data = tensor<uint6, [2052653]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177220992))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178760576))))[name = string("module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_13_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_13_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170089152)))];
+            tensor<fp16, [1024]> module_layers_13_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_13_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170091264)))];
+            tensor<fp16, [1, 128, 1024]> input_697_cast_fp16 = layer_norm(axes = input_697_axes_0, beta = module_layers_13_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_feed_forward1_weight_to_fp16, x = input_695_cast_fp16)[name = string("input_697_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2052653]> module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170093376))), indices_nonzero_data = tensor<uint6, [2052653]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170617728))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172157312))))[name = string("module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_118_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_697_cast_fp16)[name = string("linear_118_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_701_cast_fp16 = silu(x = linear_118_cast_fp16)[name = string("input_701_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1946459]> module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178826176))), indices_nonzero_data = tensor<uint6, [1946459]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179350528))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180810496))))[name = string("module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_118_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_697_cast_fp16)[name = string("linear_118_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_701_cast_fp16 = silu(x = linear_118_cast_fp16)[name = string("input_701_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1946459]> module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172222912))), indices_nonzero_data = tensor<uint6, [1946459]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172747264))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174207232))))[name = string("module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_119_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_701_cast_fp16)[name = string("linear_119_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_119_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_701_cast_fp16)[name = string("linear_119_cast_fp16")];
             fp16 var_2351_to_fp16 = const()[name = string("op_2351_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_2352_cast_fp16 = mul(x = linear_119_cast_fp16, y = var_2351_to_fp16)[name = string("op_2352_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_707_cast_fp16 = add(x = input_695_cast_fp16, y = var_2352_cast_fp16)[name = string("input_707_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_2352_cast_fp16 = mul(x = linear_119_cast_fp16, y = var_2351_to_fp16)[name = string("op_2352_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_707_cast_fp16 = add(x = input_695_cast_fp16, y = var_2352_cast_fp16)[name = string("input_707_cast_fp16")];
             tensor<int32, [1]> query_27_axes_0 = const()[name = string("query_27_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_13_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_13_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180826944)))];
-            tensor<fp16, [1024]> module_layers_13_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_13_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180829056)))];
-            tensor<fp16, [1, 188, 1024]> query_27_cast_fp16 = layer_norm(axes = query_27_axes_0, beta = module_layers_13_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_self_att_weight_to_fp16, x = input_707_cast_fp16)[name = string("query_27_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524122]> module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180831168))), indices_nonzero_data = tensor<uint6, [524122]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180962304))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181355520))))[name = string("module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_13_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_13_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174223680)))];
+            tensor<fp16, [1024]> module_layers_13_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_13_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174225792)))];
+            tensor<fp16, [1, 128, 1024]> query_27_cast_fp16 = layer_norm(axes = query_27_axes_0, beta = module_layers_13_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_self_att_weight_to_fp16, x = input_707_cast_fp16)[name = string("query_27_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524122]> module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174227904))), indices_nonzero_data = tensor<uint6, [524122]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174359040))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174752256))))[name = string("module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_120_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_27_cast_fp16)[name = string("linear_120_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_120_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_27_cast_fp16)[name = string("linear_120_cast_fp16")];
             tensor<int32, [4]> var_2368 = const()[name = string("op_2368"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_79_cast_fp16 = reshape(shape = var_2368, x = linear_120_cast_fp16)[name = string("q_79_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521808]> module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181371968))), indices_nonzero_data = tensor<uint6, [521808]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181503104))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181894528))))[name = string("module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> q_79_cast_fp16 = reshape(shape = var_2368, x = linear_120_cast_fp16)[name = string("q_79_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521808]> module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174768704))), indices_nonzero_data = tensor<uint6, [521808]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174899840))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175291264))))[name = string("module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_121_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_27_cast_fp16)[name = string("linear_121_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_121_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_27_cast_fp16)[name = string("linear_121_cast_fp16")];
             tensor<int32, [4]> var_2372 = const()[name = string("op_2372"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_53_cast_fp16 = reshape(shape = var_2372, x = linear_121_cast_fp16)[name = string("k_53_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521835]> module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181910976))), indices_nonzero_data = tensor<uint6, [521835]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182042112))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182433600))))[name = string("module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> k_53_cast_fp16 = reshape(shape = var_2372, x = linear_121_cast_fp16)[name = string("k_53_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521835]> module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175307712))), indices_nonzero_data = tensor<uint6, [521835]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175438848))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175830336))))[name = string("module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_122_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_27_cast_fp16)[name = string("linear_122_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_122_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_27_cast_fp16)[name = string("linear_122_cast_fp16")];
             tensor<int32, [4]> var_2376 = const()[name = string("op_2376"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_27_cast_fp16 = reshape(shape = var_2376, x = linear_122_cast_fp16)[name = string("v_27_cast_fp16")];
+            tensor<fp16, [1, 128, 8, 128]> v_27_cast_fp16 = reshape(shape = var_2376, x = linear_122_cast_fp16)[name = string("v_27_cast_fp16")];
             tensor<int32, [4]> value_29_perm_0 = const()[name = string("value_29_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_13_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_13_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182450048)))];
-            tensor<fp16, [1, 188, 8, 128]> var_2388_cast_fp16 = add(x = q_79_cast_fp16, y = module_layers_13_self_attn_pos_bias_u_to_fp16)[name = string("op_2388_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_13_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_13_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182452160)))];
-            tensor<fp16, [1, 188, 8, 128]> var_2390_cast_fp16 = add(x = q_79_cast_fp16, y = module_layers_13_self_attn_pos_bias_v_to_fp16)[name = string("op_2390_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_13_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_13_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175846784)))];
+            tensor<fp16, [1, 128, 8, 128]> var_2388_cast_fp16 = add(x = q_79_cast_fp16, y = module_layers_13_self_attn_pos_bias_u_to_fp16)[name = string("op_2388_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_13_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_13_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175848896)))];
+            tensor<fp16, [1, 128, 8, 128]> var_2390_cast_fp16 = add(x = q_79_cast_fp16, y = module_layers_13_self_attn_pos_bias_v_to_fp16)[name = string("op_2390_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_27_perm_0 = const()[name = string("q_with_bias_v_27_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_293_transpose_x_0 = const()[name = string("x_293_transpose_x_0"), val = bool(false)];
             bool x_293_transpose_y_0 = const()[name = string("x_293_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_2392_to_fp16 = const()[name = string("op_2392_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182454272)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_27_cast_fp16 = transpose(perm = q_with_bias_v_27_perm_0, x = var_2390_cast_fp16)[name = string("transpose_221")];
-            tensor<fp16, [1, 8, 188, 375]> x_293_cast_fp16 = matmul(transpose_x = x_293_transpose_x_0, transpose_y = x_293_transpose_y_0, x = q_with_bias_v_27_cast_fp16, y = var_2392_to_fp16)[name = string("x_293_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 255]> var_2392_to_fp16 = const()[name = string("op_2392_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314211776)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_27_cast_fp16 = transpose(perm = q_with_bias_v_27_perm_0, x = var_2390_cast_fp16)[name = string("transpose_221")];
+            tensor<fp16, [1, 8, 128, 255]> x_293_cast_fp16 = matmul(transpose_x = x_293_transpose_x_0, transpose_y = x_293_transpose_y_0, x = q_with_bias_v_27_cast_fp16, y = var_2392_to_fp16)[name = string("x_293_cast_fp16")];
             tensor<int32, [8]> x_295_pad_0 = const()[name = string("x_295_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_295_mode_0 = const()[name = string("x_295_mode_0"), val = string("constant")];
             fp16 const_144_to_fp16 = const()[name = string("const_144_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_295_cast_fp16 = pad(constant_val = const_144_to_fp16, mode = x_295_mode_0, pad = x_295_pad_0, x = x_293_cast_fp16)[name = string("x_295_cast_fp16")];
-            tensor<int32, [4]> var_2400 = const()[name = string("op_2400"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_297_cast_fp16 = reshape(shape = var_2400, x = x_295_cast_fp16)[name = string("x_297_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 256]> x_295_cast_fp16 = pad(constant_val = const_144_to_fp16, mode = x_295_mode_0, pad = x_295_pad_0, x = x_293_cast_fp16)[name = string("x_295_cast_fp16")];
+            tensor<int32, [4]> var_2400 = const()[name = string("op_2400"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_297_cast_fp16 = reshape(shape = var_2400, x = x_295_cast_fp16)[name = string("x_297_cast_fp16")];
             tensor<int32, [4]> var_2404_begin_0 = const()[name = string("op_2404_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_2404_end_0 = const()[name = string("op_2404_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_2404_end_0 = const()[name = string("op_2404_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
             tensor<bool, [4]> var_2404_end_mask_0 = const()[name = string("op_2404_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_2404_cast_fp16 = slice_by_index(begin = var_2404_begin_0, end = var_2404_end_0, end_mask = var_2404_end_mask_0, x = x_297_cast_fp16)[name = string("op_2404_cast_fp16")];
-            tensor<int32, [4]> var_2405 = const()[name = string("op_2405"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_53_cast_fp16 = reshape(shape = var_2405, x = var_2404_cast_fp16)[name = string("matrix_bd_53_cast_fp16")];
+            tensor<fp16, [1, 8, 255, 128]> var_2404_cast_fp16 = slice_by_index(begin = var_2404_begin_0, end = var_2404_end_0, end_mask = var_2404_end_mask_0, x = x_297_cast_fp16)[name = string("op_2404_cast_fp16")];
+            tensor<int32, [4]> var_2405 = const()[name = string("op_2405"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_53_cast_fp16 = reshape(shape = var_2405, x = var_2404_cast_fp16)[name = string("matrix_bd_53_cast_fp16")];
             bool matrix_ac_27_transpose_x_0 = const()[name = string("matrix_ac_27_transpose_x_0"), val = bool(false)];
             bool matrix_ac_27_transpose_y_0 = const()[name = string("matrix_ac_27_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_122_perm_0 = const()[name = string("transpose_122_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_123_perm_0 = const()[name = string("transpose_123_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_123 = transpose(perm = transpose_123_perm_0, x = k_53_cast_fp16)[name = string("transpose_219")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_122 = transpose(perm = transpose_122_perm_0, x = var_2388_cast_fp16)[name = string("transpose_220")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_27_cast_fp16 = matmul(transpose_x = matrix_ac_27_transpose_x_0, transpose_y = matrix_ac_27_transpose_y_0, x = transpose_122, y = transpose_123)[name = string("matrix_ac_27_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_123 = transpose(perm = transpose_123_perm_0, x = k_53_cast_fp16)[name = string("transpose_219")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_122 = transpose(perm = transpose_122_perm_0, x = var_2388_cast_fp16)[name = string("transpose_220")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_27_cast_fp16 = matmul(transpose_x = matrix_ac_27_transpose_x_0, transpose_y = matrix_ac_27_transpose_y_0, x = transpose_122, y = transpose_123)[name = string("matrix_ac_27_cast_fp16")];
             tensor<int32, [4]> matrix_bd_55_begin_0 = const()[name = string("matrix_bd_55_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_55_end_0 = const()[name = string("matrix_bd_55_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_55_end_0 = const()[name = string("matrix_bd_55_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
             tensor<bool, [4]> matrix_bd_55_end_mask_0 = const()[name = string("matrix_bd_55_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_55_cast_fp16 = slice_by_index(begin = matrix_bd_55_begin_0, end = matrix_bd_55_end_0, end_mask = matrix_bd_55_end_mask_0, x = matrix_bd_53_cast_fp16)[name = string("matrix_bd_55_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_2414_cast_fp16 = add(x = matrix_ac_27_cast_fp16, y = matrix_bd_55_cast_fp16)[name = string("op_2414_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_55_cast_fp16 = slice_by_index(begin = matrix_bd_55_begin_0, end = matrix_bd_55_end_0, end_mask = matrix_bd_55_end_mask_0, x = matrix_bd_53_cast_fp16)[name = string("matrix_bd_55_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_2414_cast_fp16 = add(x = matrix_ac_27_cast_fp16, y = matrix_bd_55_cast_fp16)[name = string("op_2414_cast_fp16")];
             fp16 _inversed_scores_53_y_0_to_fp16 = const()[name = string("_inversed_scores_53_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_53_cast_fp16 = mul(x = var_2414_cast_fp16, y = _inversed_scores_53_y_0_to_fp16)[name = string("_inversed_scores_53_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_55_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_53_cast_fp16, cond = mask_3)[name = string("scores_55_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_2420_cast_fp16 = softmax(axis = var_30, x = scores_55_cast_fp16)[name = string("op_2420_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_709_cast_fp16 = select(a = var_11_to_fp16, b = var_2420_cast_fp16, cond = mask_3)[name = string("input_709_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_53_cast_fp16 = mul(x = var_2414_cast_fp16, y = _inversed_scores_53_y_0_to_fp16)[name = string("_inversed_scores_53_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_55_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_53_cast_fp16, cond = mask_3)[name = string("scores_55_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_2420_cast_fp16 = softmax(axis = var_30, x = scores_55_cast_fp16)[name = string("op_2420_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_709_cast_fp16 = select(a = var_11_to_fp16, b = var_2420_cast_fp16, cond = mask_3)[name = string("input_709_cast_fp16")];
             bool x_299_transpose_x_0 = const()[name = string("x_299_transpose_x_0"), val = bool(false)];
             bool x_299_transpose_y_0 = const()[name = string("x_299_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_29_cast_fp16 = transpose(perm = value_29_perm_0, x = v_27_cast_fp16)[name = string("transpose_218")];
-            tensor<fp16, [1, 8, 188, 128]> x_299_cast_fp16 = matmul(transpose_x = x_299_transpose_x_0, transpose_y = x_299_transpose_y_0, x = input_709_cast_fp16, y = value_29_cast_fp16)[name = string("x_299_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> value_29_cast_fp16 = transpose(perm = value_29_perm_0, x = v_27_cast_fp16)[name = string("transpose_218")];
+            tensor<fp16, [1, 8, 128, 128]> x_299_cast_fp16 = matmul(transpose_x = x_299_transpose_x_0, transpose_y = x_299_transpose_y_0, x = input_709_cast_fp16, y = value_29_cast_fp16)[name = string("x_299_cast_fp16")];
             tensor<int32, [4]> var_2424_perm_0 = const()[name = string("op_2424_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_2425 = const()[name = string("op_2425"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_2424_cast_fp16 = transpose(perm = var_2424_perm_0, x = x_299_cast_fp16)[name = string("transpose_217")];
-            tensor<fp16, [1, 188, 1024]> input_711_cast_fp16 = reshape(shape = var_2425, x = var_2424_cast_fp16)[name = string("input_711_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [528285]> module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183222336))), indices_nonzero_data = tensor<uint6, [528285]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183353472))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183749760))))[name = string("module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> var_2424_cast_fp16 = transpose(perm = var_2424_perm_0, x = x_299_cast_fp16)[name = string("transpose_217")];
+            tensor<fp16, [1, 128, 1024]> input_711_cast_fp16 = reshape(shape = var_2425, x = var_2424_cast_fp16)[name = string("input_711_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [528285]> module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176111168))), indices_nonzero_data = tensor<uint6, [528285]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176242304))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176638592))))[name = string("module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_124_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_711_cast_fp16)[name = string("linear_124_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_715_cast_fp16 = add(x = input_707_cast_fp16, y = linear_124_cast_fp16)[name = string("input_715_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_124_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_711_cast_fp16)[name = string("linear_124_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_715_cast_fp16 = add(x = input_707_cast_fp16, y = linear_124_cast_fp16)[name = string("input_715_cast_fp16")];
             tensor<int32, [1]> x_303_axes_0 = const()[name = string("x_303_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_13_norm_conv_weight_to_fp16 = const()[name = string("module_layers_13_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183766208)))];
-            tensor<fp16, [1024]> module_layers_13_norm_conv_bias_to_fp16 = const()[name = string("module_layers_13_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183768320)))];
-            tensor<fp16, [1, 188, 1024]> x_303_cast_fp16 = layer_norm(axes = x_303_axes_0, beta = module_layers_13_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_conv_weight_to_fp16, x = input_715_cast_fp16)[name = string("x_303_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_13_norm_conv_weight_to_fp16 = const()[name = string("module_layers_13_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176655040)))];
+            tensor<fp16, [1024]> module_layers_13_norm_conv_bias_to_fp16 = const()[name = string("module_layers_13_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176657152)))];
+            tensor<fp16, [1, 128, 1024]> x_303_cast_fp16 = layer_norm(axes = x_303_axes_0, beta = module_layers_13_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_conv_weight_to_fp16, x = input_715_cast_fp16)[name = string("x_303_cast_fp16")];
             tensor<int32, [3]> input_717_perm_0 = const()[name = string("input_717_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_719_pad_type_0 = const()[name = string("input_719_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_719_strides_0 = const()[name = string("input_719_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_719_pad_0 = const()[name = string("input_719_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_719_dilations_0 = const()[name = string("input_719_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_719_groups_0 = const()[name = string("input_719_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1045290]> module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183770432))), indices_nonzero_data = tensor<uint6, [1045290]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184032640))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184816704))))[name = string("module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1045290]> module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176659264))), indices_nonzero_data = tensor<uint6, [1045290]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176921472))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177705536))))[name = string("module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_717_cast_fp16 = transpose(perm = input_717_perm_0, x = x_303_cast_fp16)[name = string("transpose_216")];
-            tensor<fp16, [1, 2048, 188]> input_719_cast_fp16 = conv(dilations = input_719_dilations_0, groups = input_719_groups_0, pad = input_719_pad_0, pad_type = input_719_pad_type_0, strides = input_719_strides_0, weight = module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_717_cast_fp16)[name = string("input_719_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_717_cast_fp16 = transpose(perm = input_717_perm_0, x = x_303_cast_fp16)[name = string("transpose_216")];
+            tensor<fp16, [1, 2048, 128]> input_719_cast_fp16 = conv(dilations = input_719_dilations_0, groups = input_719_groups_0, pad = input_719_pad_0, pad_type = input_719_pad_type_0, strides = input_719_strides_0, weight = module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_717_cast_fp16)[name = string("input_719_cast_fp16")];
             int32 x_305_split_num_splits_0 = const()[name = string("x_305_split_num_splits_0"), val = int32(2)];
             int32 x_305_split_axis_0 = const()[name = string("x_305_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_305_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_305_split_cast_fp16_1 = split(axis = x_305_split_axis_0, num_splits = x_305_split_num_splits_0, x = input_719_cast_fp16)[name = string("x_305_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_305_split_1_sigmoid_cast_fp16 = sigmoid(x = x_305_split_cast_fp16_1)[name = string("x_305_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_305_cast_fp16 = mul(x = x_305_split_cast_fp16_0, y = x_305_split_1_sigmoid_cast_fp16)[name = string("x_305_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_721_cast_fp16 = select(a = var_11_to_fp16, b = x_305_cast_fp16, cond = var_328)[name = string("input_721_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_305_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_305_split_cast_fp16_1 = split(axis = x_305_split_axis_0, num_splits = x_305_split_num_splits_0, x = input_719_cast_fp16)[name = string("x_305_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_305_split_1_sigmoid_cast_fp16 = sigmoid(x = x_305_split_cast_fp16_1)[name = string("x_305_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_305_cast_fp16 = mul(x = x_305_split_cast_fp16_0, y = x_305_split_1_sigmoid_cast_fp16)[name = string("x_305_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_721_cast_fp16 = select(a = var_11_to_fp16, b = x_305_cast_fp16, cond = var_328)[name = string("input_721_cast_fp16")];
             tensor<int32, [6]> input_723_pad_0 = const()[name = string("input_723_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_723_mode_0 = const()[name = string("input_723_mode_0"), val = string("constant")];
             fp16 const_147_to_fp16 = const()[name = string("const_147_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_723_cast_fp16 = pad(constant_val = const_147_to_fp16, mode = input_723_mode_0, pad = input_723_pad_0, x = input_721_cast_fp16)[name = string("input_723_cast_fp16")];
+            tensor<fp16, [1, 1024, 136]> input_723_cast_fp16 = pad(constant_val = const_147_to_fp16, mode = input_723_mode_0, pad = input_723_pad_0, x = input_721_cast_fp16)[name = string("input_723_cast_fp16")];
             string input_725_pad_type_0 = const()[name = string("input_725_pad_type_0"), val = string("valid")];
             int32 input_725_groups_0 = const()[name = string("input_725_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_725_strides_0 = const()[name = string("input_725_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_725_pad_0 = const()[name = string("input_725_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_725_dilations_0 = const()[name = string("input_725_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_274_to_fp16 = const()[name = string("const_274_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184849536)))];
-            tensor<fp16, [1024]> const_275_to_fp16 = const()[name = string("const_275_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184868032)))];
-            tensor<fp16, [1, 1024, 188]> input_727_cast_fp16 = conv(bias = const_275_to_fp16, dilations = input_725_dilations_0, groups = input_725_groups_0, pad = input_725_pad_0, pad_type = input_725_pad_type_0, strides = input_725_strides_0, weight = const_274_to_fp16, x = input_723_cast_fp16)[name = string("input_727_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_729_cast_fp16 = silu(x = input_727_cast_fp16)[name = string("input_729_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_274_to_fp16 = const()[name = string("const_274_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177738368)))];
+            tensor<fp16, [1024]> const_275_to_fp16 = const()[name = string("const_275_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177756864)))];
+            tensor<fp16, [1, 1024, 128]> input_727_cast_fp16 = conv(bias = const_275_to_fp16, dilations = input_725_dilations_0, groups = input_725_groups_0, pad = input_725_pad_0, pad_type = input_725_pad_type_0, strides = input_725_strides_0, weight = const_274_to_fp16, x = input_723_cast_fp16)[name = string("input_727_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_729_cast_fp16 = silu(x = input_727_cast_fp16)[name = string("input_729_cast_fp16")];
             string x_307_pad_type_0 = const()[name = string("x_307_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_307_strides_0 = const()[name = string("x_307_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_307_pad_0 = const()[name = string("x_307_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_307_dilations_0 = const()[name = string("x_307_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_307_groups_0 = const()[name = string("x_307_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524845]> module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184870144))), indices_nonzero_data = tensor<uint6, [524845]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185001280))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185395008))))[name = string("module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524845]> module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177758976))), indices_nonzero_data = tensor<uint6, [524845]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177890112))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178283840))))[name = string("module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_307_cast_fp16 = conv(dilations = x_307_dilations_0, groups = x_307_groups_0, pad = x_307_pad_0, pad_type = x_307_pad_type_0, strides = x_307_strides_0, weight = module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_729_cast_fp16)[name = string("x_307_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_307_cast_fp16 = conv(dilations = x_307_dilations_0, groups = x_307_groups_0, pad = x_307_pad_0, pad_type = x_307_pad_type_0, strides = x_307_strides_0, weight = module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_729_cast_fp16)[name = string("x_307_cast_fp16")];
             tensor<int32, [3]> input_731_perm_0 = const()[name = string("input_731_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_731_cast_fp16 = transpose(perm = input_731_perm_0, x = x_307_cast_fp16)[name = string("transpose_215")];
-            tensor<fp16, [1, 188, 1024]> input_733_cast_fp16 = add(x = input_715_cast_fp16, y = input_731_cast_fp16)[name = string("input_733_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_731_cast_fp16 = transpose(perm = input_731_perm_0, x = x_307_cast_fp16)[name = string("transpose_215")];
+            tensor<fp16, [1, 128, 1024]> input_733_cast_fp16 = add(x = input_715_cast_fp16, y = input_731_cast_fp16)[name = string("input_733_cast_fp16")];
             tensor<int32, [1]> input_735_axes_0 = const()[name = string("input_735_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_13_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_13_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185411456)))];
-            tensor<fp16, [1024]> module_layers_13_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_13_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185413568)))];
-            tensor<fp16, [1, 188, 1024]> input_735_cast_fp16 = layer_norm(axes = input_735_axes_0, beta = module_layers_13_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_feed_forward2_weight_to_fp16, x = input_733_cast_fp16)[name = string("input_735_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2074364]> module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185415680))), indices_nonzero_data = tensor<uint6, [2074364]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185940032))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187495872))))[name = string("module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_13_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_13_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178300288)))];
+            tensor<fp16, [1024]> module_layers_13_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_13_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178302400)))];
+            tensor<fp16, [1, 128, 1024]> input_735_cast_fp16 = layer_norm(axes = input_735_axes_0, beta = module_layers_13_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_feed_forward2_weight_to_fp16, x = input_733_cast_fp16)[name = string("input_735_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2074364]> module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178304512))), indices_nonzero_data = tensor<uint6, [2074364]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178828864))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180384704))))[name = string("module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_125_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_735_cast_fp16)[name = string("linear_125_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_739_cast_fp16 = silu(x = linear_125_cast_fp16)[name = string("input_739_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1986819]> module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187561472))), indices_nonzero_data = tensor<uint6, [1986819]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188085824))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189576064))))[name = string("module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_125_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_735_cast_fp16)[name = string("linear_125_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_739_cast_fp16 = silu(x = linear_125_cast_fp16)[name = string("input_739_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1986819]> module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180450304))), indices_nonzero_data = tensor<uint6, [1986819]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180974656))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182464896))))[name = string("module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_126_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_739_cast_fp16)[name = string("linear_126_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_126_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_739_cast_fp16)[name = string("linear_126_cast_fp16")];
             fp16 var_2485_to_fp16 = const()[name = string("op_2485_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_2486_cast_fp16 = mul(x = linear_126_cast_fp16, y = var_2485_to_fp16)[name = string("op_2486_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_745_cast_fp16 = add(x = input_733_cast_fp16, y = var_2486_cast_fp16)[name = string("input_745_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_2486_cast_fp16 = mul(x = linear_126_cast_fp16, y = var_2485_to_fp16)[name = string("op_2486_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_745_cast_fp16 = add(x = input_733_cast_fp16, y = var_2486_cast_fp16)[name = string("input_745_cast_fp16")];
             tensor<int32, [1]> input_747_axes_0 = const()[name = string("input_747_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_13_norm_out_weight_to_fp16 = const()[name = string("module_layers_13_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189592512)))];
-            tensor<fp16, [1024]> module_layers_13_norm_out_bias_to_fp16 = const()[name = string("module_layers_13_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189594624)))];
-            tensor<fp16, [1, 188, 1024]> input_747_cast_fp16 = layer_norm(axes = input_747_axes_0, beta = module_layers_13_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_out_weight_to_fp16, x = input_745_cast_fp16)[name = string("input_747_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_13_norm_out_weight_to_fp16 = const()[name = string("module_layers_13_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182481344)))];
+            tensor<fp16, [1024]> module_layers_13_norm_out_bias_to_fp16 = const()[name = string("module_layers_13_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182483456)))];
+            tensor<fp16, [1, 128, 1024]> input_747_cast_fp16 = layer_norm(axes = input_747_axes_0, beta = module_layers_13_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_out_weight_to_fp16, x = input_745_cast_fp16)[name = string("input_747_cast_fp16")];
             tensor<int32, [1]> input_749_axes_0 = const()[name = string("input_749_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_14_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_14_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189596736)))];
-            tensor<fp16, [1024]> module_layers_14_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_14_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189598848)))];
-            tensor<fp16, [1, 188, 1024]> input_749_cast_fp16 = layer_norm(axes = input_749_axes_0, beta = module_layers_14_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_feed_forward1_weight_to_fp16, x = input_747_cast_fp16)[name = string("input_749_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2059717]> module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189600960))), indices_nonzero_data = tensor<uint6, [2059717]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190125312))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191670208))))[name = string("module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_14_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_14_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182485568)))];
+            tensor<fp16, [1024]> module_layers_14_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_14_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182487680)))];
+            tensor<fp16, [1, 128, 1024]> input_749_cast_fp16 = layer_norm(axes = input_749_axes_0, beta = module_layers_14_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_feed_forward1_weight_to_fp16, x = input_747_cast_fp16)[name = string("input_749_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2059717]> module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182489792))), indices_nonzero_data = tensor<uint6, [2059717]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183014144))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184559040))))[name = string("module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_127_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_749_cast_fp16)[name = string("linear_127_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_753_cast_fp16 = silu(x = linear_127_cast_fp16)[name = string("input_753_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1924564]> module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191735808))), indices_nonzero_data = tensor<uint6, [1924564]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192260160))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193703680))))[name = string("module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_127_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_749_cast_fp16)[name = string("linear_127_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_753_cast_fp16 = silu(x = linear_127_cast_fp16)[name = string("input_753_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1924564]> module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184624640))), indices_nonzero_data = tensor<uint6, [1924564]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185148992))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186592512))))[name = string("module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_128_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_753_cast_fp16)[name = string("linear_128_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_128_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_753_cast_fp16)[name = string("linear_128_cast_fp16")];
             fp16 var_2514_to_fp16 = const()[name = string("op_2514_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_2515_cast_fp16 = mul(x = linear_128_cast_fp16, y = var_2514_to_fp16)[name = string("op_2515_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_759_cast_fp16 = add(x = input_747_cast_fp16, y = var_2515_cast_fp16)[name = string("input_759_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_2515_cast_fp16 = mul(x = linear_128_cast_fp16, y = var_2514_to_fp16)[name = string("op_2515_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_759_cast_fp16 = add(x = input_747_cast_fp16, y = var_2515_cast_fp16)[name = string("input_759_cast_fp16")];
             tensor<int32, [1]> query_29_axes_0 = const()[name = string("query_29_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_14_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_14_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193720128)))];
-            tensor<fp16, [1024]> module_layers_14_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_14_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193722240)))];
-            tensor<fp16, [1, 188, 1024]> query_29_cast_fp16 = layer_norm(axes = query_29_axes_0, beta = module_layers_14_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_self_att_weight_to_fp16, x = input_759_cast_fp16)[name = string("query_29_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525274]> module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193724352))), indices_nonzero_data = tensor<uint6, [525274]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193855488))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194249536))))[name = string("module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_14_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_14_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186608960)))];
+            tensor<fp16, [1024]> module_layers_14_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_14_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186611072)))];
+            tensor<fp16, [1, 128, 1024]> query_29_cast_fp16 = layer_norm(axes = query_29_axes_0, beta = module_layers_14_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_self_att_weight_to_fp16, x = input_759_cast_fp16)[name = string("query_29_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525274]> module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186613184))), indices_nonzero_data = tensor<uint6, [525274]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186744320))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187138368))))[name = string("module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_129_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_29_cast_fp16)[name = string("linear_129_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_129_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_29_cast_fp16)[name = string("linear_129_cast_fp16")];
             tensor<int32, [4]> var_2531 = const()[name = string("op_2531"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_85_cast_fp16 = reshape(shape = var_2531, x = linear_129_cast_fp16)[name = string("q_85_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522998]> module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194265984))), indices_nonzero_data = tensor<uint6, [522998]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194397120))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194789440))))[name = string("module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> q_85_cast_fp16 = reshape(shape = var_2531, x = linear_129_cast_fp16)[name = string("q_85_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522998]> module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187154816))), indices_nonzero_data = tensor<uint6, [522998]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187285952))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187678272))))[name = string("module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_130_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_29_cast_fp16)[name = string("linear_130_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_130_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_29_cast_fp16)[name = string("linear_130_cast_fp16")];
             tensor<int32, [4]> var_2535 = const()[name = string("op_2535"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_57_cast_fp16 = reshape(shape = var_2535, x = linear_130_cast_fp16)[name = string("k_57_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522224]> module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194805888))), indices_nonzero_data = tensor<uint6, [522224]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194937024))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195328768))))[name = string("module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> k_57_cast_fp16 = reshape(shape = var_2535, x = linear_130_cast_fp16)[name = string("k_57_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522224]> module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187694720))), indices_nonzero_data = tensor<uint6, [522224]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187825856))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188217600))))[name = string("module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_131_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_29_cast_fp16)[name = string("linear_131_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_131_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_29_cast_fp16)[name = string("linear_131_cast_fp16")];
             tensor<int32, [4]> var_2539 = const()[name = string("op_2539"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_29_cast_fp16 = reshape(shape = var_2539, x = linear_131_cast_fp16)[name = string("v_29_cast_fp16")];
+            tensor<fp16, [1, 128, 8, 128]> v_29_cast_fp16 = reshape(shape = var_2539, x = linear_131_cast_fp16)[name = string("v_29_cast_fp16")];
             tensor<int32, [4]> value_31_perm_0 = const()[name = string("value_31_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_14_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_14_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195345216)))];
-            tensor<fp16, [1, 188, 8, 128]> var_2551_cast_fp16 = add(x = q_85_cast_fp16, y = module_layers_14_self_attn_pos_bias_u_to_fp16)[name = string("op_2551_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_14_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_14_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195347328)))];
-            tensor<fp16, [1, 188, 8, 128]> var_2553_cast_fp16 = add(x = q_85_cast_fp16, y = module_layers_14_self_attn_pos_bias_v_to_fp16)[name = string("op_2553_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_14_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_14_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188234048)))];
+            tensor<fp16, [1, 128, 8, 128]> var_2551_cast_fp16 = add(x = q_85_cast_fp16, y = module_layers_14_self_attn_pos_bias_u_to_fp16)[name = string("op_2551_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_14_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_14_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188236160)))];
+            tensor<fp16, [1, 128, 8, 128]> var_2553_cast_fp16 = add(x = q_85_cast_fp16, y = module_layers_14_self_attn_pos_bias_v_to_fp16)[name = string("op_2553_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_29_perm_0 = const()[name = string("q_with_bias_v_29_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_315_transpose_x_0 = const()[name = string("x_315_transpose_x_0"), val = bool(false)];
             bool x_315_transpose_y_0 = const()[name = string("x_315_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_2555_to_fp16 = const()[name = string("op_2555_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195349440)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_29_cast_fp16 = transpose(perm = q_with_bias_v_29_perm_0, x = var_2553_cast_fp16)[name = string("transpose_214")];
-            tensor<fp16, [1, 8, 188, 375]> x_315_cast_fp16 = matmul(transpose_x = x_315_transpose_x_0, transpose_y = x_315_transpose_y_0, x = q_with_bias_v_29_cast_fp16, y = var_2555_to_fp16)[name = string("x_315_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 255]> var_2555_to_fp16 = const()[name = string("op_2555_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314734080)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_29_cast_fp16 = transpose(perm = q_with_bias_v_29_perm_0, x = var_2553_cast_fp16)[name = string("transpose_214")];
+            tensor<fp16, [1, 8, 128, 255]> x_315_cast_fp16 = matmul(transpose_x = x_315_transpose_x_0, transpose_y = x_315_transpose_y_0, x = q_with_bias_v_29_cast_fp16, y = var_2555_to_fp16)[name = string("x_315_cast_fp16")];
             tensor<int32, [8]> x_317_pad_0 = const()[name = string("x_317_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_317_mode_0 = const()[name = string("x_317_mode_0"), val = string("constant")];
             fp16 const_154_to_fp16 = const()[name = string("const_154_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_317_cast_fp16 = pad(constant_val = const_154_to_fp16, mode = x_317_mode_0, pad = x_317_pad_0, x = x_315_cast_fp16)[name = string("x_317_cast_fp16")];
-            tensor<int32, [4]> var_2563 = const()[name = string("op_2563"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_319_cast_fp16 = reshape(shape = var_2563, x = x_317_cast_fp16)[name = string("x_319_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 256]> x_317_cast_fp16 = pad(constant_val = const_154_to_fp16, mode = x_317_mode_0, pad = x_317_pad_0, x = x_315_cast_fp16)[name = string("x_317_cast_fp16")];
+            tensor<int32, [4]> var_2563 = const()[name = string("op_2563"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_319_cast_fp16 = reshape(shape = var_2563, x = x_317_cast_fp16)[name = string("x_319_cast_fp16")];
             tensor<int32, [4]> var_2567_begin_0 = const()[name = string("op_2567_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_2567_end_0 = const()[name = string("op_2567_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_2567_end_0 = const()[name = string("op_2567_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
             tensor<bool, [4]> var_2567_end_mask_0 = const()[name = string("op_2567_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_2567_cast_fp16 = slice_by_index(begin = var_2567_begin_0, end = var_2567_end_0, end_mask = var_2567_end_mask_0, x = x_319_cast_fp16)[name = string("op_2567_cast_fp16")];
-            tensor<int32, [4]> var_2568 = const()[name = string("op_2568"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_57_cast_fp16 = reshape(shape = var_2568, x = var_2567_cast_fp16)[name = string("matrix_bd_57_cast_fp16")];
+            tensor<fp16, [1, 8, 255, 128]> var_2567_cast_fp16 = slice_by_index(begin = var_2567_begin_0, end = var_2567_end_0, end_mask = var_2567_end_mask_0, x = x_319_cast_fp16)[name = string("op_2567_cast_fp16")];
+            tensor<int32, [4]> var_2568 = const()[name = string("op_2568"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_57_cast_fp16 = reshape(shape = var_2568, x = var_2567_cast_fp16)[name = string("matrix_bd_57_cast_fp16")];
             bool matrix_ac_29_transpose_x_0 = const()[name = string("matrix_ac_29_transpose_x_0"), val = bool(false)];
             bool matrix_ac_29_transpose_y_0 = const()[name = string("matrix_ac_29_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_124_perm_0 = const()[name = string("transpose_124_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_125_perm_0 = const()[name = string("transpose_125_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_125 = transpose(perm = transpose_125_perm_0, x = k_57_cast_fp16)[name = string("transpose_212")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_124 = transpose(perm = transpose_124_perm_0, x = var_2551_cast_fp16)[name = string("transpose_213")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_29_cast_fp16 = matmul(transpose_x = matrix_ac_29_transpose_x_0, transpose_y = matrix_ac_29_transpose_y_0, x = transpose_124, y = transpose_125)[name = string("matrix_ac_29_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_125 = transpose(perm = transpose_125_perm_0, x = k_57_cast_fp16)[name = string("transpose_212")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_124 = transpose(perm = transpose_124_perm_0, x = var_2551_cast_fp16)[name = string("transpose_213")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_29_cast_fp16 = matmul(transpose_x = matrix_ac_29_transpose_x_0, transpose_y = matrix_ac_29_transpose_y_0, x = transpose_124, y = transpose_125)[name = string("matrix_ac_29_cast_fp16")];
             tensor<int32, [4]> matrix_bd_59_begin_0 = const()[name = string("matrix_bd_59_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_59_end_0 = const()[name = string("matrix_bd_59_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_59_end_0 = const()[name = string("matrix_bd_59_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
             tensor<bool, [4]> matrix_bd_59_end_mask_0 = const()[name = string("matrix_bd_59_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_59_cast_fp16 = slice_by_index(begin = matrix_bd_59_begin_0, end = matrix_bd_59_end_0, end_mask = matrix_bd_59_end_mask_0, x = matrix_bd_57_cast_fp16)[name = string("matrix_bd_59_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_2577_cast_fp16 = add(x = matrix_ac_29_cast_fp16, y = matrix_bd_59_cast_fp16)[name = string("op_2577_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_59_cast_fp16 = slice_by_index(begin = matrix_bd_59_begin_0, end = matrix_bd_59_end_0, end_mask = matrix_bd_59_end_mask_0, x = matrix_bd_57_cast_fp16)[name = string("matrix_bd_59_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_2577_cast_fp16 = add(x = matrix_ac_29_cast_fp16, y = matrix_bd_59_cast_fp16)[name = string("op_2577_cast_fp16")];
             fp16 _inversed_scores_57_y_0_to_fp16 = const()[name = string("_inversed_scores_57_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_57_cast_fp16 = mul(x = var_2577_cast_fp16, y = _inversed_scores_57_y_0_to_fp16)[name = string("_inversed_scores_57_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_59_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_57_cast_fp16, cond = mask_3)[name = string("scores_59_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_2583_cast_fp16 = softmax(axis = var_30, x = scores_59_cast_fp16)[name = string("op_2583_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_761_cast_fp16 = select(a = var_11_to_fp16, b = var_2583_cast_fp16, cond = mask_3)[name = string("input_761_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_57_cast_fp16 = mul(x = var_2577_cast_fp16, y = _inversed_scores_57_y_0_to_fp16)[name = string("_inversed_scores_57_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_59_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_57_cast_fp16, cond = mask_3)[name = string("scores_59_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_2583_cast_fp16 = softmax(axis = var_30, x = scores_59_cast_fp16)[name = string("op_2583_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_761_cast_fp16 = select(a = var_11_to_fp16, b = var_2583_cast_fp16, cond = mask_3)[name = string("input_761_cast_fp16")];
             bool x_321_transpose_x_0 = const()[name = string("x_321_transpose_x_0"), val = bool(false)];
             bool x_321_transpose_y_0 = const()[name = string("x_321_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_31_cast_fp16 = transpose(perm = value_31_perm_0, x = v_29_cast_fp16)[name = string("transpose_211")];
-            tensor<fp16, [1, 8, 188, 128]> x_321_cast_fp16 = matmul(transpose_x = x_321_transpose_x_0, transpose_y = x_321_transpose_y_0, x = input_761_cast_fp16, y = value_31_cast_fp16)[name = string("x_321_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> value_31_cast_fp16 = transpose(perm = value_31_perm_0, x = v_29_cast_fp16)[name = string("transpose_211")];
+            tensor<fp16, [1, 8, 128, 128]> x_321_cast_fp16 = matmul(transpose_x = x_321_transpose_x_0, transpose_y = x_321_transpose_y_0, x = input_761_cast_fp16, y = value_31_cast_fp16)[name = string("x_321_cast_fp16")];
             tensor<int32, [4]> var_2587_perm_0 = const()[name = string("op_2587_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_2588 = const()[name = string("op_2588"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_2587_cast_fp16 = transpose(perm = var_2587_perm_0, x = x_321_cast_fp16)[name = string("transpose_210")];
-            tensor<fp16, [1, 188, 1024]> input_763_cast_fp16 = reshape(shape = var_2588, x = var_2587_cast_fp16)[name = string("input_763_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524764]> module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196117504))), indices_nonzero_data = tensor<uint6, [524764]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196248640))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196642304))))[name = string("module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> var_2587_cast_fp16 = transpose(perm = var_2587_perm_0, x = x_321_cast_fp16)[name = string("transpose_210")];
+            tensor<fp16, [1, 128, 1024]> input_763_cast_fp16 = reshape(shape = var_2588, x = var_2587_cast_fp16)[name = string("input_763_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524764]> module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188498432))), indices_nonzero_data = tensor<uint6, [524764]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188629568))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189023232))))[name = string("module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_133_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_763_cast_fp16)[name = string("linear_133_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_767_cast_fp16 = add(x = input_759_cast_fp16, y = linear_133_cast_fp16)[name = string("input_767_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_133_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_763_cast_fp16)[name = string("linear_133_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_767_cast_fp16 = add(x = input_759_cast_fp16, y = linear_133_cast_fp16)[name = string("input_767_cast_fp16")];
             tensor<int32, [1]> x_325_axes_0 = const()[name = string("x_325_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_14_norm_conv_weight_to_fp16 = const()[name = string("module_layers_14_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196658752)))];
-            tensor<fp16, [1024]> module_layers_14_norm_conv_bias_to_fp16 = const()[name = string("module_layers_14_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196660864)))];
-            tensor<fp16, [1, 188, 1024]> x_325_cast_fp16 = layer_norm(axes = x_325_axes_0, beta = module_layers_14_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_conv_weight_to_fp16, x = input_767_cast_fp16)[name = string("x_325_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_14_norm_conv_weight_to_fp16 = const()[name = string("module_layers_14_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189039680)))];
+            tensor<fp16, [1024]> module_layers_14_norm_conv_bias_to_fp16 = const()[name = string("module_layers_14_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189041792)))];
+            tensor<fp16, [1, 128, 1024]> x_325_cast_fp16 = layer_norm(axes = x_325_axes_0, beta = module_layers_14_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_conv_weight_to_fp16, x = input_767_cast_fp16)[name = string("x_325_cast_fp16")];
             tensor<int32, [3]> input_769_perm_0 = const()[name = string("input_769_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_771_pad_type_0 = const()[name = string("input_771_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_771_strides_0 = const()[name = string("input_771_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_771_pad_0 = const()[name = string("input_771_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_771_dilations_0 = const()[name = string("input_771_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_771_groups_0 = const()[name = string("input_771_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1043300]> module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196662976))), indices_nonzero_data = tensor<uint6, [1043300]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196925184))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197707776))))[name = string("module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1043300]> module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189043904))), indices_nonzero_data = tensor<uint6, [1043300]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189306112))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190088704))))[name = string("module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_769_cast_fp16 = transpose(perm = input_769_perm_0, x = x_325_cast_fp16)[name = string("transpose_209")];
-            tensor<fp16, [1, 2048, 188]> input_771_cast_fp16 = conv(dilations = input_771_dilations_0, groups = input_771_groups_0, pad = input_771_pad_0, pad_type = input_771_pad_type_0, strides = input_771_strides_0, weight = module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_769_cast_fp16)[name = string("input_771_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_769_cast_fp16 = transpose(perm = input_769_perm_0, x = x_325_cast_fp16)[name = string("transpose_209")];
+            tensor<fp16, [1, 2048, 128]> input_771_cast_fp16 = conv(dilations = input_771_dilations_0, groups = input_771_groups_0, pad = input_771_pad_0, pad_type = input_771_pad_type_0, strides = input_771_strides_0, weight = module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_769_cast_fp16)[name = string("input_771_cast_fp16")];
             int32 x_327_split_num_splits_0 = const()[name = string("x_327_split_num_splits_0"), val = int32(2)];
             int32 x_327_split_axis_0 = const()[name = string("x_327_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_327_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_327_split_cast_fp16_1 = split(axis = x_327_split_axis_0, num_splits = x_327_split_num_splits_0, x = input_771_cast_fp16)[name = string("x_327_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_327_split_1_sigmoid_cast_fp16 = sigmoid(x = x_327_split_cast_fp16_1)[name = string("x_327_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_327_cast_fp16 = mul(x = x_327_split_cast_fp16_0, y = x_327_split_1_sigmoid_cast_fp16)[name = string("x_327_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_773_cast_fp16 = select(a = var_11_to_fp16, b = x_327_cast_fp16, cond = var_328)[name = string("input_773_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_327_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_327_split_cast_fp16_1 = split(axis = x_327_split_axis_0, num_splits = x_327_split_num_splits_0, x = input_771_cast_fp16)[name = string("x_327_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_327_split_1_sigmoid_cast_fp16 = sigmoid(x = x_327_split_cast_fp16_1)[name = string("x_327_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_327_cast_fp16 = mul(x = x_327_split_cast_fp16_0, y = x_327_split_1_sigmoid_cast_fp16)[name = string("x_327_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_773_cast_fp16 = select(a = var_11_to_fp16, b = x_327_cast_fp16, cond = var_328)[name = string("input_773_cast_fp16")];
             tensor<int32, [6]> input_775_pad_0 = const()[name = string("input_775_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_775_mode_0 = const()[name = string("input_775_mode_0"), val = string("constant")];
             fp16 const_157_to_fp16 = const()[name = string("const_157_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_775_cast_fp16 = pad(constant_val = const_157_to_fp16, mode = input_775_mode_0, pad = input_775_pad_0, x = input_773_cast_fp16)[name = string("input_775_cast_fp16")];
+            tensor<fp16, [1, 1024, 136]> input_775_cast_fp16 = pad(constant_val = const_157_to_fp16, mode = input_775_mode_0, pad = input_775_pad_0, x = input_773_cast_fp16)[name = string("input_775_cast_fp16")];
             string input_777_pad_type_0 = const()[name = string("input_777_pad_type_0"), val = string("valid")];
             int32 input_777_groups_0 = const()[name = string("input_777_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_777_strides_0 = const()[name = string("input_777_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_777_pad_0 = const()[name = string("input_777_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_777_dilations_0 = const()[name = string("input_777_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_276_to_fp16 = const()[name = string("const_276_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197740608)))];
-            tensor<fp16, [1024]> const_277_to_fp16 = const()[name = string("const_277_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197759104)))];
-            tensor<fp16, [1, 1024, 188]> input_779_cast_fp16 = conv(bias = const_277_to_fp16, dilations = input_777_dilations_0, groups = input_777_groups_0, pad = input_777_pad_0, pad_type = input_777_pad_type_0, strides = input_777_strides_0, weight = const_276_to_fp16, x = input_775_cast_fp16)[name = string("input_779_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_781_cast_fp16 = silu(x = input_779_cast_fp16)[name = string("input_781_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_276_to_fp16 = const()[name = string("const_276_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190121536)))];
+            tensor<fp16, [1024]> const_277_to_fp16 = const()[name = string("const_277_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190140032)))];
+            tensor<fp16, [1, 1024, 128]> input_779_cast_fp16 = conv(bias = const_277_to_fp16, dilations = input_777_dilations_0, groups = input_777_groups_0, pad = input_777_pad_0, pad_type = input_777_pad_type_0, strides = input_777_strides_0, weight = const_276_to_fp16, x = input_775_cast_fp16)[name = string("input_779_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_781_cast_fp16 = silu(x = input_779_cast_fp16)[name = string("input_781_cast_fp16")];
             string x_329_pad_type_0 = const()[name = string("x_329_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_329_strides_0 = const()[name = string("x_329_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_329_pad_0 = const()[name = string("x_329_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_329_dilations_0 = const()[name = string("x_329_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_329_groups_0 = const()[name = string("x_329_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525590]> module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197761216))), indices_nonzero_data = tensor<uint6, [525590]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197892352))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198286656))))[name = string("module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525590]> module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190142144))), indices_nonzero_data = tensor<uint6, [525590]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190273280))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190667584))))[name = string("module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_329_cast_fp16 = conv(dilations = x_329_dilations_0, groups = x_329_groups_0, pad = x_329_pad_0, pad_type = x_329_pad_type_0, strides = x_329_strides_0, weight = module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_781_cast_fp16)[name = string("x_329_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_329_cast_fp16 = conv(dilations = x_329_dilations_0, groups = x_329_groups_0, pad = x_329_pad_0, pad_type = x_329_pad_type_0, strides = x_329_strides_0, weight = module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_781_cast_fp16)[name = string("x_329_cast_fp16")];
             tensor<int32, [3]> input_783_perm_0 = const()[name = string("input_783_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_783_cast_fp16 = transpose(perm = input_783_perm_0, x = x_329_cast_fp16)[name = string("transpose_208")];
-            tensor<fp16, [1, 188, 1024]> input_785_cast_fp16 = add(x = input_767_cast_fp16, y = input_783_cast_fp16)[name = string("input_785_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_783_cast_fp16 = transpose(perm = input_783_perm_0, x = x_329_cast_fp16)[name = string("transpose_208")];
+            tensor<fp16, [1, 128, 1024]> input_785_cast_fp16 = add(x = input_767_cast_fp16, y = input_783_cast_fp16)[name = string("input_785_cast_fp16")];
             tensor<int32, [1]> input_787_axes_0 = const()[name = string("input_787_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_14_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_14_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198303104)))];
-            tensor<fp16, [1024]> module_layers_14_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_14_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198305216)))];
-            tensor<fp16, [1, 188, 1024]> input_787_cast_fp16 = layer_norm(axes = input_787_axes_0, beta = module_layers_14_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_feed_forward2_weight_to_fp16, x = input_785_cast_fp16)[name = string("input_787_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2073313]> module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198307328))), indices_nonzero_data = tensor<uint6, [2073313]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198831680))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200386752))))[name = string("module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_14_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_14_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190684032)))];
+            tensor<fp16, [1024]> module_layers_14_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_14_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190686144)))];
+            tensor<fp16, [1, 128, 1024]> input_787_cast_fp16 = layer_norm(axes = input_787_axes_0, beta = module_layers_14_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_feed_forward2_weight_to_fp16, x = input_785_cast_fp16)[name = string("input_787_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2073313]> module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190688256))), indices_nonzero_data = tensor<uint6, [2073313]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191212608))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192767680))))[name = string("module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_134_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_787_cast_fp16)[name = string("linear_134_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_791_cast_fp16 = silu(x = linear_134_cast_fp16)[name = string("input_791_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1989781]> module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200452352))), indices_nonzero_data = tensor<uint6, [1989781]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200976704))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202469120))))[name = string("module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_134_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_787_cast_fp16)[name = string("linear_134_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_791_cast_fp16 = silu(x = linear_134_cast_fp16)[name = string("input_791_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1989781]> module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192833280))), indices_nonzero_data = tensor<uint6, [1989781]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193357632))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194850048))))[name = string("module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_135_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_791_cast_fp16)[name = string("linear_135_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_135_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_791_cast_fp16)[name = string("linear_135_cast_fp16")];
             fp16 var_2648_to_fp16 = const()[name = string("op_2648_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_2649_cast_fp16 = mul(x = linear_135_cast_fp16, y = var_2648_to_fp16)[name = string("op_2649_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_797_cast_fp16 = add(x = input_785_cast_fp16, y = var_2649_cast_fp16)[name = string("input_797_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_2649_cast_fp16 = mul(x = linear_135_cast_fp16, y = var_2648_to_fp16)[name = string("op_2649_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_797_cast_fp16 = add(x = input_785_cast_fp16, y = var_2649_cast_fp16)[name = string("input_797_cast_fp16")];
             tensor<int32, [1]> input_799_axes_0 = const()[name = string("input_799_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_14_norm_out_weight_to_fp16 = const()[name = string("module_layers_14_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202485568)))];
-            tensor<fp16, [1024]> module_layers_14_norm_out_bias_to_fp16 = const()[name = string("module_layers_14_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202487680)))];
-            tensor<fp16, [1, 188, 1024]> input_799_cast_fp16 = layer_norm(axes = input_799_axes_0, beta = module_layers_14_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_out_weight_to_fp16, x = input_797_cast_fp16)[name = string("input_799_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_14_norm_out_weight_to_fp16 = const()[name = string("module_layers_14_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194866496)))];
+            tensor<fp16, [1024]> module_layers_14_norm_out_bias_to_fp16 = const()[name = string("module_layers_14_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194868608)))];
+            tensor<fp16, [1, 128, 1024]> input_799_cast_fp16 = layer_norm(axes = input_799_axes_0, beta = module_layers_14_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_out_weight_to_fp16, x = input_797_cast_fp16)[name = string("input_799_cast_fp16")];
             tensor<int32, [1]> input_801_axes_0 = const()[name = string("input_801_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_15_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_15_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202489792)))];
-            tensor<fp16, [1024]> module_layers_15_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_15_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202491904)))];
-            tensor<fp16, [1, 188, 1024]> input_801_cast_fp16 = layer_norm(axes = input_801_axes_0, beta = module_layers_15_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_feed_forward1_weight_to_fp16, x = input_799_cast_fp16)[name = string("input_801_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2064877]> module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202494016))), indices_nonzero_data = tensor<uint6, [2064877]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203018368))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204567104))))[name = string("module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_15_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_15_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194870720)))];
+            tensor<fp16, [1024]> module_layers_15_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_15_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194872832)))];
+            tensor<fp16, [1, 128, 1024]> input_801_cast_fp16 = layer_norm(axes = input_801_axes_0, beta = module_layers_15_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_feed_forward1_weight_to_fp16, x = input_799_cast_fp16)[name = string("input_801_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2064877]> module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194874944))), indices_nonzero_data = tensor<uint6, [2064877]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195399296))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196948032))))[name = string("module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_136_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_801_cast_fp16)[name = string("linear_136_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_805_cast_fp16 = silu(x = linear_136_cast_fp16)[name = string("input_805_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1919271]> module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204632704))), indices_nonzero_data = tensor<uint6, [1919271]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205157056))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206596608))))[name = string("module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 4096]> linear_136_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_801_cast_fp16)[name = string("linear_136_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_805_cast_fp16 = silu(x = linear_136_cast_fp16)[name = string("input_805_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1919271]> module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197013632))), indices_nonzero_data = tensor<uint6, [1919271]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197537984))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198977536))))[name = string("module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_137_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_805_cast_fp16)[name = string("linear_137_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_137_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_805_cast_fp16)[name = string("linear_137_cast_fp16")];
             fp16 var_2677_to_fp16 = const()[name = string("op_2677_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_2678_cast_fp16 = mul(x = linear_137_cast_fp16, y = var_2677_to_fp16)[name = string("op_2678_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_811_cast_fp16 = add(x = input_799_cast_fp16, y = var_2678_cast_fp16)[name = string("input_811_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> var_2678_cast_fp16 = mul(x = linear_137_cast_fp16, y = var_2677_to_fp16)[name = string("op_2678_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_811_cast_fp16 = add(x = input_799_cast_fp16, y = var_2678_cast_fp16)[name = string("input_811_cast_fp16")];
             tensor<int32, [1]> query_31_axes_0 = const()[name = string("query_31_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_15_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_15_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206613056)))];
-            tensor<fp16, [1024]> module_layers_15_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_15_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206615168)))];
-            tensor<fp16, [1, 188, 1024]> query_31_cast_fp16 = layer_norm(axes = query_31_axes_0, beta = module_layers_15_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_self_att_weight_to_fp16, x = input_811_cast_fp16)[name = string("query_31_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525795]> module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206617280))), indices_nonzero_data = tensor<uint6, [525795]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206748416))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207142848))))[name = string("module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_15_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_15_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198993984)))];
+            tensor<fp16, [1024]> module_layers_15_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_15_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198996096)))];
+            tensor<fp16, [1, 128, 1024]> query_31_cast_fp16 = layer_norm(axes = query_31_axes_0, beta = module_layers_15_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_self_att_weight_to_fp16, x = input_811_cast_fp16)[name = string("query_31_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525795]> module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198998208))), indices_nonzero_data = tensor<uint6, [525795]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199129344))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199523776))))[name = string("module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_138_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_31_cast_fp16)[name = string("linear_138_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_138_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_31_cast_fp16)[name = string("linear_138_cast_fp16")];
             tensor<int32, [4]> var_2694 = const()[name = string("op_2694"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_91_cast_fp16 = reshape(shape = var_2694, x = linear_138_cast_fp16)[name = string("q_91_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523094]> module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207159296))), indices_nonzero_data = tensor<uint6, [523094]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207290432))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207682880))))[name = string("module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> q_91_cast_fp16 = reshape(shape = var_2694, x = linear_138_cast_fp16)[name = string("q_91_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523094]> module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199540224))), indices_nonzero_data = tensor<uint6, [523094]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199671360))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200063808))))[name = string("module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_139_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_31_cast_fp16)[name = string("linear_139_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_139_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_31_cast_fp16)[name = string("linear_139_cast_fp16")];
             tensor<int32, [4]> var_2698 = const()[name = string("op_2698"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_61_cast_fp16 = reshape(shape = var_2698, x = linear_139_cast_fp16)[name = string("k_61_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523685]> module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207699328))), indices_nonzero_data = tensor<uint6, [523685]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207830464))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208223296))))[name = string("module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 128, 8, 128]> k_61_cast_fp16 = reshape(shape = var_2698, x = linear_139_cast_fp16)[name = string("k_61_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523685]> module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200080256))), indices_nonzero_data = tensor<uint6, [523685]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200211392))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200604224))))[name = string("module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_140_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_31_cast_fp16)[name = string("linear_140_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> linear_140_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_31_cast_fp16)[name = string("linear_140_cast_fp16")];
             tensor<int32, [4]> var_2702 = const()[name = string("op_2702"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_31_cast_fp16 = reshape(shape = var_2702, x = linear_140_cast_fp16)[name = string("v_31_cast_fp16")];
+            tensor<fp16, [1, 128, 8, 128]> v_31_cast_fp16 = reshape(shape = var_2702, x = linear_140_cast_fp16)[name = string("v_31_cast_fp16")];
             tensor<int32, [4]> value_33_perm_0 = const()[name = string("value_33_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_15_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_15_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208239744)))];
-            tensor<fp16, [1, 188, 8, 128]> var_2714_cast_fp16 = add(x = q_91_cast_fp16, y = module_layers_15_self_attn_pos_bias_u_to_fp16)[name = string("op_2714_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_15_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_15_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208241856)))];
-            tensor<fp16, [1, 188, 8, 128]> var_2716_cast_fp16 = add(x = q_91_cast_fp16, y = module_layers_15_self_attn_pos_bias_v_to_fp16)[name = string("op_2716_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_15_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_15_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200620672)))];
+            tensor<fp16, [1, 128, 8, 128]> var_2714_cast_fp16 = add(x = q_91_cast_fp16, y = module_layers_15_self_attn_pos_bias_u_to_fp16)[name = string("op_2714_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_15_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_15_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200622784)))];
+            tensor<fp16, [1, 128, 8, 128]> var_2716_cast_fp16 = add(x = q_91_cast_fp16, y = module_layers_15_self_attn_pos_bias_v_to_fp16)[name = string("op_2716_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_31_perm_0 = const()[name = string("q_with_bias_v_31_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_337_transpose_x_0 = const()[name = string("x_337_transpose_x_0"), val = bool(false)];
             bool x_337_transpose_y_0 = const()[name = string("x_337_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_2718_to_fp16 = const()[name = string("op_2718_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208243968)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_31_cast_fp16 = transpose(perm = q_with_bias_v_31_perm_0, x = var_2716_cast_fp16)[name = string("transpose_207")];
-            tensor<fp16, [1, 8, 188, 375]> x_337_cast_fp16 = matmul(transpose_x = x_337_transpose_x_0, transpose_y = x_337_transpose_y_0, x = q_with_bias_v_31_cast_fp16, y = var_2718_to_fp16)[name = string("x_337_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 255]> var_2718_to_fp16 = const()[name = string("op_2718_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315256384)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_31_cast_fp16 = transpose(perm = q_with_bias_v_31_perm_0, x = var_2716_cast_fp16)[name = string("transpose_207")];
+            tensor<fp16, [1, 8, 128, 255]> x_337_cast_fp16 = matmul(transpose_x = x_337_transpose_x_0, transpose_y = x_337_transpose_y_0, x = q_with_bias_v_31_cast_fp16, y = var_2718_to_fp16)[name = string("x_337_cast_fp16")];
             tensor<int32, [8]> x_339_pad_0 = const()[name = string("x_339_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_339_mode_0 = const()[name = string("x_339_mode_0"), val = string("constant")];
             fp16 const_164_to_fp16 = const()[name = string("const_164_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_339_cast_fp16 = pad(constant_val = const_164_to_fp16, mode = x_339_mode_0, pad = x_339_pad_0, x = x_337_cast_fp16)[name = string("x_339_cast_fp16")];
-            tensor<int32, [4]> var_2726 = const()[name = string("op_2726"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_341_cast_fp16 = reshape(shape = var_2726, x = x_339_cast_fp16)[name = string("x_341_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 256]> x_339_cast_fp16 = pad(constant_val = const_164_to_fp16, mode = x_339_mode_0, pad = x_339_pad_0, x = x_337_cast_fp16)[name = string("x_339_cast_fp16")];
+            tensor<int32, [4]> var_2726 = const()[name = string("op_2726"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_341_cast_fp16 = reshape(shape = var_2726, x = x_339_cast_fp16)[name = string("x_341_cast_fp16")];
             tensor<int32, [4]> var_2730_begin_0 = const()[name = string("op_2730_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_2730_end_0 = const()[name = string("op_2730_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_2730_end_0 = const()[name = string("op_2730_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
             tensor<bool, [4]> var_2730_end_mask_0 = const()[name = string("op_2730_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_2730_cast_fp16 = slice_by_index(begin = var_2730_begin_0, end = var_2730_end_0, end_mask = var_2730_end_mask_0, x = x_341_cast_fp16)[name = string("op_2730_cast_fp16")];
-            tensor<int32, [4]> var_2731 = const()[name = string("op_2731"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_61_cast_fp16 = reshape(shape = var_2731, x = var_2730_cast_fp16)[name = string("matrix_bd_61_cast_fp16")];
+            tensor<fp16, [1, 8, 255, 128]> var_2730_cast_fp16 = slice_by_index(begin = var_2730_begin_0, end = var_2730_end_0, end_mask = var_2730_end_mask_0, x = x_341_cast_fp16)[name = string("op_2730_cast_fp16")];
+            tensor<int32, [4]> var_2731 = const()[name = string("op_2731"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_61_cast_fp16 = reshape(shape = var_2731, x = var_2730_cast_fp16)[name = string("matrix_bd_61_cast_fp16")];
             bool matrix_ac_31_transpose_x_0 = const()[name = string("matrix_ac_31_transpose_x_0"), val = bool(false)];
             bool matrix_ac_31_transpose_y_0 = const()[name = string("matrix_ac_31_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_126_perm_0 = const()[name = string("transpose_126_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_127_perm_0 = const()[name = string("transpose_127_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_127 = transpose(perm = transpose_127_perm_0, x = k_61_cast_fp16)[name = string("transpose_205")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_126 = transpose(perm = transpose_126_perm_0, x = var_2714_cast_fp16)[name = string("transpose_206")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_31_cast_fp16 = matmul(transpose_x = matrix_ac_31_transpose_x_0, transpose_y = matrix_ac_31_transpose_y_0, x = transpose_126, y = transpose_127)[name = string("matrix_ac_31_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_127 = transpose(perm = transpose_127_perm_0, x = k_61_cast_fp16)[name = string("transpose_205")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_126 = transpose(perm = transpose_126_perm_0, x = var_2714_cast_fp16)[name = string("transpose_206")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_31_cast_fp16 = matmul(transpose_x = matrix_ac_31_transpose_x_0, transpose_y = matrix_ac_31_transpose_y_0, x = transpose_126, y = transpose_127)[name = string("matrix_ac_31_cast_fp16")];
             tensor<int32, [4]> matrix_bd_63_begin_0 = const()[name = string("matrix_bd_63_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_63_end_0 = const()[name = string("matrix_bd_63_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_63_end_0 = const()[name = string("matrix_bd_63_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
             tensor<bool, [4]> matrix_bd_63_end_mask_0 = const()[name = string("matrix_bd_63_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_63_cast_fp16 = slice_by_index(begin = matrix_bd_63_begin_0, end = matrix_bd_63_end_0, end_mask = matrix_bd_63_end_mask_0, x = matrix_bd_61_cast_fp16)[name = string("matrix_bd_63_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_2740_cast_fp16 = add(x = matrix_ac_31_cast_fp16, y = matrix_bd_63_cast_fp16)[name = string("op_2740_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_63_cast_fp16 = slice_by_index(begin = matrix_bd_63_begin_0, end = matrix_bd_63_end_0, end_mask = matrix_bd_63_end_mask_0, x = matrix_bd_61_cast_fp16)[name = string("matrix_bd_63_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_2740_cast_fp16 = add(x = matrix_ac_31_cast_fp16, y = matrix_bd_63_cast_fp16)[name = string("op_2740_cast_fp16")];
             fp16 _inversed_scores_61_y_0_to_fp16 = const()[name = string("_inversed_scores_61_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_61_cast_fp16 = mul(x = var_2740_cast_fp16, y = _inversed_scores_61_y_0_to_fp16)[name = string("_inversed_scores_61_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_63_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_61_cast_fp16, cond = mask_3)[name = string("scores_63_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_2746_cast_fp16 = softmax(axis = var_30, x = scores_63_cast_fp16)[name = string("op_2746_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_813_cast_fp16 = select(a = var_11_to_fp16, b = var_2746_cast_fp16, cond = mask_3)[name = string("input_813_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_61_cast_fp16 = mul(x = var_2740_cast_fp16, y = _inversed_scores_61_y_0_to_fp16)[name = string("_inversed_scores_61_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_63_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_61_cast_fp16, cond = mask_3)[name = string("scores_63_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_2746_cast_fp16 = softmax(axis = var_30, x = scores_63_cast_fp16)[name = string("op_2746_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_813_cast_fp16 = select(a = var_11_to_fp16, b = var_2746_cast_fp16, cond = mask_3)[name = string("input_813_cast_fp16")];
+            bool x_343_transpose_x_0 = const()[name = string("x_343_transpose_x_0"), val = bool(false)];
+            bool x_343_transpose_y_0 = const()[name = string("x_343_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 128]> value_33_cast_fp16 = transpose(perm = value_33_perm_0, x = v_31_cast_fp16)[name = string("transpose_204")];
+            tensor<fp16, [1, 8, 128, 128]> x_343_cast_fp16 = matmul(transpose_x = x_343_transpose_x_0, transpose_y = x_343_transpose_y_0, x = input_813_cast_fp16, y = value_33_cast_fp16)[name = string("x_343_cast_fp16")];
+            tensor<int32, [4]> var_2750_perm_0 = const()[name = string("op_2750_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2751 = const()[name = string("op_2751"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 128, 8, 128]> var_2750_cast_fp16 = transpose(perm = var_2750_perm_0, x = x_343_cast_fp16)[name = string("transpose_203")];
+            tensor<fp16, [1, 128, 1024]> input_815_cast_fp16 = reshape(shape = var_2751, x = var_2750_cast_fp16)[name = string("input_815_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [526464]> module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200885056))), indices_nonzero_data = tensor<uint6, [526464]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201016192))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201411136))))[name = string("module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_142_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_815_cast_fp16)[name = string("linear_142_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_819_cast_fp16 = add(x = input_811_cast_fp16, y = linear_142_cast_fp16)[name = string("input_819_cast_fp16")];
+            tensor<int32, [1]> x_347_axes_0 = const()[name = string("x_347_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_15_norm_conv_weight_to_fp16 = const()[name = string("module_layers_15_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201427584)))];
+            tensor<fp16, [1024]> module_layers_15_norm_conv_bias_to_fp16 = const()[name = string("module_layers_15_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201429696)))];
+            tensor<fp16, [1, 128, 1024]> x_347_cast_fp16 = layer_norm(axes = x_347_axes_0, beta = module_layers_15_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_conv_weight_to_fp16, x = input_819_cast_fp16)[name = string("x_347_cast_fp16")];
+            tensor<int32, [3]> input_821_perm_0 = const()[name = string("input_821_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_823_pad_type_0 = const()[name = string("input_823_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_823_strides_0 = const()[name = string("input_823_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_823_pad_0 = const()[name = string("input_823_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_823_dilations_0 = const()[name = string("input_823_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_823_groups_0 = const()[name = string("input_823_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1046264]> module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201431808))), indices_nonzero_data = tensor<uint6, [1046264]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201694016))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202478784))))[name = string("module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 128]> input_821_cast_fp16 = transpose(perm = input_821_perm_0, x = x_347_cast_fp16)[name = string("transpose_202")];
+            tensor<fp16, [1, 2048, 128]> input_823_cast_fp16 = conv(dilations = input_823_dilations_0, groups = input_823_groups_0, pad = input_823_pad_0, pad_type = input_823_pad_type_0, strides = input_823_strides_0, weight = module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_821_cast_fp16)[name = string("input_823_cast_fp16")];
+            int32 x_349_split_num_splits_0 = const()[name = string("x_349_split_num_splits_0"), val = int32(2)];
+            int32 x_349_split_axis_0 = const()[name = string("x_349_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 128]> x_349_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_349_split_cast_fp16_1 = split(axis = x_349_split_axis_0, num_splits = x_349_split_num_splits_0, x = input_823_cast_fp16)[name = string("x_349_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_349_split_1_sigmoid_cast_fp16 = sigmoid(x = x_349_split_cast_fp16_1)[name = string("x_349_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_349_cast_fp16 = mul(x = x_349_split_cast_fp16_0, y = x_349_split_1_sigmoid_cast_fp16)[name = string("x_349_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_825_cast_fp16 = select(a = var_11_to_fp16, b = x_349_cast_fp16, cond = var_328)[name = string("input_825_cast_fp16")];
+            tensor<int32, [6]> input_827_pad_0 = const()[name = string("input_827_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_827_mode_0 = const()[name = string("input_827_mode_0"), val = string("constant")];
+            fp16 const_167_to_fp16 = const()[name = string("const_167_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 136]> input_827_cast_fp16 = pad(constant_val = const_167_to_fp16, mode = input_827_mode_0, pad = input_827_pad_0, x = input_825_cast_fp16)[name = string("input_827_cast_fp16")];
+            string input_829_pad_type_0 = const()[name = string("input_829_pad_type_0"), val = string("valid")];
+            int32 input_829_groups_0 = const()[name = string("input_829_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_829_strides_0 = const()[name = string("input_829_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_829_pad_0 = const()[name = string("input_829_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_829_dilations_0 = const()[name = string("input_829_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_278_to_fp16 = const()[name = string("const_278_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202511616)))];
+            tensor<fp16, [1024]> const_279_to_fp16 = const()[name = string("const_279_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202530112)))];
+            tensor<fp16, [1, 1024, 128]> input_831_cast_fp16 = conv(bias = const_279_to_fp16, dilations = input_829_dilations_0, groups = input_829_groups_0, pad = input_829_pad_0, pad_type = input_829_pad_type_0, strides = input_829_strides_0, weight = const_278_to_fp16, x = input_827_cast_fp16)[name = string("input_831_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_833_cast_fp16 = silu(x = input_831_cast_fp16)[name = string("input_833_cast_fp16")];
+            string x_351_pad_type_0 = const()[name = string("x_351_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_351_strides_0 = const()[name = string("x_351_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_351_pad_0 = const()[name = string("x_351_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_351_dilations_0 = const()[name = string("x_351_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_351_groups_0 = const()[name = string("x_351_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524462]> module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202532224))), indices_nonzero_data = tensor<uint6, [524462]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202663360))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203056832))))[name = string("module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 128]> x_351_cast_fp16 = conv(dilations = x_351_dilations_0, groups = x_351_groups_0, pad = x_351_pad_0, pad_type = x_351_pad_type_0, strides = x_351_strides_0, weight = module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_833_cast_fp16)[name = string("x_351_cast_fp16")];
+            tensor<int32, [3]> input_835_perm_0 = const()[name = string("input_835_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 128, 1024]> input_835_cast_fp16 = transpose(perm = input_835_perm_0, x = x_351_cast_fp16)[name = string("transpose_201")];
+            tensor<fp16, [1, 128, 1024]> input_837_cast_fp16 = add(x = input_819_cast_fp16, y = input_835_cast_fp16)[name = string("input_837_cast_fp16")];
+            tensor<int32, [1]> input_839_axes_0 = const()[name = string("input_839_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_15_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_15_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203073280)))];
+            tensor<fp16, [1024]> module_layers_15_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_15_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203075392)))];
+            tensor<fp16, [1, 128, 1024]> input_839_cast_fp16 = layer_norm(axes = input_839_axes_0, beta = module_layers_15_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_feed_forward2_weight_to_fp16, x = input_837_cast_fp16)[name = string("input_839_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2078588]> module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203077504))), indices_nonzero_data = tensor<uint6, [2078588]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203601856))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205160896))))[name = string("module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 4096]> linear_143_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_839_cast_fp16)[name = string("linear_143_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_843_cast_fp16 = silu(x = linear_143_cast_fp16)[name = string("input_843_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1967312]> module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205226496))), indices_nonzero_data = tensor<uint6, [1967312]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205750848))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207226432))))[name = string("module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_144_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_843_cast_fp16)[name = string("linear_144_cast_fp16")];
+            fp16 var_2811_to_fp16 = const()[name = string("op_2811_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 128, 1024]> var_2812_cast_fp16 = mul(x = linear_144_cast_fp16, y = var_2811_to_fp16)[name = string("op_2812_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_849_cast_fp16 = add(x = input_837_cast_fp16, y = var_2812_cast_fp16)[name = string("input_849_cast_fp16")];
+            tensor<int32, [1]> input_851_axes_0 = const()[name = string("input_851_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_15_norm_out_weight_to_fp16 = const()[name = string("module_layers_15_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207242880)))];
+            tensor<fp16, [1024]> module_layers_15_norm_out_bias_to_fp16 = const()[name = string("module_layers_15_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207244992)))];
+            tensor<fp16, [1, 128, 1024]> input_851_cast_fp16 = layer_norm(axes = input_851_axes_0, beta = module_layers_15_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_out_weight_to_fp16, x = input_849_cast_fp16)[name = string("input_851_cast_fp16")];
+            tensor<int32, [1]> input_853_axes_0 = const()[name = string("input_853_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_16_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_16_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207247104)))];
+            tensor<fp16, [1024]> module_layers_16_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_16_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207249216)))];
+            tensor<fp16, [1, 128, 1024]> input_853_cast_fp16 = layer_norm(axes = input_853_axes_0, beta = module_layers_16_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_feed_forward1_weight_to_fp16, x = input_851_cast_fp16)[name = string("input_853_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2069902]> module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207251328))), indices_nonzero_data = tensor<uint6, [2069902]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207775680))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209328192))))[name = string("module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 4096]> linear_145_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_853_cast_fp16)[name = string("linear_145_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_857_cast_fp16 = silu(x = linear_145_cast_fp16)[name = string("input_857_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1927421]> module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209393792))), indices_nonzero_data = tensor<uint6, [1927421]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209918144))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211363776))))[name = string("module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_146_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_857_cast_fp16)[name = string("linear_146_cast_fp16")];
+            fp16 var_2840_to_fp16 = const()[name = string("op_2840_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 128, 1024]> var_2841_cast_fp16 = mul(x = linear_146_cast_fp16, y = var_2840_to_fp16)[name = string("op_2841_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_863_cast_fp16 = add(x = input_851_cast_fp16, y = var_2841_cast_fp16)[name = string("input_863_cast_fp16")];
+            tensor<int32, [1]> query_33_axes_0 = const()[name = string("query_33_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_16_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_16_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211380224)))];
+            tensor<fp16, [1024]> module_layers_16_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_16_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211382336)))];
+            tensor<fp16, [1, 128, 1024]> query_33_cast_fp16 = layer_norm(axes = query_33_axes_0, beta = module_layers_16_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_self_att_weight_to_fp16, x = input_863_cast_fp16)[name = string("query_33_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524478]> module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211384448))), indices_nonzero_data = tensor<uint6, [524478]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211515584))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211909056))))[name = string("module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_147_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_33_cast_fp16)[name = string("linear_147_cast_fp16")];
+            tensor<int32, [4]> var_2857 = const()[name = string("op_2857"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> q_97_cast_fp16 = reshape(shape = var_2857, x = linear_147_cast_fp16)[name = string("q_97_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524779]> module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211925504))), indices_nonzero_data = tensor<uint6, [524779]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212056640))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212450304))))[name = string("module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_148_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_33_cast_fp16)[name = string("linear_148_cast_fp16")];
+            tensor<int32, [4]> var_2861 = const()[name = string("op_2861"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> k_65_cast_fp16 = reshape(shape = var_2861, x = linear_148_cast_fp16)[name = string("k_65_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525290]> module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212466752))), indices_nonzero_data = tensor<uint6, [525290]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212597888))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212991936))))[name = string("module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_149_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_33_cast_fp16)[name = string("linear_149_cast_fp16")];
+            tensor<int32, [4]> var_2865 = const()[name = string("op_2865"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> v_33_cast_fp16 = reshape(shape = var_2865, x = linear_149_cast_fp16)[name = string("v_33_cast_fp16")];
+            tensor<int32, [4]> value_35_perm_0 = const()[name = string("value_35_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_16_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_16_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213008384)))];
+            tensor<fp16, [1, 128, 8, 128]> var_2877_cast_fp16 = add(x = q_97_cast_fp16, y = module_layers_16_self_attn_pos_bias_u_to_fp16)[name = string("op_2877_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_16_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_16_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213010496)))];
+            tensor<fp16, [1, 128, 8, 128]> var_2879_cast_fp16 = add(x = q_97_cast_fp16, y = module_layers_16_self_attn_pos_bias_v_to_fp16)[name = string("op_2879_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_33_perm_0 = const()[name = string("q_with_bias_v_33_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_359_transpose_x_0 = const()[name = string("x_359_transpose_x_0"), val = bool(false)];
+            bool x_359_transpose_y_0 = const()[name = string("x_359_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 255]> var_2881_to_fp16 = const()[name = string("op_2881_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315778688)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_33_cast_fp16 = transpose(perm = q_with_bias_v_33_perm_0, x = var_2879_cast_fp16)[name = string("transpose_200")];
+            tensor<fp16, [1, 8, 128, 255]> x_359_cast_fp16 = matmul(transpose_x = x_359_transpose_x_0, transpose_y = x_359_transpose_y_0, x = q_with_bias_v_33_cast_fp16, y = var_2881_to_fp16)[name = string("x_359_cast_fp16")];
+            tensor<int32, [8]> x_361_pad_0 = const()[name = string("x_361_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_361_mode_0 = const()[name = string("x_361_mode_0"), val = string("constant")];
+            fp16 const_174_to_fp16 = const()[name = string("const_174_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 128, 256]> x_361_cast_fp16 = pad(constant_val = const_174_to_fp16, mode = x_361_mode_0, pad = x_361_pad_0, x = x_359_cast_fp16)[name = string("x_361_cast_fp16")];
+            tensor<int32, [4]> var_2889 = const()[name = string("op_2889"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_363_cast_fp16 = reshape(shape = var_2889, x = x_361_cast_fp16)[name = string("x_363_cast_fp16")];
+            tensor<int32, [4]> var_2893_begin_0 = const()[name = string("op_2893_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2893_end_0 = const()[name = string("op_2893_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
+            tensor<bool, [4]> var_2893_end_mask_0 = const()[name = string("op_2893_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 255, 128]> var_2893_cast_fp16 = slice_by_index(begin = var_2893_begin_0, end = var_2893_end_0, end_mask = var_2893_end_mask_0, x = x_363_cast_fp16)[name = string("op_2893_cast_fp16")];
+            tensor<int32, [4]> var_2894 = const()[name = string("op_2894"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_65_cast_fp16 = reshape(shape = var_2894, x = var_2893_cast_fp16)[name = string("matrix_bd_65_cast_fp16")];
+            bool matrix_ac_33_transpose_x_0 = const()[name = string("matrix_ac_33_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_33_transpose_y_0 = const()[name = string("matrix_ac_33_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_128_perm_0 = const()[name = string("transpose_128_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_129_perm_0 = const()[name = string("transpose_129_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 128]> transpose_129 = transpose(perm = transpose_129_perm_0, x = k_65_cast_fp16)[name = string("transpose_198")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_128 = transpose(perm = transpose_128_perm_0, x = var_2877_cast_fp16)[name = string("transpose_199")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_33_cast_fp16 = matmul(transpose_x = matrix_ac_33_transpose_x_0, transpose_y = matrix_ac_33_transpose_y_0, x = transpose_128, y = transpose_129)[name = string("matrix_ac_33_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_67_begin_0 = const()[name = string("matrix_bd_67_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_67_end_0 = const()[name = string("matrix_bd_67_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
+            tensor<bool, [4]> matrix_bd_67_end_mask_0 = const()[name = string("matrix_bd_67_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_67_cast_fp16 = slice_by_index(begin = matrix_bd_67_begin_0, end = matrix_bd_67_end_0, end_mask = matrix_bd_67_end_mask_0, x = matrix_bd_65_cast_fp16)[name = string("matrix_bd_67_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_2903_cast_fp16 = add(x = matrix_ac_33_cast_fp16, y = matrix_bd_67_cast_fp16)[name = string("op_2903_cast_fp16")];
+            fp16 _inversed_scores_65_y_0_to_fp16 = const()[name = string("_inversed_scores_65_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_65_cast_fp16 = mul(x = var_2903_cast_fp16, y = _inversed_scores_65_y_0_to_fp16)[name = string("_inversed_scores_65_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_67_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_65_cast_fp16, cond = mask_3)[name = string("scores_67_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_2909_cast_fp16 = softmax(axis = var_30, x = scores_67_cast_fp16)[name = string("op_2909_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_865_cast_fp16 = select(a = var_11_to_fp16, b = var_2909_cast_fp16, cond = mask_3)[name = string("input_865_cast_fp16")];
+            bool x_365_transpose_x_0 = const()[name = string("x_365_transpose_x_0"), val = bool(false)];
+            bool x_365_transpose_y_0 = const()[name = string("x_365_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 128]> value_35_cast_fp16 = transpose(perm = value_35_perm_0, x = v_33_cast_fp16)[name = string("transpose_197")];
+            tensor<fp16, [1, 8, 128, 128]> x_365_cast_fp16 = matmul(transpose_x = x_365_transpose_x_0, transpose_y = x_365_transpose_y_0, x = input_865_cast_fp16, y = value_35_cast_fp16)[name = string("x_365_cast_fp16")];
+            tensor<int32, [4]> var_2913_perm_0 = const()[name = string("op_2913_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2914 = const()[name = string("op_2914"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 128, 8, 128]> var_2913_cast_fp16 = transpose(perm = var_2913_perm_0, x = x_365_cast_fp16)[name = string("transpose_196")];
+            tensor<fp16, [1, 128, 1024]> input_867_cast_fp16 = reshape(shape = var_2914, x = var_2913_cast_fp16)[name = string("input_867_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523537]> module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213272768))), indices_nonzero_data = tensor<uint6, [523537]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213403904))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213796672))))[name = string("module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_151_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_867_cast_fp16)[name = string("linear_151_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_871_cast_fp16 = add(x = input_863_cast_fp16, y = linear_151_cast_fp16)[name = string("input_871_cast_fp16")];
+            tensor<int32, [1]> x_369_axes_0 = const()[name = string("x_369_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_16_norm_conv_weight_to_fp16 = const()[name = string("module_layers_16_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213813120)))];
+            tensor<fp16, [1024]> module_layers_16_norm_conv_bias_to_fp16 = const()[name = string("module_layers_16_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213815232)))];
+            tensor<fp16, [1, 128, 1024]> x_369_cast_fp16 = layer_norm(axes = x_369_axes_0, beta = module_layers_16_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_conv_weight_to_fp16, x = input_871_cast_fp16)[name = string("x_369_cast_fp16")];
+            tensor<int32, [3]> input_873_perm_0 = const()[name = string("input_873_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_875_pad_type_0 = const()[name = string("input_875_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_875_strides_0 = const()[name = string("input_875_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_875_pad_0 = const()[name = string("input_875_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_875_dilations_0 = const()[name = string("input_875_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_875_groups_0 = const()[name = string("input_875_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047665]> module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213817344))), indices_nonzero_data = tensor<uint6, [1047665]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214079552))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214865408))))[name = string("module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 128]> input_873_cast_fp16 = transpose(perm = input_873_perm_0, x = x_369_cast_fp16)[name = string("transpose_195")];
+            tensor<fp16, [1, 2048, 128]> input_875_cast_fp16 = conv(dilations = input_875_dilations_0, groups = input_875_groups_0, pad = input_875_pad_0, pad_type = input_875_pad_type_0, strides = input_875_strides_0, weight = module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_873_cast_fp16)[name = string("input_875_cast_fp16")];
+            int32 x_371_split_num_splits_0 = const()[name = string("x_371_split_num_splits_0"), val = int32(2)];
+            int32 x_371_split_axis_0 = const()[name = string("x_371_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 128]> x_371_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_371_split_cast_fp16_1 = split(axis = x_371_split_axis_0, num_splits = x_371_split_num_splits_0, x = input_875_cast_fp16)[name = string("x_371_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_371_split_1_sigmoid_cast_fp16 = sigmoid(x = x_371_split_cast_fp16_1)[name = string("x_371_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_371_cast_fp16 = mul(x = x_371_split_cast_fp16_0, y = x_371_split_1_sigmoid_cast_fp16)[name = string("x_371_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_877_cast_fp16 = select(a = var_11_to_fp16, b = x_371_cast_fp16, cond = var_328)[name = string("input_877_cast_fp16")];
+            tensor<int32, [6]> input_879_pad_0 = const()[name = string("input_879_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_879_mode_0 = const()[name = string("input_879_mode_0"), val = string("constant")];
+            fp16 const_177_to_fp16 = const()[name = string("const_177_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 136]> input_879_cast_fp16 = pad(constant_val = const_177_to_fp16, mode = input_879_mode_0, pad = input_879_pad_0, x = input_877_cast_fp16)[name = string("input_879_cast_fp16")];
+            string input_881_pad_type_0 = const()[name = string("input_881_pad_type_0"), val = string("valid")];
+            int32 input_881_groups_0 = const()[name = string("input_881_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_881_strides_0 = const()[name = string("input_881_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_881_pad_0 = const()[name = string("input_881_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_881_dilations_0 = const()[name = string("input_881_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_280_to_fp16 = const()[name = string("const_280_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214898240)))];
+            tensor<fp16, [1024]> const_281_to_fp16 = const()[name = string("const_281_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214916736)))];
+            tensor<fp16, [1, 1024, 128]> input_883_cast_fp16 = conv(bias = const_281_to_fp16, dilations = input_881_dilations_0, groups = input_881_groups_0, pad = input_881_pad_0, pad_type = input_881_pad_type_0, strides = input_881_strides_0, weight = const_280_to_fp16, x = input_879_cast_fp16)[name = string("input_883_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_885_cast_fp16 = silu(x = input_883_cast_fp16)[name = string("input_885_cast_fp16")];
+            string x_373_pad_type_0 = const()[name = string("x_373_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_373_strides_0 = const()[name = string("x_373_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_373_pad_0 = const()[name = string("x_373_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_373_dilations_0 = const()[name = string("x_373_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_373_groups_0 = const()[name = string("x_373_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523727]> module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214918848))), indices_nonzero_data = tensor<uint6, [523727]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215049984))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215442880))))[name = string("module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 128]> x_373_cast_fp16 = conv(dilations = x_373_dilations_0, groups = x_373_groups_0, pad = x_373_pad_0, pad_type = x_373_pad_type_0, strides = x_373_strides_0, weight = module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_885_cast_fp16)[name = string("x_373_cast_fp16")];
+            tensor<int32, [3]> input_887_perm_0 = const()[name = string("input_887_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 128, 1024]> input_887_cast_fp16 = transpose(perm = input_887_perm_0, x = x_373_cast_fp16)[name = string("transpose_194")];
+            tensor<fp16, [1, 128, 1024]> input_889_cast_fp16 = add(x = input_871_cast_fp16, y = input_887_cast_fp16)[name = string("input_889_cast_fp16")];
+            tensor<int32, [1]> input_891_axes_0 = const()[name = string("input_891_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_16_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_16_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215459328)))];
+            tensor<fp16, [1024]> module_layers_16_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_16_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215461440)))];
+            tensor<fp16, [1, 128, 1024]> input_891_cast_fp16 = layer_norm(axes = input_891_axes_0, beta = module_layers_16_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_feed_forward2_weight_to_fp16, x = input_889_cast_fp16)[name = string("input_891_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2085434]> module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215463552))), indices_nonzero_data = tensor<uint6, [2085434]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215987904))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217552064))))[name = string("module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 4096]> linear_152_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_891_cast_fp16)[name = string("linear_152_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_895_cast_fp16 = silu(x = linear_152_cast_fp16)[name = string("input_895_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2028445]> module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217617664))), indices_nonzero_data = tensor<uint6, [2028445]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218142016))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219663424))))[name = string("module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_153_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_895_cast_fp16)[name = string("linear_153_cast_fp16")];
+            fp16 var_2974_to_fp16 = const()[name = string("op_2974_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 128, 1024]> var_2975_cast_fp16 = mul(x = linear_153_cast_fp16, y = var_2974_to_fp16)[name = string("op_2975_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_901_cast_fp16 = add(x = input_889_cast_fp16, y = var_2975_cast_fp16)[name = string("input_901_cast_fp16")];
+            tensor<int32, [1]> input_903_axes_0 = const()[name = string("input_903_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_16_norm_out_weight_to_fp16 = const()[name = string("module_layers_16_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219679872)))];
+            tensor<fp16, [1024]> module_layers_16_norm_out_bias_to_fp16 = const()[name = string("module_layers_16_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219681984)))];
+            tensor<fp16, [1, 128, 1024]> input_903_cast_fp16 = layer_norm(axes = input_903_axes_0, beta = module_layers_16_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_out_weight_to_fp16, x = input_901_cast_fp16)[name = string("input_903_cast_fp16")];
+            tensor<int32, [1]> input_905_axes_0 = const()[name = string("input_905_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_17_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_17_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219684096)))];
+            tensor<fp16, [1024]> module_layers_17_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_17_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219686208)))];
+            tensor<fp16, [1, 128, 1024]> input_905_cast_fp16 = layer_norm(axes = input_905_axes_0, beta = module_layers_17_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_feed_forward1_weight_to_fp16, x = input_903_cast_fp16)[name = string("input_905_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2078697]> module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219688320))), indices_nonzero_data = tensor<uint6, [2078697]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220212672))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221771776))))[name = string("module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 4096]> linear_154_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_905_cast_fp16)[name = string("linear_154_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_909_cast_fp16 = silu(x = linear_154_cast_fp16)[name = string("input_909_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2015245]> module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221837376))), indices_nonzero_data = tensor<uint6, [2015245]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222361728))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223873280))))[name = string("module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_155_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_909_cast_fp16)[name = string("linear_155_cast_fp16")];
+            fp16 var_3003_to_fp16 = const()[name = string("op_3003_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 128, 1024]> var_3004_cast_fp16 = mul(x = linear_155_cast_fp16, y = var_3003_to_fp16)[name = string("op_3004_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_915_cast_fp16 = add(x = input_903_cast_fp16, y = var_3004_cast_fp16)[name = string("input_915_cast_fp16")];
+            tensor<int32, [1]> query_35_axes_0 = const()[name = string("query_35_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_17_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_17_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223889728)))];
+            tensor<fp16, [1024]> module_layers_17_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_17_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223891840)))];
+            tensor<fp16, [1, 128, 1024]> query_35_cast_fp16 = layer_norm(axes = query_35_axes_0, beta = module_layers_17_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_self_att_weight_to_fp16, x = input_915_cast_fp16)[name = string("query_35_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525433]> module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223893952))), indices_nonzero_data = tensor<uint6, [525433]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224025088))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224419264))))[name = string("module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_156_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_35_cast_fp16)[name = string("linear_156_cast_fp16")];
+            tensor<int32, [4]> var_3020 = const()[name = string("op_3020"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> q_103_cast_fp16 = reshape(shape = var_3020, x = linear_156_cast_fp16)[name = string("q_103_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521661]> module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224435712))), indices_nonzero_data = tensor<uint6, [521661]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224566848))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224958208))))[name = string("module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_157_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_35_cast_fp16)[name = string("linear_157_cast_fp16")];
+            tensor<int32, [4]> var_3024 = const()[name = string("op_3024"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> k_69_cast_fp16 = reshape(shape = var_3024, x = linear_157_cast_fp16)[name = string("k_69_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521931]> module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224974656))), indices_nonzero_data = tensor<uint6, [521931]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225105792))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225497344))))[name = string("module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_158_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_35_cast_fp16)[name = string("linear_158_cast_fp16")];
+            tensor<int32, [4]> var_3028 = const()[name = string("op_3028"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> v_35_cast_fp16 = reshape(shape = var_3028, x = linear_158_cast_fp16)[name = string("v_35_cast_fp16")];
+            tensor<int32, [4]> value_37_perm_0 = const()[name = string("value_37_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_17_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_17_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225513792)))];
+            tensor<fp16, [1, 128, 8, 128]> var_3040_cast_fp16 = add(x = q_103_cast_fp16, y = module_layers_17_self_attn_pos_bias_u_to_fp16)[name = string("op_3040_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_17_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_17_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225515904)))];
+            tensor<fp16, [1, 128, 8, 128]> var_3042_cast_fp16 = add(x = q_103_cast_fp16, y = module_layers_17_self_attn_pos_bias_v_to_fp16)[name = string("op_3042_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_35_perm_0 = const()[name = string("q_with_bias_v_35_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_381_transpose_x_0 = const()[name = string("x_381_transpose_x_0"), val = bool(false)];
+            bool x_381_transpose_y_0 = const()[name = string("x_381_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 255]> var_3044_to_fp16 = const()[name = string("op_3044_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316300992)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_35_cast_fp16 = transpose(perm = q_with_bias_v_35_perm_0, x = var_3042_cast_fp16)[name = string("transpose_193")];
+            tensor<fp16, [1, 8, 128, 255]> x_381_cast_fp16 = matmul(transpose_x = x_381_transpose_x_0, transpose_y = x_381_transpose_y_0, x = q_with_bias_v_35_cast_fp16, y = var_3044_to_fp16)[name = string("x_381_cast_fp16")];
+            tensor<int32, [8]> x_383_pad_0 = const()[name = string("x_383_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_383_mode_0 = const()[name = string("x_383_mode_0"), val = string("constant")];
+            fp16 const_184_to_fp16 = const()[name = string("const_184_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 128, 256]> x_383_cast_fp16 = pad(constant_val = const_184_to_fp16, mode = x_383_mode_0, pad = x_383_pad_0, x = x_381_cast_fp16)[name = string("x_383_cast_fp16")];
+            tensor<int32, [4]> var_3052 = const()[name = string("op_3052"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_385_cast_fp16 = reshape(shape = var_3052, x = x_383_cast_fp16)[name = string("x_385_cast_fp16")];
+            tensor<int32, [4]> var_3056_begin_0 = const()[name = string("op_3056_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3056_end_0 = const()[name = string("op_3056_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
+            tensor<bool, [4]> var_3056_end_mask_0 = const()[name = string("op_3056_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 255, 128]> var_3056_cast_fp16 = slice_by_index(begin = var_3056_begin_0, end = var_3056_end_0, end_mask = var_3056_end_mask_0, x = x_385_cast_fp16)[name = string("op_3056_cast_fp16")];
+            tensor<int32, [4]> var_3057 = const()[name = string("op_3057"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_69_cast_fp16 = reshape(shape = var_3057, x = var_3056_cast_fp16)[name = string("matrix_bd_69_cast_fp16")];
+            bool matrix_ac_35_transpose_x_0 = const()[name = string("matrix_ac_35_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_35_transpose_y_0 = const()[name = string("matrix_ac_35_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_130_perm_0 = const()[name = string("transpose_130_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_131_perm_0 = const()[name = string("transpose_131_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 128]> transpose_131 = transpose(perm = transpose_131_perm_0, x = k_69_cast_fp16)[name = string("transpose_191")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_130 = transpose(perm = transpose_130_perm_0, x = var_3040_cast_fp16)[name = string("transpose_192")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_35_cast_fp16 = matmul(transpose_x = matrix_ac_35_transpose_x_0, transpose_y = matrix_ac_35_transpose_y_0, x = transpose_130, y = transpose_131)[name = string("matrix_ac_35_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_71_begin_0 = const()[name = string("matrix_bd_71_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_71_end_0 = const()[name = string("matrix_bd_71_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
+            tensor<bool, [4]> matrix_bd_71_end_mask_0 = const()[name = string("matrix_bd_71_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_71_cast_fp16 = slice_by_index(begin = matrix_bd_71_begin_0, end = matrix_bd_71_end_0, end_mask = matrix_bd_71_end_mask_0, x = matrix_bd_69_cast_fp16)[name = string("matrix_bd_71_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_3066_cast_fp16 = add(x = matrix_ac_35_cast_fp16, y = matrix_bd_71_cast_fp16)[name = string("op_3066_cast_fp16")];
+            fp16 _inversed_scores_69_y_0_to_fp16 = const()[name = string("_inversed_scores_69_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_69_cast_fp16 = mul(x = var_3066_cast_fp16, y = _inversed_scores_69_y_0_to_fp16)[name = string("_inversed_scores_69_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_71_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_69_cast_fp16, cond = mask_3)[name = string("scores_71_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_3072_cast_fp16 = softmax(axis = var_30, x = scores_71_cast_fp16)[name = string("op_3072_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_917_cast_fp16 = select(a = var_11_to_fp16, b = var_3072_cast_fp16, cond = mask_3)[name = string("input_917_cast_fp16")];
+            bool x_387_transpose_x_0 = const()[name = string("x_387_transpose_x_0"), val = bool(false)];
+            bool x_387_transpose_y_0 = const()[name = string("x_387_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 128]> value_37_cast_fp16 = transpose(perm = value_37_perm_0, x = v_35_cast_fp16)[name = string("transpose_190")];
+            tensor<fp16, [1, 8, 128, 128]> x_387_cast_fp16 = matmul(transpose_x = x_387_transpose_x_0, transpose_y = x_387_transpose_y_0, x = input_917_cast_fp16, y = value_37_cast_fp16)[name = string("x_387_cast_fp16")];
+            tensor<int32, [4]> var_3076_perm_0 = const()[name = string("op_3076_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3077 = const()[name = string("op_3077"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 128, 8, 128]> var_3076_cast_fp16 = transpose(perm = var_3076_perm_0, x = x_387_cast_fp16)[name = string("transpose_189")];
+            tensor<fp16, [1, 128, 1024]> input_919_cast_fp16 = reshape(shape = var_3077, x = var_3076_cast_fp16)[name = string("input_919_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [518816]> module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225778176))), indices_nonzero_data = tensor<uint6, [518816]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225909312))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226298496))))[name = string("module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_160_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_919_cast_fp16)[name = string("linear_160_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_923_cast_fp16 = add(x = input_915_cast_fp16, y = linear_160_cast_fp16)[name = string("input_923_cast_fp16")];
+            tensor<int32, [1]> x_391_axes_0 = const()[name = string("x_391_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_17_norm_conv_weight_to_fp16 = const()[name = string("module_layers_17_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226314944)))];
+            tensor<fp16, [1024]> module_layers_17_norm_conv_bias_to_fp16 = const()[name = string("module_layers_17_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226317056)))];
+            tensor<fp16, [1, 128, 1024]> x_391_cast_fp16 = layer_norm(axes = x_391_axes_0, beta = module_layers_17_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_conv_weight_to_fp16, x = input_923_cast_fp16)[name = string("x_391_cast_fp16")];
+            tensor<int32, [3]> input_925_perm_0 = const()[name = string("input_925_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_927_pad_type_0 = const()[name = string("input_927_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_927_strides_0 = const()[name = string("input_927_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_927_pad_0 = const()[name = string("input_927_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_927_dilations_0 = const()[name = string("input_927_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_927_groups_0 = const()[name = string("input_927_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047749]> module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226319168))), indices_nonzero_data = tensor<uint6, [1047749]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226581376))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227367296))))[name = string("module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 128]> input_925_cast_fp16 = transpose(perm = input_925_perm_0, x = x_391_cast_fp16)[name = string("transpose_188")];
+            tensor<fp16, [1, 2048, 128]> input_927_cast_fp16 = conv(dilations = input_927_dilations_0, groups = input_927_groups_0, pad = input_927_pad_0, pad_type = input_927_pad_type_0, strides = input_927_strides_0, weight = module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_925_cast_fp16)[name = string("input_927_cast_fp16")];
+            int32 x_393_split_num_splits_0 = const()[name = string("x_393_split_num_splits_0"), val = int32(2)];
+            int32 x_393_split_axis_0 = const()[name = string("x_393_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 128]> x_393_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_393_split_cast_fp16_1 = split(axis = x_393_split_axis_0, num_splits = x_393_split_num_splits_0, x = input_927_cast_fp16)[name = string("x_393_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_393_split_1_sigmoid_cast_fp16 = sigmoid(x = x_393_split_cast_fp16_1)[name = string("x_393_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_393_cast_fp16 = mul(x = x_393_split_cast_fp16_0, y = x_393_split_1_sigmoid_cast_fp16)[name = string("x_393_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_929_cast_fp16 = select(a = var_11_to_fp16, b = x_393_cast_fp16, cond = var_328)[name = string("input_929_cast_fp16")];
+            tensor<int32, [6]> input_931_pad_0 = const()[name = string("input_931_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_931_mode_0 = const()[name = string("input_931_mode_0"), val = string("constant")];
+            fp16 const_187_to_fp16 = const()[name = string("const_187_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 136]> input_931_cast_fp16 = pad(constant_val = const_187_to_fp16, mode = input_931_mode_0, pad = input_931_pad_0, x = input_929_cast_fp16)[name = string("input_931_cast_fp16")];
+            string input_933_pad_type_0 = const()[name = string("input_933_pad_type_0"), val = string("valid")];
+            int32 input_933_groups_0 = const()[name = string("input_933_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_933_strides_0 = const()[name = string("input_933_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_933_pad_0 = const()[name = string("input_933_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_933_dilations_0 = const()[name = string("input_933_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_282_to_fp16 = const()[name = string("const_282_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227400128)))];
+            tensor<fp16, [1024]> const_283_to_fp16 = const()[name = string("const_283_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227418624)))];
+            tensor<fp16, [1, 1024, 128]> input_935_cast_fp16 = conv(bias = const_283_to_fp16, dilations = input_933_dilations_0, groups = input_933_groups_0, pad = input_933_pad_0, pad_type = input_933_pad_type_0, strides = input_933_strides_0, weight = const_282_to_fp16, x = input_931_cast_fp16)[name = string("input_935_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_937_cast_fp16 = silu(x = input_935_cast_fp16)[name = string("input_937_cast_fp16")];
+            string x_395_pad_type_0 = const()[name = string("x_395_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_395_strides_0 = const()[name = string("x_395_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_395_pad_0 = const()[name = string("x_395_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_395_dilations_0 = const()[name = string("x_395_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_395_groups_0 = const()[name = string("x_395_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521069]> module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227420736))), indices_nonzero_data = tensor<uint6, [521069]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227551872))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227942784))))[name = string("module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 128]> x_395_cast_fp16 = conv(dilations = x_395_dilations_0, groups = x_395_groups_0, pad = x_395_pad_0, pad_type = x_395_pad_type_0, strides = x_395_strides_0, weight = module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_937_cast_fp16)[name = string("x_395_cast_fp16")];
+            tensor<int32, [3]> input_939_perm_0 = const()[name = string("input_939_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 128, 1024]> input_939_cast_fp16 = transpose(perm = input_939_perm_0, x = x_395_cast_fp16)[name = string("transpose_187")];
+            tensor<fp16, [1, 128, 1024]> input_941_cast_fp16 = add(x = input_923_cast_fp16, y = input_939_cast_fp16)[name = string("input_941_cast_fp16")];
+            tensor<int32, [1]> input_943_axes_0 = const()[name = string("input_943_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_17_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_17_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227959232)))];
+            tensor<fp16, [1024]> module_layers_17_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_17_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227961344)))];
+            tensor<fp16, [1, 128, 1024]> input_943_cast_fp16 = layer_norm(axes = input_943_axes_0, beta = module_layers_17_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_feed_forward2_weight_to_fp16, x = input_941_cast_fp16)[name = string("input_943_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2085405]> module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227963456))), indices_nonzero_data = tensor<uint6, [2085405]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228487808))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230051968))))[name = string("module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 4096]> linear_161_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_943_cast_fp16)[name = string("linear_161_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_947_cast_fp16 = silu(x = linear_161_cast_fp16)[name = string("input_947_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2032104]> module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230117568))), indices_nonzero_data = tensor<uint6, [2032104]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230641920))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232166080))))[name = string("module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_162_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_947_cast_fp16)[name = string("linear_162_cast_fp16")];
+            fp16 var_3137_to_fp16 = const()[name = string("op_3137_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 128, 1024]> var_3138_cast_fp16 = mul(x = linear_162_cast_fp16, y = var_3137_to_fp16)[name = string("op_3138_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_953_cast_fp16 = add(x = input_941_cast_fp16, y = var_3138_cast_fp16)[name = string("input_953_cast_fp16")];
+            tensor<int32, [1]> input_955_axes_0 = const()[name = string("input_955_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_17_norm_out_weight_to_fp16 = const()[name = string("module_layers_17_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232182528)))];
+            tensor<fp16, [1024]> module_layers_17_norm_out_bias_to_fp16 = const()[name = string("module_layers_17_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232184640)))];
+            tensor<fp16, [1, 128, 1024]> input_955_cast_fp16 = layer_norm(axes = input_955_axes_0, beta = module_layers_17_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_out_weight_to_fp16, x = input_953_cast_fp16)[name = string("input_955_cast_fp16")];
+            tensor<int32, [1]> input_957_axes_0 = const()[name = string("input_957_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_18_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_18_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232186752)))];
+            tensor<fp16, [1024]> module_layers_18_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_18_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232188864)))];
+            tensor<fp16, [1, 128, 1024]> input_957_cast_fp16 = layer_norm(axes = input_957_axes_0, beta = module_layers_18_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_feed_forward1_weight_to_fp16, x = input_955_cast_fp16)[name = string("input_957_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2077029]> module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232190976))), indices_nonzero_data = tensor<uint6, [2077029]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232715328))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234273216))))[name = string("module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 4096]> linear_163_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_957_cast_fp16)[name = string("linear_163_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_961_cast_fp16 = silu(x = linear_163_cast_fp16)[name = string("input_961_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2028343]> module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234338816))), indices_nonzero_data = tensor<uint6, [2028343]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234863168))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236384512))))[name = string("module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_164_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_961_cast_fp16)[name = string("linear_164_cast_fp16")];
+            fp16 var_3166_to_fp16 = const()[name = string("op_3166_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 128, 1024]> var_3167_cast_fp16 = mul(x = linear_164_cast_fp16, y = var_3166_to_fp16)[name = string("op_3167_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_967_cast_fp16 = add(x = input_955_cast_fp16, y = var_3167_cast_fp16)[name = string("input_967_cast_fp16")];
+            tensor<int32, [1]> query_37_axes_0 = const()[name = string("query_37_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_18_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_18_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236400960)))];
+            tensor<fp16, [1024]> module_layers_18_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_18_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236403072)))];
+            tensor<fp16, [1, 128, 1024]> query_37_cast_fp16 = layer_norm(axes = query_37_axes_0, beta = module_layers_18_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_self_att_weight_to_fp16, x = input_967_cast_fp16)[name = string("query_37_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525392]> module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236405184))), indices_nonzero_data = tensor<uint6, [525392]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236536320))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236930432))))[name = string("module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_165_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_37_cast_fp16)[name = string("linear_165_cast_fp16")];
+            tensor<int32, [4]> var_3183 = const()[name = string("op_3183"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> q_109_cast_fp16 = reshape(shape = var_3183, x = linear_165_cast_fp16)[name = string("q_109_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524473]> module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236946880))), indices_nonzero_data = tensor<uint6, [524473]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237078016))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237471488))))[name = string("module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_166_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_37_cast_fp16)[name = string("linear_166_cast_fp16")];
+            tensor<int32, [4]> var_3187 = const()[name = string("op_3187"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> k_73_cast_fp16 = reshape(shape = var_3187, x = linear_166_cast_fp16)[name = string("k_73_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523564]> module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237487936))), indices_nonzero_data = tensor<uint6, [523564]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237619072))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238011840))))[name = string("module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_167_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_37_cast_fp16)[name = string("linear_167_cast_fp16")];
+            tensor<int32, [4]> var_3191 = const()[name = string("op_3191"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> v_37_cast_fp16 = reshape(shape = var_3191, x = linear_167_cast_fp16)[name = string("v_37_cast_fp16")];
+            tensor<int32, [4]> value_39_perm_0 = const()[name = string("value_39_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_18_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_18_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238028288)))];
+            tensor<fp16, [1, 128, 8, 128]> var_3203_cast_fp16 = add(x = q_109_cast_fp16, y = module_layers_18_self_attn_pos_bias_u_to_fp16)[name = string("op_3203_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_18_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_18_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238030400)))];
+            tensor<fp16, [1, 128, 8, 128]> var_3205_cast_fp16 = add(x = q_109_cast_fp16, y = module_layers_18_self_attn_pos_bias_v_to_fp16)[name = string("op_3205_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_37_perm_0 = const()[name = string("q_with_bias_v_37_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_403_transpose_x_0 = const()[name = string("x_403_transpose_x_0"), val = bool(false)];
+            bool x_403_transpose_y_0 = const()[name = string("x_403_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 255]> var_3207_to_fp16 = const()[name = string("op_3207_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316823296)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_37_cast_fp16 = transpose(perm = q_with_bias_v_37_perm_0, x = var_3205_cast_fp16)[name = string("transpose_186")];
+            tensor<fp16, [1, 8, 128, 255]> x_403_cast_fp16 = matmul(transpose_x = x_403_transpose_x_0, transpose_y = x_403_transpose_y_0, x = q_with_bias_v_37_cast_fp16, y = var_3207_to_fp16)[name = string("x_403_cast_fp16")];
+            tensor<int32, [8]> x_405_pad_0 = const()[name = string("x_405_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_405_mode_0 = const()[name = string("x_405_mode_0"), val = string("constant")];
+            fp16 const_194_to_fp16 = const()[name = string("const_194_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 128, 256]> x_405_cast_fp16 = pad(constant_val = const_194_to_fp16, mode = x_405_mode_0, pad = x_405_pad_0, x = x_403_cast_fp16)[name = string("x_405_cast_fp16")];
+            tensor<int32, [4]> var_3215 = const()[name = string("op_3215"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_407_cast_fp16 = reshape(shape = var_3215, x = x_405_cast_fp16)[name = string("x_407_cast_fp16")];
+            tensor<int32, [4]> var_3219_begin_0 = const()[name = string("op_3219_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3219_end_0 = const()[name = string("op_3219_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
+            tensor<bool, [4]> var_3219_end_mask_0 = const()[name = string("op_3219_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 255, 128]> var_3219_cast_fp16 = slice_by_index(begin = var_3219_begin_0, end = var_3219_end_0, end_mask = var_3219_end_mask_0, x = x_407_cast_fp16)[name = string("op_3219_cast_fp16")];
+            tensor<int32, [4]> var_3220 = const()[name = string("op_3220"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_73_cast_fp16 = reshape(shape = var_3220, x = var_3219_cast_fp16)[name = string("matrix_bd_73_cast_fp16")];
+            bool matrix_ac_37_transpose_x_0 = const()[name = string("matrix_ac_37_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_37_transpose_y_0 = const()[name = string("matrix_ac_37_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_132_perm_0 = const()[name = string("transpose_132_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_133_perm_0 = const()[name = string("transpose_133_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 128]> transpose_133 = transpose(perm = transpose_133_perm_0, x = k_73_cast_fp16)[name = string("transpose_184")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_132 = transpose(perm = transpose_132_perm_0, x = var_3203_cast_fp16)[name = string("transpose_185")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_37_cast_fp16 = matmul(transpose_x = matrix_ac_37_transpose_x_0, transpose_y = matrix_ac_37_transpose_y_0, x = transpose_132, y = transpose_133)[name = string("matrix_ac_37_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_75_begin_0 = const()[name = string("matrix_bd_75_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_75_end_0 = const()[name = string("matrix_bd_75_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
+            tensor<bool, [4]> matrix_bd_75_end_mask_0 = const()[name = string("matrix_bd_75_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_75_cast_fp16 = slice_by_index(begin = matrix_bd_75_begin_0, end = matrix_bd_75_end_0, end_mask = matrix_bd_75_end_mask_0, x = matrix_bd_73_cast_fp16)[name = string("matrix_bd_75_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_3229_cast_fp16 = add(x = matrix_ac_37_cast_fp16, y = matrix_bd_75_cast_fp16)[name = string("op_3229_cast_fp16")];
+            fp16 _inversed_scores_73_y_0_to_fp16 = const()[name = string("_inversed_scores_73_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_73_cast_fp16 = mul(x = var_3229_cast_fp16, y = _inversed_scores_73_y_0_to_fp16)[name = string("_inversed_scores_73_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_75_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_73_cast_fp16, cond = mask_3)[name = string("scores_75_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_3235_cast_fp16 = softmax(axis = var_30, x = scores_75_cast_fp16)[name = string("op_3235_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_969_cast_fp16 = select(a = var_11_to_fp16, b = var_3235_cast_fp16, cond = mask_3)[name = string("input_969_cast_fp16")];
+            bool x_409_transpose_x_0 = const()[name = string("x_409_transpose_x_0"), val = bool(false)];
+            bool x_409_transpose_y_0 = const()[name = string("x_409_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 128]> value_39_cast_fp16 = transpose(perm = value_39_perm_0, x = v_37_cast_fp16)[name = string("transpose_183")];
+            tensor<fp16, [1, 8, 128, 128]> x_409_cast_fp16 = matmul(transpose_x = x_409_transpose_x_0, transpose_y = x_409_transpose_y_0, x = input_969_cast_fp16, y = value_39_cast_fp16)[name = string("x_409_cast_fp16")];
+            tensor<int32, [4]> var_3239_perm_0 = const()[name = string("op_3239_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3240 = const()[name = string("op_3240"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 128, 8, 128]> var_3239_cast_fp16 = transpose(perm = var_3239_perm_0, x = x_409_cast_fp16)[name = string("transpose_182")];
+            tensor<fp16, [1, 128, 1024]> input_971_cast_fp16 = reshape(shape = var_3240, x = var_3239_cast_fp16)[name = string("input_971_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520692]> module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238292672))), indices_nonzero_data = tensor<uint6, [520692]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238423808))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238814400))))[name = string("module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_169_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_971_cast_fp16)[name = string("linear_169_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_975_cast_fp16 = add(x = input_967_cast_fp16, y = linear_169_cast_fp16)[name = string("input_975_cast_fp16")];
+            tensor<int32, [1]> x_413_axes_0 = const()[name = string("x_413_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_18_norm_conv_weight_to_fp16 = const()[name = string("module_layers_18_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238830848)))];
+            tensor<fp16, [1024]> module_layers_18_norm_conv_bias_to_fp16 = const()[name = string("module_layers_18_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238832960)))];
+            tensor<fp16, [1, 128, 1024]> x_413_cast_fp16 = layer_norm(axes = x_413_axes_0, beta = module_layers_18_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_conv_weight_to_fp16, x = input_975_cast_fp16)[name = string("x_413_cast_fp16")];
+            tensor<int32, [3]> input_977_perm_0 = const()[name = string("input_977_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_979_pad_type_0 = const()[name = string("input_979_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_979_strides_0 = const()[name = string("input_979_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_979_pad_0 = const()[name = string("input_979_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_979_dilations_0 = const()[name = string("input_979_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_979_groups_0 = const()[name = string("input_979_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047234]> module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238835072))), indices_nonzero_data = tensor<uint6, [1047234]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239097280))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239882816))))[name = string("module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 128]> input_977_cast_fp16 = transpose(perm = input_977_perm_0, x = x_413_cast_fp16)[name = string("transpose_181")];
+            tensor<fp16, [1, 2048, 128]> input_979_cast_fp16 = conv(dilations = input_979_dilations_0, groups = input_979_groups_0, pad = input_979_pad_0, pad_type = input_979_pad_type_0, strides = input_979_strides_0, weight = module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_977_cast_fp16)[name = string("input_979_cast_fp16")];
+            int32 x_415_split_num_splits_0 = const()[name = string("x_415_split_num_splits_0"), val = int32(2)];
+            int32 x_415_split_axis_0 = const()[name = string("x_415_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 128]> x_415_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_415_split_cast_fp16_1 = split(axis = x_415_split_axis_0, num_splits = x_415_split_num_splits_0, x = input_979_cast_fp16)[name = string("x_415_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_415_split_1_sigmoid_cast_fp16 = sigmoid(x = x_415_split_cast_fp16_1)[name = string("x_415_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_415_cast_fp16 = mul(x = x_415_split_cast_fp16_0, y = x_415_split_1_sigmoid_cast_fp16)[name = string("x_415_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_981_cast_fp16 = select(a = var_11_to_fp16, b = x_415_cast_fp16, cond = var_328)[name = string("input_981_cast_fp16")];
+            tensor<int32, [6]> input_983_pad_0 = const()[name = string("input_983_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_983_mode_0 = const()[name = string("input_983_mode_0"), val = string("constant")];
+            fp16 const_197_to_fp16 = const()[name = string("const_197_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 136]> input_983_cast_fp16 = pad(constant_val = const_197_to_fp16, mode = input_983_mode_0, pad = input_983_pad_0, x = input_981_cast_fp16)[name = string("input_983_cast_fp16")];
+            string input_985_pad_type_0 = const()[name = string("input_985_pad_type_0"), val = string("valid")];
+            int32 input_985_groups_0 = const()[name = string("input_985_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_985_strides_0 = const()[name = string("input_985_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_985_pad_0 = const()[name = string("input_985_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_985_dilations_0 = const()[name = string("input_985_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_284_to_fp16 = const()[name = string("const_284_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239915648)))];
+            tensor<fp16, [1024]> const_285_to_fp16 = const()[name = string("const_285_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239934144)))];
+            tensor<fp16, [1, 1024, 128]> input_987_cast_fp16 = conv(bias = const_285_to_fp16, dilations = input_985_dilations_0, groups = input_985_groups_0, pad = input_985_pad_0, pad_type = input_985_pad_type_0, strides = input_985_strides_0, weight = const_284_to_fp16, x = input_983_cast_fp16)[name = string("input_987_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_989_cast_fp16 = silu(x = input_987_cast_fp16)[name = string("input_989_cast_fp16")];
+            string x_417_pad_type_0 = const()[name = string("x_417_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_417_strides_0 = const()[name = string("x_417_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_417_pad_0 = const()[name = string("x_417_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_417_dilations_0 = const()[name = string("x_417_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_417_groups_0 = const()[name = string("x_417_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521466]> module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239936256))), indices_nonzero_data = tensor<uint6, [521466]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240067392))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240458560))))[name = string("module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 128]> x_417_cast_fp16 = conv(dilations = x_417_dilations_0, groups = x_417_groups_0, pad = x_417_pad_0, pad_type = x_417_pad_type_0, strides = x_417_strides_0, weight = module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_989_cast_fp16)[name = string("x_417_cast_fp16")];
+            tensor<int32, [3]> input_991_perm_0 = const()[name = string("input_991_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 128, 1024]> input_991_cast_fp16 = transpose(perm = input_991_perm_0, x = x_417_cast_fp16)[name = string("transpose_180")];
+            tensor<fp16, [1, 128, 1024]> input_993_cast_fp16 = add(x = input_975_cast_fp16, y = input_991_cast_fp16)[name = string("input_993_cast_fp16")];
+            tensor<int32, [1]> input_995_axes_0 = const()[name = string("input_995_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_18_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_18_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240475008)))];
+            tensor<fp16, [1024]> module_layers_18_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_18_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240477120)))];
+            tensor<fp16, [1, 128, 1024]> input_995_cast_fp16 = layer_norm(axes = input_995_axes_0, beta = module_layers_18_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_feed_forward2_weight_to_fp16, x = input_993_cast_fp16)[name = string("input_995_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2081130]> module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240479232))), indices_nonzero_data = tensor<uint6, [2081130]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241003584))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242564544))))[name = string("module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 4096]> linear_170_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_995_cast_fp16)[name = string("linear_170_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_999_cast_fp16 = silu(x = linear_170_cast_fp16)[name = string("input_999_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2068517]> module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242630144))), indices_nonzero_data = tensor<uint6, [2068517]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243154496))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244705984))))[name = string("module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_171_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_999_cast_fp16)[name = string("linear_171_cast_fp16")];
+            fp16 var_3300_to_fp16 = const()[name = string("op_3300_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 128, 1024]> var_3301_cast_fp16 = mul(x = linear_171_cast_fp16, y = var_3300_to_fp16)[name = string("op_3301_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_1005_cast_fp16 = add(x = input_993_cast_fp16, y = var_3301_cast_fp16)[name = string("input_1005_cast_fp16")];
+            tensor<int32, [1]> input_1007_axes_0 = const()[name = string("input_1007_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_18_norm_out_weight_to_fp16 = const()[name = string("module_layers_18_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244722432)))];
+            tensor<fp16, [1024]> module_layers_18_norm_out_bias_to_fp16 = const()[name = string("module_layers_18_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244724544)))];
+            tensor<fp16, [1, 128, 1024]> input_1007_cast_fp16 = layer_norm(axes = input_1007_axes_0, beta = module_layers_18_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_out_weight_to_fp16, x = input_1005_cast_fp16)[name = string("input_1007_cast_fp16")];
+            tensor<int32, [1]> input_1009_axes_0 = const()[name = string("input_1009_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_19_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_19_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244726656)))];
+            tensor<fp16, [1024]> module_layers_19_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_19_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244728768)))];
+            tensor<fp16, [1, 128, 1024]> input_1009_cast_fp16 = layer_norm(axes = input_1009_axes_0, beta = module_layers_19_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_feed_forward1_weight_to_fp16, x = input_1007_cast_fp16)[name = string("input_1009_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2077149]> module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244730880))), indices_nonzero_data = tensor<uint6, [2077149]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245255232))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246813184))))[name = string("module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 4096]> linear_172_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_1009_cast_fp16)[name = string("linear_172_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_1013_cast_fp16 = silu(x = linear_172_cast_fp16)[name = string("input_1013_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2050240]> module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246878784))), indices_nonzero_data = tensor<uint6, [2050240]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247403136))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248940928))))[name = string("module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_173_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_1013_cast_fp16)[name = string("linear_173_cast_fp16")];
+            fp16 var_3329_to_fp16 = const()[name = string("op_3329_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 128, 1024]> var_3330_cast_fp16 = mul(x = linear_173_cast_fp16, y = var_3329_to_fp16)[name = string("op_3330_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_1019_cast_fp16 = add(x = input_1007_cast_fp16, y = var_3330_cast_fp16)[name = string("input_1019_cast_fp16")];
+            tensor<int32, [1]> query_39_axes_0 = const()[name = string("query_39_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_19_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_19_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248957376)))];
+            tensor<fp16, [1024]> module_layers_19_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_19_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248959488)))];
+            tensor<fp16, [1, 128, 1024]> query_39_cast_fp16 = layer_norm(axes = query_39_axes_0, beta = module_layers_19_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_self_att_weight_to_fp16, x = input_1019_cast_fp16)[name = string("query_39_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524663]> module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248961600))), indices_nonzero_data = tensor<uint6, [524663]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249092736))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249486336))))[name = string("module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_174_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_39_cast_fp16)[name = string("linear_174_cast_fp16")];
+            tensor<int32, [4]> var_3346 = const()[name = string("op_3346"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> q_115_cast_fp16 = reshape(shape = var_3346, x = linear_174_cast_fp16)[name = string("q_115_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525503]> module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249502784))), indices_nonzero_data = tensor<uint6, [525503]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249633920))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250028160))))[name = string("module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_175_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_39_cast_fp16)[name = string("linear_175_cast_fp16")];
+            tensor<int32, [4]> var_3350 = const()[name = string("op_3350"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> k_77_cast_fp16 = reshape(shape = var_3350, x = linear_175_cast_fp16)[name = string("k_77_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523873]> module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250044608))), indices_nonzero_data = tensor<uint6, [523873]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250175744))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250568768))))[name = string("module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_176_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_39_cast_fp16)[name = string("linear_176_cast_fp16")];
+            tensor<int32, [4]> var_3354 = const()[name = string("op_3354"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> v_39_cast_fp16 = reshape(shape = var_3354, x = linear_176_cast_fp16)[name = string("v_39_cast_fp16")];
+            tensor<int32, [4]> value_41_perm_0 = const()[name = string("value_41_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_19_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_19_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250585216)))];
+            tensor<fp16, [1, 128, 8, 128]> var_3366_cast_fp16 = add(x = q_115_cast_fp16, y = module_layers_19_self_attn_pos_bias_u_to_fp16)[name = string("op_3366_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_19_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_19_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250587328)))];
+            tensor<fp16, [1, 128, 8, 128]> var_3368_cast_fp16 = add(x = q_115_cast_fp16, y = module_layers_19_self_attn_pos_bias_v_to_fp16)[name = string("op_3368_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_39_perm_0 = const()[name = string("q_with_bias_v_39_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_425_transpose_x_0 = const()[name = string("x_425_transpose_x_0"), val = bool(false)];
+            bool x_425_transpose_y_0 = const()[name = string("x_425_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 255]> var_3370_to_fp16 = const()[name = string("op_3370_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(317345600)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_39_cast_fp16 = transpose(perm = q_with_bias_v_39_perm_0, x = var_3368_cast_fp16)[name = string("transpose_179")];
+            tensor<fp16, [1, 8, 128, 255]> x_425_cast_fp16 = matmul(transpose_x = x_425_transpose_x_0, transpose_y = x_425_transpose_y_0, x = q_with_bias_v_39_cast_fp16, y = var_3370_to_fp16)[name = string("x_425_cast_fp16")];
+            tensor<int32, [8]> x_427_pad_0 = const()[name = string("x_427_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_427_mode_0 = const()[name = string("x_427_mode_0"), val = string("constant")];
+            fp16 const_204_to_fp16 = const()[name = string("const_204_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 128, 256]> x_427_cast_fp16 = pad(constant_val = const_204_to_fp16, mode = x_427_mode_0, pad = x_427_pad_0, x = x_425_cast_fp16)[name = string("x_427_cast_fp16")];
+            tensor<int32, [4]> var_3378 = const()[name = string("op_3378"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_429_cast_fp16 = reshape(shape = var_3378, x = x_427_cast_fp16)[name = string("x_429_cast_fp16")];
+            tensor<int32, [4]> var_3382_begin_0 = const()[name = string("op_3382_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3382_end_0 = const()[name = string("op_3382_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
+            tensor<bool, [4]> var_3382_end_mask_0 = const()[name = string("op_3382_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 255, 128]> var_3382_cast_fp16 = slice_by_index(begin = var_3382_begin_0, end = var_3382_end_0, end_mask = var_3382_end_mask_0, x = x_429_cast_fp16)[name = string("op_3382_cast_fp16")];
+            tensor<int32, [4]> var_3383 = const()[name = string("op_3383"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_77_cast_fp16 = reshape(shape = var_3383, x = var_3382_cast_fp16)[name = string("matrix_bd_77_cast_fp16")];
+            bool matrix_ac_39_transpose_x_0 = const()[name = string("matrix_ac_39_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_39_transpose_y_0 = const()[name = string("matrix_ac_39_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_134_perm_0 = const()[name = string("transpose_134_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_135_perm_0 = const()[name = string("transpose_135_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 128]> transpose_135 = transpose(perm = transpose_135_perm_0, x = k_77_cast_fp16)[name = string("transpose_177")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_134 = transpose(perm = transpose_134_perm_0, x = var_3366_cast_fp16)[name = string("transpose_178")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_39_cast_fp16 = matmul(transpose_x = matrix_ac_39_transpose_x_0, transpose_y = matrix_ac_39_transpose_y_0, x = transpose_134, y = transpose_135)[name = string("matrix_ac_39_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_79_begin_0 = const()[name = string("matrix_bd_79_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_79_end_0 = const()[name = string("matrix_bd_79_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
+            tensor<bool, [4]> matrix_bd_79_end_mask_0 = const()[name = string("matrix_bd_79_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_79_cast_fp16 = slice_by_index(begin = matrix_bd_79_begin_0, end = matrix_bd_79_end_0, end_mask = matrix_bd_79_end_mask_0, x = matrix_bd_77_cast_fp16)[name = string("matrix_bd_79_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_3392_cast_fp16 = add(x = matrix_ac_39_cast_fp16, y = matrix_bd_79_cast_fp16)[name = string("op_3392_cast_fp16")];
+            fp16 _inversed_scores_77_y_0_to_fp16 = const()[name = string("_inversed_scores_77_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_77_cast_fp16 = mul(x = var_3392_cast_fp16, y = _inversed_scores_77_y_0_to_fp16)[name = string("_inversed_scores_77_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_79_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_77_cast_fp16, cond = mask_3)[name = string("scores_79_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_3398_cast_fp16 = softmax(axis = var_30, x = scores_79_cast_fp16)[name = string("op_3398_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_1021_cast_fp16 = select(a = var_11_to_fp16, b = var_3398_cast_fp16, cond = mask_3)[name = string("input_1021_cast_fp16")];
+            bool x_431_transpose_x_0 = const()[name = string("x_431_transpose_x_0"), val = bool(false)];
+            bool x_431_transpose_y_0 = const()[name = string("x_431_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 128]> value_41_cast_fp16 = transpose(perm = value_41_perm_0, x = v_39_cast_fp16)[name = string("transpose_176")];
+            tensor<fp16, [1, 8, 128, 128]> x_431_cast_fp16 = matmul(transpose_x = x_431_transpose_x_0, transpose_y = x_431_transpose_y_0, x = input_1021_cast_fp16, y = value_41_cast_fp16)[name = string("x_431_cast_fp16")];
+            tensor<int32, [4]> var_3402_perm_0 = const()[name = string("op_3402_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3403 = const()[name = string("op_3403"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 128, 8, 128]> var_3402_cast_fp16 = transpose(perm = var_3402_perm_0, x = x_431_cast_fp16)[name = string("transpose_175")];
+            tensor<fp16, [1, 128, 1024]> input_1023_cast_fp16 = reshape(shape = var_3403, x = var_3402_cast_fp16)[name = string("input_1023_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523881]> module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250849600))), indices_nonzero_data = tensor<uint6, [523881]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250980736))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251373760))))[name = string("module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_178_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_1023_cast_fp16)[name = string("linear_178_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_1027_cast_fp16 = add(x = input_1019_cast_fp16, y = linear_178_cast_fp16)[name = string("input_1027_cast_fp16")];
+            tensor<int32, [1]> x_435_axes_0 = const()[name = string("x_435_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_19_norm_conv_weight_to_fp16 = const()[name = string("module_layers_19_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251390208)))];
+            tensor<fp16, [1024]> module_layers_19_norm_conv_bias_to_fp16 = const()[name = string("module_layers_19_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251392320)))];
+            tensor<fp16, [1, 128, 1024]> x_435_cast_fp16 = layer_norm(axes = x_435_axes_0, beta = module_layers_19_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_conv_weight_to_fp16, x = input_1027_cast_fp16)[name = string("x_435_cast_fp16")];
+            tensor<int32, [3]> input_1029_perm_0 = const()[name = string("input_1029_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_1031_pad_type_0 = const()[name = string("input_1031_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_1031_strides_0 = const()[name = string("input_1031_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1031_pad_0 = const()[name = string("input_1031_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1031_dilations_0 = const()[name = string("input_1031_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_1031_groups_0 = const()[name = string("input_1031_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047423]> module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251394432))), indices_nonzero_data = tensor<uint6, [1047423]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251656640))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252442304))))[name = string("module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 128]> input_1029_cast_fp16 = transpose(perm = input_1029_perm_0, x = x_435_cast_fp16)[name = string("transpose_174")];
+            tensor<fp16, [1, 2048, 128]> input_1031_cast_fp16 = conv(dilations = input_1031_dilations_0, groups = input_1031_groups_0, pad = input_1031_pad_0, pad_type = input_1031_pad_type_0, strides = input_1031_strides_0, weight = module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_1029_cast_fp16)[name = string("input_1031_cast_fp16")];
+            int32 x_437_split_num_splits_0 = const()[name = string("x_437_split_num_splits_0"), val = int32(2)];
+            int32 x_437_split_axis_0 = const()[name = string("x_437_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 128]> x_437_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_437_split_cast_fp16_1 = split(axis = x_437_split_axis_0, num_splits = x_437_split_num_splits_0, x = input_1031_cast_fp16)[name = string("x_437_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_437_split_1_sigmoid_cast_fp16 = sigmoid(x = x_437_split_cast_fp16_1)[name = string("x_437_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_437_cast_fp16 = mul(x = x_437_split_cast_fp16_0, y = x_437_split_1_sigmoid_cast_fp16)[name = string("x_437_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_1033_cast_fp16 = select(a = var_11_to_fp16, b = x_437_cast_fp16, cond = var_328)[name = string("input_1033_cast_fp16")];
+            tensor<int32, [6]> input_1035_pad_0 = const()[name = string("input_1035_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_1035_mode_0 = const()[name = string("input_1035_mode_0"), val = string("constant")];
+            fp16 const_207_to_fp16 = const()[name = string("const_207_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 136]> input_1035_cast_fp16 = pad(constant_val = const_207_to_fp16, mode = input_1035_mode_0, pad = input_1035_pad_0, x = input_1033_cast_fp16)[name = string("input_1035_cast_fp16")];
+            string input_1037_pad_type_0 = const()[name = string("input_1037_pad_type_0"), val = string("valid")];
+            int32 input_1037_groups_0 = const()[name = string("input_1037_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_1037_strides_0 = const()[name = string("input_1037_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1037_pad_0 = const()[name = string("input_1037_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1037_dilations_0 = const()[name = string("input_1037_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_286_to_fp16 = const()[name = string("const_286_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252475136)))];
+            tensor<fp16, [1024]> const_287_to_fp16 = const()[name = string("const_287_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252493632)))];
+            tensor<fp16, [1, 1024, 128]> input_1039_cast_fp16 = conv(bias = const_287_to_fp16, dilations = input_1037_dilations_0, groups = input_1037_groups_0, pad = input_1037_pad_0, pad_type = input_1037_pad_type_0, strides = input_1037_strides_0, weight = const_286_to_fp16, x = input_1035_cast_fp16)[name = string("input_1039_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_1041_cast_fp16 = silu(x = input_1039_cast_fp16)[name = string("input_1041_cast_fp16")];
+            string x_439_pad_type_0 = const()[name = string("x_439_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_439_strides_0 = const()[name = string("x_439_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_439_pad_0 = const()[name = string("x_439_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_439_dilations_0 = const()[name = string("x_439_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_439_groups_0 = const()[name = string("x_439_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519548]> module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252495744))), indices_nonzero_data = tensor<uint6, [519548]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252626880))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253016640))))[name = string("module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 128]> x_439_cast_fp16 = conv(dilations = x_439_dilations_0, groups = x_439_groups_0, pad = x_439_pad_0, pad_type = x_439_pad_type_0, strides = x_439_strides_0, weight = module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_1041_cast_fp16)[name = string("x_439_cast_fp16")];
+            tensor<int32, [3]> input_1043_perm_0 = const()[name = string("input_1043_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 128, 1024]> input_1043_cast_fp16 = transpose(perm = input_1043_perm_0, x = x_439_cast_fp16)[name = string("transpose_173")];
+            tensor<fp16, [1, 128, 1024]> input_1045_cast_fp16 = add(x = input_1027_cast_fp16, y = input_1043_cast_fp16)[name = string("input_1045_cast_fp16")];
+            tensor<int32, [1]> input_1047_axes_0 = const()[name = string("input_1047_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_19_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_19_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253033088)))];
+            tensor<fp16, [1024]> module_layers_19_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_19_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253035200)))];
+            tensor<fp16, [1, 128, 1024]> input_1047_cast_fp16 = layer_norm(axes = input_1047_axes_0, beta = module_layers_19_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_feed_forward2_weight_to_fp16, x = input_1045_cast_fp16)[name = string("input_1047_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2091309]> module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253037312))), indices_nonzero_data = tensor<uint6, [2091309]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253561664))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(255130240))))[name = string("module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 4096]> linear_179_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_1047_cast_fp16)[name = string("linear_179_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_1051_cast_fp16 = silu(x = linear_179_cast_fp16)[name = string("input_1051_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2085070]> module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(255195840))), indices_nonzero_data = tensor<uint6, [2085070]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(255720192))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257284096))))[name = string("module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_180_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_1051_cast_fp16)[name = string("linear_180_cast_fp16")];
+            fp16 var_3463_to_fp16 = const()[name = string("op_3463_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 128, 1024]> var_3464_cast_fp16 = mul(x = linear_180_cast_fp16, y = var_3463_to_fp16)[name = string("op_3464_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_1057_cast_fp16 = add(x = input_1045_cast_fp16, y = var_3464_cast_fp16)[name = string("input_1057_cast_fp16")];
+            tensor<int32, [1]> input_1059_axes_0 = const()[name = string("input_1059_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_19_norm_out_weight_to_fp16 = const()[name = string("module_layers_19_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257300544)))];
+            tensor<fp16, [1024]> module_layers_19_norm_out_bias_to_fp16 = const()[name = string("module_layers_19_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257302656)))];
+            tensor<fp16, [1, 128, 1024]> input_1059_cast_fp16 = layer_norm(axes = input_1059_axes_0, beta = module_layers_19_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_out_weight_to_fp16, x = input_1057_cast_fp16)[name = string("input_1059_cast_fp16")];
+            tensor<int32, [1]> input_1061_axes_0 = const()[name = string("input_1061_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_20_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_20_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257304768)))];
+            tensor<fp16, [1024]> module_layers_20_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_20_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257306880)))];
+            tensor<fp16, [1, 128, 1024]> input_1061_cast_fp16 = layer_norm(axes = input_1061_axes_0, beta = module_layers_20_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_feed_forward1_weight_to_fp16, x = input_1059_cast_fp16)[name = string("input_1061_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2088569]> module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257308992))), indices_nonzero_data = tensor<uint6, [2088569]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257833344))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259399872))))[name = string("module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 4096]> linear_181_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_1061_cast_fp16)[name = string("linear_181_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_1065_cast_fp16 = silu(x = linear_181_cast_fp16)[name = string("input_1065_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2053925]> module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259465472))), indices_nonzero_data = tensor<uint6, [2053925]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259989824))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261530368))))[name = string("module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_182_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_1065_cast_fp16)[name = string("linear_182_cast_fp16")];
+            fp16 var_3492_to_fp16 = const()[name = string("op_3492_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 128, 1024]> var_3493_cast_fp16 = mul(x = linear_182_cast_fp16, y = var_3492_to_fp16)[name = string("op_3493_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_1071_cast_fp16 = add(x = input_1059_cast_fp16, y = var_3493_cast_fp16)[name = string("input_1071_cast_fp16")];
+            tensor<int32, [1]> query_41_axes_0 = const()[name = string("query_41_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_20_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_20_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261546816)))];
+            tensor<fp16, [1024]> module_layers_20_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_20_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261548928)))];
+            tensor<fp16, [1, 128, 1024]> query_41_cast_fp16 = layer_norm(axes = query_41_axes_0, beta = module_layers_20_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_self_att_weight_to_fp16, x = input_1071_cast_fp16)[name = string("query_41_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525329]> module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261551040))), indices_nonzero_data = tensor<uint6, [525329]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261682176))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262076288))))[name = string("module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_183_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_41_cast_fp16)[name = string("linear_183_cast_fp16")];
+            tensor<int32, [4]> var_3509 = const()[name = string("op_3509"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> q_121_cast_fp16 = reshape(shape = var_3509, x = linear_183_cast_fp16)[name = string("q_121_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525107]> module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262092736))), indices_nonzero_data = tensor<uint6, [525107]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262223872))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262617792))))[name = string("module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_184_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_41_cast_fp16)[name = string("linear_184_cast_fp16")];
+            tensor<int32, [4]> var_3513 = const()[name = string("op_3513"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> k_81_cast_fp16 = reshape(shape = var_3513, x = linear_184_cast_fp16)[name = string("k_81_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525707]> module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262634240))), indices_nonzero_data = tensor<uint6, [525707]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262765376))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263159744))))[name = string("module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_185_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_41_cast_fp16)[name = string("linear_185_cast_fp16")];
+            tensor<int32, [4]> var_3517 = const()[name = string("op_3517"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> v_41_cast_fp16 = reshape(shape = var_3517, x = linear_185_cast_fp16)[name = string("v_41_cast_fp16")];
+            tensor<int32, [4]> value_43_perm_0 = const()[name = string("value_43_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_20_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_20_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263176192)))];
+            tensor<fp16, [1, 128, 8, 128]> var_3529_cast_fp16 = add(x = q_121_cast_fp16, y = module_layers_20_self_attn_pos_bias_u_to_fp16)[name = string("op_3529_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_20_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_20_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263178304)))];
+            tensor<fp16, [1, 128, 8, 128]> var_3531_cast_fp16 = add(x = q_121_cast_fp16, y = module_layers_20_self_attn_pos_bias_v_to_fp16)[name = string("op_3531_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_41_perm_0 = const()[name = string("q_with_bias_v_41_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_447_transpose_x_0 = const()[name = string("x_447_transpose_x_0"), val = bool(false)];
+            bool x_447_transpose_y_0 = const()[name = string("x_447_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 255]> var_3533_to_fp16 = const()[name = string("op_3533_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(317867904)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_41_cast_fp16 = transpose(perm = q_with_bias_v_41_perm_0, x = var_3531_cast_fp16)[name = string("transpose_172")];
+            tensor<fp16, [1, 8, 128, 255]> x_447_cast_fp16 = matmul(transpose_x = x_447_transpose_x_0, transpose_y = x_447_transpose_y_0, x = q_with_bias_v_41_cast_fp16, y = var_3533_to_fp16)[name = string("x_447_cast_fp16")];
+            tensor<int32, [8]> x_449_pad_0 = const()[name = string("x_449_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_449_mode_0 = const()[name = string("x_449_mode_0"), val = string("constant")];
+            fp16 const_214_to_fp16 = const()[name = string("const_214_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 128, 256]> x_449_cast_fp16 = pad(constant_val = const_214_to_fp16, mode = x_449_mode_0, pad = x_449_pad_0, x = x_447_cast_fp16)[name = string("x_449_cast_fp16")];
+            tensor<int32, [4]> var_3541 = const()[name = string("op_3541"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_451_cast_fp16 = reshape(shape = var_3541, x = x_449_cast_fp16)[name = string("x_451_cast_fp16")];
+            tensor<int32, [4]> var_3545_begin_0 = const()[name = string("op_3545_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3545_end_0 = const()[name = string("op_3545_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
+            tensor<bool, [4]> var_3545_end_mask_0 = const()[name = string("op_3545_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 255, 128]> var_3545_cast_fp16 = slice_by_index(begin = var_3545_begin_0, end = var_3545_end_0, end_mask = var_3545_end_mask_0, x = x_451_cast_fp16)[name = string("op_3545_cast_fp16")];
+            tensor<int32, [4]> var_3546 = const()[name = string("op_3546"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_81_cast_fp16 = reshape(shape = var_3546, x = var_3545_cast_fp16)[name = string("matrix_bd_81_cast_fp16")];
+            bool matrix_ac_41_transpose_x_0 = const()[name = string("matrix_ac_41_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_41_transpose_y_0 = const()[name = string("matrix_ac_41_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_136_perm_0 = const()[name = string("transpose_136_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_137_perm_0 = const()[name = string("transpose_137_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 128]> transpose_137 = transpose(perm = transpose_137_perm_0, x = k_81_cast_fp16)[name = string("transpose_170")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_136 = transpose(perm = transpose_136_perm_0, x = var_3529_cast_fp16)[name = string("transpose_171")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_41_cast_fp16 = matmul(transpose_x = matrix_ac_41_transpose_x_0, transpose_y = matrix_ac_41_transpose_y_0, x = transpose_136, y = transpose_137)[name = string("matrix_ac_41_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_83_begin_0 = const()[name = string("matrix_bd_83_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_83_end_0 = const()[name = string("matrix_bd_83_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
+            tensor<bool, [4]> matrix_bd_83_end_mask_0 = const()[name = string("matrix_bd_83_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_83_cast_fp16 = slice_by_index(begin = matrix_bd_83_begin_0, end = matrix_bd_83_end_0, end_mask = matrix_bd_83_end_mask_0, x = matrix_bd_81_cast_fp16)[name = string("matrix_bd_83_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_3555_cast_fp16 = add(x = matrix_ac_41_cast_fp16, y = matrix_bd_83_cast_fp16)[name = string("op_3555_cast_fp16")];
+            fp16 _inversed_scores_81_y_0_to_fp16 = const()[name = string("_inversed_scores_81_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_81_cast_fp16 = mul(x = var_3555_cast_fp16, y = _inversed_scores_81_y_0_to_fp16)[name = string("_inversed_scores_81_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_83_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_81_cast_fp16, cond = mask_3)[name = string("scores_83_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_3561_cast_fp16 = softmax(axis = var_30, x = scores_83_cast_fp16)[name = string("op_3561_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_1073_cast_fp16 = select(a = var_11_to_fp16, b = var_3561_cast_fp16, cond = mask_3)[name = string("input_1073_cast_fp16")];
+            bool x_453_transpose_x_0 = const()[name = string("x_453_transpose_x_0"), val = bool(false)];
+            bool x_453_transpose_y_0 = const()[name = string("x_453_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 128]> value_43_cast_fp16 = transpose(perm = value_43_perm_0, x = v_41_cast_fp16)[name = string("transpose_169")];
+            tensor<fp16, [1, 8, 128, 128]> x_453_cast_fp16 = matmul(transpose_x = x_453_transpose_x_0, transpose_y = x_453_transpose_y_0, x = input_1073_cast_fp16, y = value_43_cast_fp16)[name = string("x_453_cast_fp16")];
+            tensor<int32, [4]> var_3565_perm_0 = const()[name = string("op_3565_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3566 = const()[name = string("op_3566"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 128, 8, 128]> var_3565_cast_fp16 = transpose(perm = var_3565_perm_0, x = x_453_cast_fp16)[name = string("transpose_168")];
+            tensor<fp16, [1, 128, 1024]> input_1075_cast_fp16 = reshape(shape = var_3566, x = var_3565_cast_fp16)[name = string("input_1075_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [526127]> module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263440576))), indices_nonzero_data = tensor<uint6, [526127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263571712))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263966400))))[name = string("module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_187_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_1075_cast_fp16)[name = string("linear_187_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_1079_cast_fp16 = add(x = input_1071_cast_fp16, y = linear_187_cast_fp16)[name = string("input_1079_cast_fp16")];
+            tensor<int32, [1]> x_457_axes_0 = const()[name = string("x_457_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_20_norm_conv_weight_to_fp16 = const()[name = string("module_layers_20_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263982848)))];
+            tensor<fp16, [1024]> module_layers_20_norm_conv_bias_to_fp16 = const()[name = string("module_layers_20_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263984960)))];
+            tensor<fp16, [1, 128, 1024]> x_457_cast_fp16 = layer_norm(axes = x_457_axes_0, beta = module_layers_20_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_conv_weight_to_fp16, x = input_1079_cast_fp16)[name = string("x_457_cast_fp16")];
+            tensor<int32, [3]> input_1081_perm_0 = const()[name = string("input_1081_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_1083_pad_type_0 = const()[name = string("input_1083_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_1083_strides_0 = const()[name = string("input_1083_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1083_pad_0 = const()[name = string("input_1083_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1083_dilations_0 = const()[name = string("input_1083_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_1083_groups_0 = const()[name = string("input_1083_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1049203]> module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263987072))), indices_nonzero_data = tensor<uint6, [1049203]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264249280))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265036288))))[name = string("module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 128]> input_1081_cast_fp16 = transpose(perm = input_1081_perm_0, x = x_457_cast_fp16)[name = string("transpose_167")];
+            tensor<fp16, [1, 2048, 128]> input_1083_cast_fp16 = conv(dilations = input_1083_dilations_0, groups = input_1083_groups_0, pad = input_1083_pad_0, pad_type = input_1083_pad_type_0, strides = input_1083_strides_0, weight = module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_1081_cast_fp16)[name = string("input_1083_cast_fp16")];
+            int32 x_459_split_num_splits_0 = const()[name = string("x_459_split_num_splits_0"), val = int32(2)];
+            int32 x_459_split_axis_0 = const()[name = string("x_459_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 128]> x_459_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_459_split_cast_fp16_1 = split(axis = x_459_split_axis_0, num_splits = x_459_split_num_splits_0, x = input_1083_cast_fp16)[name = string("x_459_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_459_split_1_sigmoid_cast_fp16 = sigmoid(x = x_459_split_cast_fp16_1)[name = string("x_459_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_459_cast_fp16 = mul(x = x_459_split_cast_fp16_0, y = x_459_split_1_sigmoid_cast_fp16)[name = string("x_459_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_1085_cast_fp16 = select(a = var_11_to_fp16, b = x_459_cast_fp16, cond = var_328)[name = string("input_1085_cast_fp16")];
+            tensor<int32, [6]> input_1087_pad_0 = const()[name = string("input_1087_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_1087_mode_0 = const()[name = string("input_1087_mode_0"), val = string("constant")];
+            fp16 const_217_to_fp16 = const()[name = string("const_217_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 136]> input_1087_cast_fp16 = pad(constant_val = const_217_to_fp16, mode = input_1087_mode_0, pad = input_1087_pad_0, x = input_1085_cast_fp16)[name = string("input_1087_cast_fp16")];
+            string input_1089_pad_type_0 = const()[name = string("input_1089_pad_type_0"), val = string("valid")];
+            int32 input_1089_groups_0 = const()[name = string("input_1089_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_1089_strides_0 = const()[name = string("input_1089_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1089_pad_0 = const()[name = string("input_1089_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1089_dilations_0 = const()[name = string("input_1089_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_288_to_fp16 = const()[name = string("const_288_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265069120)))];
+            tensor<fp16, [1024]> const_289_to_fp16 = const()[name = string("const_289_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265087616)))];
+            tensor<fp16, [1, 1024, 128]> input_1091_cast_fp16 = conv(bias = const_289_to_fp16, dilations = input_1089_dilations_0, groups = input_1089_groups_0, pad = input_1089_pad_0, pad_type = input_1089_pad_type_0, strides = input_1089_strides_0, weight = const_288_to_fp16, x = input_1087_cast_fp16)[name = string("input_1091_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_1093_cast_fp16 = silu(x = input_1091_cast_fp16)[name = string("input_1093_cast_fp16")];
+            string x_461_pad_type_0 = const()[name = string("x_461_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_461_strides_0 = const()[name = string("x_461_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_461_pad_0 = const()[name = string("x_461_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_461_dilations_0 = const()[name = string("x_461_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_461_groups_0 = const()[name = string("x_461_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [518556]> module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265089728))), indices_nonzero_data = tensor<uint6, [518556]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265220864))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265609856))))[name = string("module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 128]> x_461_cast_fp16 = conv(dilations = x_461_dilations_0, groups = x_461_groups_0, pad = x_461_pad_0, pad_type = x_461_pad_type_0, strides = x_461_strides_0, weight = module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_1093_cast_fp16)[name = string("x_461_cast_fp16")];
+            tensor<int32, [3]> input_1095_perm_0 = const()[name = string("input_1095_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 128, 1024]> input_1095_cast_fp16 = transpose(perm = input_1095_perm_0, x = x_461_cast_fp16)[name = string("transpose_166")];
+            tensor<fp16, [1, 128, 1024]> input_1097_cast_fp16 = add(x = input_1079_cast_fp16, y = input_1095_cast_fp16)[name = string("input_1097_cast_fp16")];
+            tensor<int32, [1]> input_1099_axes_0 = const()[name = string("input_1099_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_20_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_20_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265626304)))];
+            tensor<fp16, [1024]> module_layers_20_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_20_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265628416)))];
+            tensor<fp16, [1, 128, 1024]> input_1099_cast_fp16 = layer_norm(axes = input_1099_axes_0, beta = module_layers_20_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_feed_forward2_weight_to_fp16, x = input_1097_cast_fp16)[name = string("input_1099_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2095351]> module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265630528))), indices_nonzero_data = tensor<uint6, [2095351]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(266154880))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267726464))))[name = string("module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 4096]> linear_188_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_1099_cast_fp16)[name = string("linear_188_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_1103_cast_fp16 = silu(x = linear_188_cast_fp16)[name = string("input_1103_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2074717]> module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267792064))), indices_nonzero_data = tensor<uint6, [2074717]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268316416))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269872576))))[name = string("module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_189_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_1103_cast_fp16)[name = string("linear_189_cast_fp16")];
+            fp16 var_3626_to_fp16 = const()[name = string("op_3626_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 128, 1024]> var_3627_cast_fp16 = mul(x = linear_189_cast_fp16, y = var_3626_to_fp16)[name = string("op_3627_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_1109_cast_fp16 = add(x = input_1097_cast_fp16, y = var_3627_cast_fp16)[name = string("input_1109_cast_fp16")];
+            tensor<int32, [1]> input_1111_axes_0 = const()[name = string("input_1111_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_20_norm_out_weight_to_fp16 = const()[name = string("module_layers_20_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269889024)))];
+            tensor<fp16, [1024]> module_layers_20_norm_out_bias_to_fp16 = const()[name = string("module_layers_20_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269891136)))];
+            tensor<fp16, [1, 128, 1024]> input_1111_cast_fp16 = layer_norm(axes = input_1111_axes_0, beta = module_layers_20_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_out_weight_to_fp16, x = input_1109_cast_fp16)[name = string("input_1111_cast_fp16")];
+            tensor<int32, [1]> input_1113_axes_0 = const()[name = string("input_1113_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_21_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_21_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269893248)))];
+            tensor<fp16, [1024]> module_layers_21_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_21_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269895360)))];
+            tensor<fp16, [1, 128, 1024]> input_1113_cast_fp16 = layer_norm(axes = input_1113_axes_0, beta = module_layers_21_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_feed_forward1_weight_to_fp16, x = input_1111_cast_fp16)[name = string("input_1113_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2093447]> module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269897472))), indices_nonzero_data = tensor<uint6, [2093447]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270421824))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271992000))))[name = string("module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 4096]> linear_190_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_1113_cast_fp16)[name = string("linear_190_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_1117_cast_fp16 = silu(x = linear_190_cast_fp16)[name = string("input_1117_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2029278]> module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272057600))), indices_nonzero_data = tensor<uint6, [2029278]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272581952))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274104000))))[name = string("module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_191_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_1117_cast_fp16)[name = string("linear_191_cast_fp16")];
+            fp16 var_3655_to_fp16 = const()[name = string("op_3655_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 128, 1024]> var_3656_cast_fp16 = mul(x = linear_191_cast_fp16, y = var_3655_to_fp16)[name = string("op_3656_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_1123_cast_fp16 = add(x = input_1111_cast_fp16, y = var_3656_cast_fp16)[name = string("input_1123_cast_fp16")];
+            tensor<int32, [1]> query_43_axes_0 = const()[name = string("query_43_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_21_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_21_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274120448)))];
+            tensor<fp16, [1024]> module_layers_21_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_21_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274122560)))];
+            tensor<fp16, [1, 128, 1024]> query_43_cast_fp16 = layer_norm(axes = query_43_axes_0, beta = module_layers_21_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_self_att_weight_to_fp16, x = input_1123_cast_fp16)[name = string("query_43_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525507]> module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274124672))), indices_nonzero_data = tensor<uint6, [525507]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274255808))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274650048))))[name = string("module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_192_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_43_cast_fp16)[name = string("linear_192_cast_fp16")];
+            tensor<int32, [4]> var_3672 = const()[name = string("op_3672"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> q_127_cast_fp16 = reshape(shape = var_3672, x = linear_192_cast_fp16)[name = string("q_127_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525093]> module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274666496))), indices_nonzero_data = tensor<uint6, [525093]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274797632))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275191552))))[name = string("module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_193_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_43_cast_fp16)[name = string("linear_193_cast_fp16")];
+            tensor<int32, [4]> var_3676 = const()[name = string("op_3676"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> k_85_cast_fp16 = reshape(shape = var_3676, x = linear_193_cast_fp16)[name = string("k_85_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525954]> module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275208000))), indices_nonzero_data = tensor<uint6, [525954]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275339136))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275733696))))[name = string("module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_194_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_43_cast_fp16)[name = string("linear_194_cast_fp16")];
+            tensor<int32, [4]> var_3680 = const()[name = string("op_3680"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> v_43_cast_fp16 = reshape(shape = var_3680, x = linear_194_cast_fp16)[name = string("v_43_cast_fp16")];
+            tensor<int32, [4]> value_45_perm_0 = const()[name = string("value_45_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_21_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_21_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275750144)))];
+            tensor<fp16, [1, 128, 8, 128]> var_3692_cast_fp16 = add(x = q_127_cast_fp16, y = module_layers_21_self_attn_pos_bias_u_to_fp16)[name = string("op_3692_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_21_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_21_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275752256)))];
+            tensor<fp16, [1, 128, 8, 128]> var_3694_cast_fp16 = add(x = q_127_cast_fp16, y = module_layers_21_self_attn_pos_bias_v_to_fp16)[name = string("op_3694_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_43_perm_0 = const()[name = string("q_with_bias_v_43_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_469_transpose_x_0 = const()[name = string("x_469_transpose_x_0"), val = bool(false)];
+            bool x_469_transpose_y_0 = const()[name = string("x_469_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 255]> var_3696_to_fp16 = const()[name = string("op_3696_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(318390208)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_43_cast_fp16 = transpose(perm = q_with_bias_v_43_perm_0, x = var_3694_cast_fp16)[name = string("transpose_165")];
+            tensor<fp16, [1, 8, 128, 255]> x_469_cast_fp16 = matmul(transpose_x = x_469_transpose_x_0, transpose_y = x_469_transpose_y_0, x = q_with_bias_v_43_cast_fp16, y = var_3696_to_fp16)[name = string("x_469_cast_fp16")];
+            tensor<int32, [8]> x_471_pad_0 = const()[name = string("x_471_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_471_mode_0 = const()[name = string("x_471_mode_0"), val = string("constant")];
+            fp16 const_224_to_fp16 = const()[name = string("const_224_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 128, 256]> x_471_cast_fp16 = pad(constant_val = const_224_to_fp16, mode = x_471_mode_0, pad = x_471_pad_0, x = x_469_cast_fp16)[name = string("x_471_cast_fp16")];
+            tensor<int32, [4]> var_3704 = const()[name = string("op_3704"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_473_cast_fp16 = reshape(shape = var_3704, x = x_471_cast_fp16)[name = string("x_473_cast_fp16")];
+            tensor<int32, [4]> var_3708_begin_0 = const()[name = string("op_3708_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3708_end_0 = const()[name = string("op_3708_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
+            tensor<bool, [4]> var_3708_end_mask_0 = const()[name = string("op_3708_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 255, 128]> var_3708_cast_fp16 = slice_by_index(begin = var_3708_begin_0, end = var_3708_end_0, end_mask = var_3708_end_mask_0, x = x_473_cast_fp16)[name = string("op_3708_cast_fp16")];
+            tensor<int32, [4]> var_3709 = const()[name = string("op_3709"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_85_cast_fp16 = reshape(shape = var_3709, x = var_3708_cast_fp16)[name = string("matrix_bd_85_cast_fp16")];
+            bool matrix_ac_43_transpose_x_0 = const()[name = string("matrix_ac_43_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_43_transpose_y_0 = const()[name = string("matrix_ac_43_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_138_perm_0 = const()[name = string("transpose_138_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_139_perm_0 = const()[name = string("transpose_139_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 128]> transpose_139 = transpose(perm = transpose_139_perm_0, x = k_85_cast_fp16)[name = string("transpose_163")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_138 = transpose(perm = transpose_138_perm_0, x = var_3692_cast_fp16)[name = string("transpose_164")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_43_cast_fp16 = matmul(transpose_x = matrix_ac_43_transpose_x_0, transpose_y = matrix_ac_43_transpose_y_0, x = transpose_138, y = transpose_139)[name = string("matrix_ac_43_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_87_begin_0 = const()[name = string("matrix_bd_87_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_87_end_0 = const()[name = string("matrix_bd_87_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
+            tensor<bool, [4]> matrix_bd_87_end_mask_0 = const()[name = string("matrix_bd_87_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_87_cast_fp16 = slice_by_index(begin = matrix_bd_87_begin_0, end = matrix_bd_87_end_0, end_mask = matrix_bd_87_end_mask_0, x = matrix_bd_85_cast_fp16)[name = string("matrix_bd_87_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_3718_cast_fp16 = add(x = matrix_ac_43_cast_fp16, y = matrix_bd_87_cast_fp16)[name = string("op_3718_cast_fp16")];
+            fp16 _inversed_scores_85_y_0_to_fp16 = const()[name = string("_inversed_scores_85_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_85_cast_fp16 = mul(x = var_3718_cast_fp16, y = _inversed_scores_85_y_0_to_fp16)[name = string("_inversed_scores_85_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_87_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_85_cast_fp16, cond = mask_3)[name = string("scores_87_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_3724_cast_fp16 = softmax(axis = var_30, x = scores_87_cast_fp16)[name = string("op_3724_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_1125_cast_fp16 = select(a = var_11_to_fp16, b = var_3724_cast_fp16, cond = mask_3)[name = string("input_1125_cast_fp16")];
+            bool x_475_transpose_x_0 = const()[name = string("x_475_transpose_x_0"), val = bool(false)];
+            bool x_475_transpose_y_0 = const()[name = string("x_475_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 128]> value_45_cast_fp16 = transpose(perm = value_45_perm_0, x = v_43_cast_fp16)[name = string("transpose_162")];
+            tensor<fp16, [1, 8, 128, 128]> x_475_cast_fp16 = matmul(transpose_x = x_475_transpose_x_0, transpose_y = x_475_transpose_y_0, x = input_1125_cast_fp16, y = value_45_cast_fp16)[name = string("x_475_cast_fp16")];
+            tensor<int32, [4]> var_3728_perm_0 = const()[name = string("op_3728_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3729 = const()[name = string("op_3729"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 128, 8, 128]> var_3728_cast_fp16 = transpose(perm = var_3728_perm_0, x = x_475_cast_fp16)[name = string("transpose_161")];
+            tensor<fp16, [1, 128, 1024]> input_1127_cast_fp16 = reshape(shape = var_3729, x = var_3728_cast_fp16)[name = string("input_1127_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [526102]> module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276014528))), indices_nonzero_data = tensor<uint6, [526102]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276145664))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276540352))))[name = string("module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_196_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_1127_cast_fp16)[name = string("linear_196_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_1131_cast_fp16 = add(x = input_1123_cast_fp16, y = linear_196_cast_fp16)[name = string("input_1131_cast_fp16")];
+            tensor<int32, [1]> x_479_axes_0 = const()[name = string("x_479_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_21_norm_conv_weight_to_fp16 = const()[name = string("module_layers_21_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276556800)))];
+            tensor<fp16, [1024]> module_layers_21_norm_conv_bias_to_fp16 = const()[name = string("module_layers_21_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276558912)))];
+            tensor<fp16, [1, 128, 1024]> x_479_cast_fp16 = layer_norm(axes = x_479_axes_0, beta = module_layers_21_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_conv_weight_to_fp16, x = input_1131_cast_fp16)[name = string("x_479_cast_fp16")];
+            tensor<int32, [3]> input_1133_perm_0 = const()[name = string("input_1133_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_1135_pad_type_0 = const()[name = string("input_1135_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_1135_strides_0 = const()[name = string("input_1135_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1135_pad_0 = const()[name = string("input_1135_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1135_dilations_0 = const()[name = string("input_1135_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_1135_groups_0 = const()[name = string("input_1135_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1049320]> module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276561024))), indices_nonzero_data = tensor<uint6, [1049320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276823232))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277610304))))[name = string("module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 128]> input_1133_cast_fp16 = transpose(perm = input_1133_perm_0, x = x_479_cast_fp16)[name = string("transpose_160")];
+            tensor<fp16, [1, 2048, 128]> input_1135_cast_fp16 = conv(dilations = input_1135_dilations_0, groups = input_1135_groups_0, pad = input_1135_pad_0, pad_type = input_1135_pad_type_0, strides = input_1135_strides_0, weight = module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_1133_cast_fp16)[name = string("input_1135_cast_fp16")];
+            int32 x_481_split_num_splits_0 = const()[name = string("x_481_split_num_splits_0"), val = int32(2)];
+            int32 x_481_split_axis_0 = const()[name = string("x_481_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 128]> x_481_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_481_split_cast_fp16_1 = split(axis = x_481_split_axis_0, num_splits = x_481_split_num_splits_0, x = input_1135_cast_fp16)[name = string("x_481_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_481_split_1_sigmoid_cast_fp16 = sigmoid(x = x_481_split_cast_fp16_1)[name = string("x_481_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_481_cast_fp16 = mul(x = x_481_split_cast_fp16_0, y = x_481_split_1_sigmoid_cast_fp16)[name = string("x_481_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_1137_cast_fp16 = select(a = var_11_to_fp16, b = x_481_cast_fp16, cond = var_328)[name = string("input_1137_cast_fp16")];
+            tensor<int32, [6]> input_1139_pad_0 = const()[name = string("input_1139_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_1139_mode_0 = const()[name = string("input_1139_mode_0"), val = string("constant")];
+            fp16 const_227_to_fp16 = const()[name = string("const_227_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 136]> input_1139_cast_fp16 = pad(constant_val = const_227_to_fp16, mode = input_1139_mode_0, pad = input_1139_pad_0, x = input_1137_cast_fp16)[name = string("input_1139_cast_fp16")];
+            string input_1141_pad_type_0 = const()[name = string("input_1141_pad_type_0"), val = string("valid")];
+            int32 input_1141_groups_0 = const()[name = string("input_1141_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_1141_strides_0 = const()[name = string("input_1141_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1141_pad_0 = const()[name = string("input_1141_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1141_dilations_0 = const()[name = string("input_1141_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_290_to_fp16 = const()[name = string("const_290_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277643136)))];
+            tensor<fp16, [1024]> const_291_to_fp16 = const()[name = string("const_291_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277661632)))];
+            tensor<fp16, [1, 1024, 128]> input_1143_cast_fp16 = conv(bias = const_291_to_fp16, dilations = input_1141_dilations_0, groups = input_1141_groups_0, pad = input_1141_pad_0, pad_type = input_1141_pad_type_0, strides = input_1141_strides_0, weight = const_290_to_fp16, x = input_1139_cast_fp16)[name = string("input_1143_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_1145_cast_fp16 = silu(x = input_1143_cast_fp16)[name = string("input_1145_cast_fp16")];
+            string x_483_pad_type_0 = const()[name = string("x_483_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_483_strides_0 = const()[name = string("x_483_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_483_pad_0 = const()[name = string("x_483_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_483_dilations_0 = const()[name = string("x_483_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_483_groups_0 = const()[name = string("x_483_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519430]> module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277663744))), indices_nonzero_data = tensor<uint6, [519430]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277794880))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278184576))))[name = string("module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 128]> x_483_cast_fp16 = conv(dilations = x_483_dilations_0, groups = x_483_groups_0, pad = x_483_pad_0, pad_type = x_483_pad_type_0, strides = x_483_strides_0, weight = module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_1145_cast_fp16)[name = string("x_483_cast_fp16")];
+            tensor<int32, [3]> input_1147_perm_0 = const()[name = string("input_1147_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 128, 1024]> input_1147_cast_fp16 = transpose(perm = input_1147_perm_0, x = x_483_cast_fp16)[name = string("transpose_159")];
+            tensor<fp16, [1, 128, 1024]> input_1149_cast_fp16 = add(x = input_1131_cast_fp16, y = input_1147_cast_fp16)[name = string("input_1149_cast_fp16")];
+            tensor<int32, [1]> input_1151_axes_0 = const()[name = string("input_1151_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_21_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_21_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278201024)))];
+            tensor<fp16, [1024]> module_layers_21_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_21_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278203136)))];
+            tensor<fp16, [1, 128, 1024]> input_1151_cast_fp16 = layer_norm(axes = input_1151_axes_0, beta = module_layers_21_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_feed_forward2_weight_to_fp16, x = input_1149_cast_fp16)[name = string("input_1151_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2100215]> module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278205248))), indices_nonzero_data = tensor<uint6, [2100215]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278729600))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280304832))))[name = string("module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 4096]> linear_197_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_1151_cast_fp16)[name = string("linear_197_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_1155_cast_fp16 = silu(x = linear_197_cast_fp16)[name = string("input_1155_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2058868]> module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280370432))), indices_nonzero_data = tensor<uint6, [2058868]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280894784))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282439040))))[name = string("module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_198_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_1155_cast_fp16)[name = string("linear_198_cast_fp16")];
+            fp16 var_3789_to_fp16 = const()[name = string("op_3789_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 128, 1024]> var_3790_cast_fp16 = mul(x = linear_198_cast_fp16, y = var_3789_to_fp16)[name = string("op_3790_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_1161_cast_fp16 = add(x = input_1149_cast_fp16, y = var_3790_cast_fp16)[name = string("input_1161_cast_fp16")];
+            tensor<int32, [1]> input_1163_axes_0 = const()[name = string("input_1163_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_21_norm_out_weight_to_fp16 = const()[name = string("module_layers_21_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282455488)))];
+            tensor<fp16, [1024]> module_layers_21_norm_out_bias_to_fp16 = const()[name = string("module_layers_21_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282457600)))];
+            tensor<fp16, [1, 128, 1024]> input_1163_cast_fp16 = layer_norm(axes = input_1163_axes_0, beta = module_layers_21_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_out_weight_to_fp16, x = input_1161_cast_fp16)[name = string("input_1163_cast_fp16")];
+            tensor<int32, [1]> input_1165_axes_0 = const()[name = string("input_1165_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_22_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_22_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282459712)))];
+            tensor<fp16, [1024]> module_layers_22_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_22_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282461824)))];
+            tensor<fp16, [1, 128, 1024]> input_1165_cast_fp16 = layer_norm(axes = input_1165_axes_0, beta = module_layers_22_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_feed_forward1_weight_to_fp16, x = input_1163_cast_fp16)[name = string("input_1165_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2098607]> module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282463936))), indices_nonzero_data = tensor<uint6, [2098607]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282988288))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284562368))))[name = string("module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 4096]> linear_199_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_1165_cast_fp16)[name = string("linear_199_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_1169_cast_fp16 = silu(x = linear_199_cast_fp16)[name = string("input_1169_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2035005]> module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284627968))), indices_nonzero_data = tensor<uint6, [2035005]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285152320))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286678656))))[name = string("module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_200_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_1169_cast_fp16)[name = string("linear_200_cast_fp16")];
+            fp16 var_3818_to_fp16 = const()[name = string("op_3818_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 128, 1024]> var_3819_cast_fp16 = mul(x = linear_200_cast_fp16, y = var_3818_to_fp16)[name = string("op_3819_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_1175_cast_fp16 = add(x = input_1163_cast_fp16, y = var_3819_cast_fp16)[name = string("input_1175_cast_fp16")];
+            tensor<int32, [1]> query_45_axes_0 = const()[name = string("query_45_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_22_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_22_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286695104)))];
+            tensor<fp16, [1024]> module_layers_22_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_22_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286697216)))];
+            tensor<fp16, [1, 128, 1024]> query_45_cast_fp16 = layer_norm(axes = query_45_axes_0, beta = module_layers_22_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_self_att_weight_to_fp16, x = input_1175_cast_fp16)[name = string("query_45_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [527072]> module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286699328))), indices_nonzero_data = tensor<uint6, [527072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286830464))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287225856))))[name = string("module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_201_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_45_cast_fp16)[name = string("linear_201_cast_fp16")];
+            tensor<int32, [4]> var_3835 = const()[name = string("op_3835"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> q_133_cast_fp16 = reshape(shape = var_3835, x = linear_201_cast_fp16)[name = string("q_133_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525242]> module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287242304))), indices_nonzero_data = tensor<uint6, [525242]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287373440))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287767488))))[name = string("module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_202_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_45_cast_fp16)[name = string("linear_202_cast_fp16")];
+            tensor<int32, [4]> var_3839 = const()[name = string("op_3839"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> k_89_cast_fp16 = reshape(shape = var_3839, x = linear_202_cast_fp16)[name = string("k_89_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523544]> module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287783936))), indices_nonzero_data = tensor<uint6, [523544]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287915072))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288307840))))[name = string("module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_203_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_45_cast_fp16)[name = string("linear_203_cast_fp16")];
+            tensor<int32, [4]> var_3843 = const()[name = string("op_3843"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> v_45_cast_fp16 = reshape(shape = var_3843, x = linear_203_cast_fp16)[name = string("v_45_cast_fp16")];
+            tensor<int32, [4]> value_47_perm_0 = const()[name = string("value_47_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_22_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_22_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288324288)))];
+            tensor<fp16, [1, 128, 8, 128]> var_3855_cast_fp16 = add(x = q_133_cast_fp16, y = module_layers_22_self_attn_pos_bias_u_to_fp16)[name = string("op_3855_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_22_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_22_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288326400)))];
+            tensor<fp16, [1, 128, 8, 128]> var_3857_cast_fp16 = add(x = q_133_cast_fp16, y = module_layers_22_self_attn_pos_bias_v_to_fp16)[name = string("op_3857_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_45_perm_0 = const()[name = string("q_with_bias_v_45_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_491_transpose_x_0 = const()[name = string("x_491_transpose_x_0"), val = bool(false)];
+            bool x_491_transpose_y_0 = const()[name = string("x_491_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 255]> var_3859_to_fp16 = const()[name = string("op_3859_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(318912512)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_45_cast_fp16 = transpose(perm = q_with_bias_v_45_perm_0, x = var_3857_cast_fp16)[name = string("transpose_158")];
+            tensor<fp16, [1, 8, 128, 255]> x_491_cast_fp16 = matmul(transpose_x = x_491_transpose_x_0, transpose_y = x_491_transpose_y_0, x = q_with_bias_v_45_cast_fp16, y = var_3859_to_fp16)[name = string("x_491_cast_fp16")];
+            tensor<int32, [8]> x_493_pad_0 = const()[name = string("x_493_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_493_mode_0 = const()[name = string("x_493_mode_0"), val = string("constant")];
+            fp16 const_234_to_fp16 = const()[name = string("const_234_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 128, 256]> x_493_cast_fp16 = pad(constant_val = const_234_to_fp16, mode = x_493_mode_0, pad = x_493_pad_0, x = x_491_cast_fp16)[name = string("x_493_cast_fp16")];
+            tensor<int32, [4]> var_3867 = const()[name = string("op_3867"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_495_cast_fp16 = reshape(shape = var_3867, x = x_493_cast_fp16)[name = string("x_495_cast_fp16")];
+            tensor<int32, [4]> var_3871_begin_0 = const()[name = string("op_3871_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3871_end_0 = const()[name = string("op_3871_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
+            tensor<bool, [4]> var_3871_end_mask_0 = const()[name = string("op_3871_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 255, 128]> var_3871_cast_fp16 = slice_by_index(begin = var_3871_begin_0, end = var_3871_end_0, end_mask = var_3871_end_mask_0, x = x_495_cast_fp16)[name = string("op_3871_cast_fp16")];
+            tensor<int32, [4]> var_3872 = const()[name = string("op_3872"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_89_cast_fp16 = reshape(shape = var_3872, x = var_3871_cast_fp16)[name = string("matrix_bd_89_cast_fp16")];
+            bool matrix_ac_45_transpose_x_0 = const()[name = string("matrix_ac_45_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_45_transpose_y_0 = const()[name = string("matrix_ac_45_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_140_perm_0 = const()[name = string("transpose_140_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_141_perm_0 = const()[name = string("transpose_141_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 128]> transpose_141 = transpose(perm = transpose_141_perm_0, x = k_89_cast_fp16)[name = string("transpose_156")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_140 = transpose(perm = transpose_140_perm_0, x = var_3855_cast_fp16)[name = string("transpose_157")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_45_cast_fp16 = matmul(transpose_x = matrix_ac_45_transpose_x_0, transpose_y = matrix_ac_45_transpose_y_0, x = transpose_140, y = transpose_141)[name = string("matrix_ac_45_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_91_begin_0 = const()[name = string("matrix_bd_91_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_91_end_0 = const()[name = string("matrix_bd_91_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
+            tensor<bool, [4]> matrix_bd_91_end_mask_0 = const()[name = string("matrix_bd_91_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_91_cast_fp16 = slice_by_index(begin = matrix_bd_91_begin_0, end = matrix_bd_91_end_0, end_mask = matrix_bd_91_end_mask_0, x = matrix_bd_89_cast_fp16)[name = string("matrix_bd_91_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_3881_cast_fp16 = add(x = matrix_ac_45_cast_fp16, y = matrix_bd_91_cast_fp16)[name = string("op_3881_cast_fp16")];
+            fp16 _inversed_scores_89_y_0_to_fp16 = const()[name = string("_inversed_scores_89_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_89_cast_fp16 = mul(x = var_3881_cast_fp16, y = _inversed_scores_89_y_0_to_fp16)[name = string("_inversed_scores_89_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_91_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_89_cast_fp16, cond = mask_3)[name = string("scores_91_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_3887_cast_fp16 = softmax(axis = var_30, x = scores_91_cast_fp16)[name = string("op_3887_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_1177_cast_fp16 = select(a = var_11_to_fp16, b = var_3887_cast_fp16, cond = mask_3)[name = string("input_1177_cast_fp16")];
+            bool x_497_transpose_x_0 = const()[name = string("x_497_transpose_x_0"), val = bool(false)];
+            bool x_497_transpose_y_0 = const()[name = string("x_497_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 128]> value_47_cast_fp16 = transpose(perm = value_47_perm_0, x = v_45_cast_fp16)[name = string("transpose_155")];
+            tensor<fp16, [1, 8, 128, 128]> x_497_cast_fp16 = matmul(transpose_x = x_497_transpose_x_0, transpose_y = x_497_transpose_y_0, x = input_1177_cast_fp16, y = value_47_cast_fp16)[name = string("x_497_cast_fp16")];
+            tensor<int32, [4]> var_3891_perm_0 = const()[name = string("op_3891_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3892 = const()[name = string("op_3892"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 128, 8, 128]> var_3891_cast_fp16 = transpose(perm = var_3891_perm_0, x = x_497_cast_fp16)[name = string("transpose_154")];
+            tensor<fp16, [1, 128, 1024]> input_1179_cast_fp16 = reshape(shape = var_3892, x = var_3891_cast_fp16)[name = string("input_1179_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523989]> module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288588672))), indices_nonzero_data = tensor<uint6, [523989]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288719808))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289112896))))[name = string("module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_205_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_1179_cast_fp16)[name = string("linear_205_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_1183_cast_fp16 = add(x = input_1175_cast_fp16, y = linear_205_cast_fp16)[name = string("input_1183_cast_fp16")];
+            tensor<int32, [1]> x_501_axes_0 = const()[name = string("x_501_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_22_norm_conv_weight_to_fp16 = const()[name = string("module_layers_22_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289129344)))];
+            tensor<fp16, [1024]> module_layers_22_norm_conv_bias_to_fp16 = const()[name = string("module_layers_22_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289131456)))];
+            tensor<fp16, [1, 128, 1024]> x_501_cast_fp16 = layer_norm(axes = x_501_axes_0, beta = module_layers_22_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_conv_weight_to_fp16, x = input_1183_cast_fp16)[name = string("x_501_cast_fp16")];
+            tensor<int32, [3]> input_1185_perm_0 = const()[name = string("input_1185_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_1187_pad_type_0 = const()[name = string("input_1187_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_1187_strides_0 = const()[name = string("input_1187_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1187_pad_0 = const()[name = string("input_1187_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1187_dilations_0 = const()[name = string("input_1187_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_1187_groups_0 = const()[name = string("input_1187_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1048742]> module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289133568))), indices_nonzero_data = tensor<uint6, [1048742]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289395776))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290182400))))[name = string("module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 128]> input_1185_cast_fp16 = transpose(perm = input_1185_perm_0, x = x_501_cast_fp16)[name = string("transpose_153")];
+            tensor<fp16, [1, 2048, 128]> input_1187_cast_fp16 = conv(dilations = input_1187_dilations_0, groups = input_1187_groups_0, pad = input_1187_pad_0, pad_type = input_1187_pad_type_0, strides = input_1187_strides_0, weight = module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_1185_cast_fp16)[name = string("input_1187_cast_fp16")];
+            int32 x_503_split_num_splits_0 = const()[name = string("x_503_split_num_splits_0"), val = int32(2)];
+            int32 x_503_split_axis_0 = const()[name = string("x_503_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 128]> x_503_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_503_split_cast_fp16_1 = split(axis = x_503_split_axis_0, num_splits = x_503_split_num_splits_0, x = input_1187_cast_fp16)[name = string("x_503_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_503_split_1_sigmoid_cast_fp16 = sigmoid(x = x_503_split_cast_fp16_1)[name = string("x_503_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_503_cast_fp16 = mul(x = x_503_split_cast_fp16_0, y = x_503_split_1_sigmoid_cast_fp16)[name = string("x_503_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_1189_cast_fp16 = select(a = var_11_to_fp16, b = x_503_cast_fp16, cond = var_328)[name = string("input_1189_cast_fp16")];
+            tensor<int32, [6]> input_1191_pad_0 = const()[name = string("input_1191_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_1191_mode_0 = const()[name = string("input_1191_mode_0"), val = string("constant")];
+            fp16 const_237_to_fp16 = const()[name = string("const_237_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 136]> input_1191_cast_fp16 = pad(constant_val = const_237_to_fp16, mode = input_1191_mode_0, pad = input_1191_pad_0, x = input_1189_cast_fp16)[name = string("input_1191_cast_fp16")];
+            string input_1193_pad_type_0 = const()[name = string("input_1193_pad_type_0"), val = string("valid")];
+            int32 input_1193_groups_0 = const()[name = string("input_1193_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_1193_strides_0 = const()[name = string("input_1193_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1193_pad_0 = const()[name = string("input_1193_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1193_dilations_0 = const()[name = string("input_1193_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_292_to_fp16 = const()[name = string("const_292_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290215232)))];
+            tensor<fp16, [1024]> const_293_to_fp16 = const()[name = string("const_293_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290233728)))];
+            tensor<fp16, [1, 1024, 128]> input_1195_cast_fp16 = conv(bias = const_293_to_fp16, dilations = input_1193_dilations_0, groups = input_1193_groups_0, pad = input_1193_pad_0, pad_type = input_1193_pad_type_0, strides = input_1193_strides_0, weight = const_292_to_fp16, x = input_1191_cast_fp16)[name = string("input_1195_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_1197_cast_fp16 = silu(x = input_1195_cast_fp16)[name = string("input_1197_cast_fp16")];
+            string x_505_pad_type_0 = const()[name = string("x_505_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_505_strides_0 = const()[name = string("x_505_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_505_pad_0 = const()[name = string("x_505_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_505_dilations_0 = const()[name = string("x_505_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_505_groups_0 = const()[name = string("x_505_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [518537]> module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290235840))), indices_nonzero_data = tensor<uint6, [518537]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290366976))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290755968))))[name = string("module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 128]> x_505_cast_fp16 = conv(dilations = x_505_dilations_0, groups = x_505_groups_0, pad = x_505_pad_0, pad_type = x_505_pad_type_0, strides = x_505_strides_0, weight = module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_1197_cast_fp16)[name = string("x_505_cast_fp16")];
+            tensor<int32, [3]> input_1199_perm_0 = const()[name = string("input_1199_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 128, 1024]> input_1199_cast_fp16 = transpose(perm = input_1199_perm_0, x = x_505_cast_fp16)[name = string("transpose_152")];
+            tensor<fp16, [1, 128, 1024]> input_1201_cast_fp16 = add(x = input_1183_cast_fp16, y = input_1199_cast_fp16)[name = string("input_1201_cast_fp16")];
+            tensor<int32, [1]> input_1203_axes_0 = const()[name = string("input_1203_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_22_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_22_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290772416)))];
+            tensor<fp16, [1024]> module_layers_22_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_22_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290774528)))];
+            tensor<fp16, [1, 128, 1024]> input_1203_cast_fp16 = layer_norm(axes = input_1203_axes_0, beta = module_layers_22_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_feed_forward2_weight_to_fp16, x = input_1201_cast_fp16)[name = string("input_1203_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2096859]> module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290776640))), indices_nonzero_data = tensor<uint6, [2096859]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(291300992))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292873728))))[name = string("module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 4096]> linear_206_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_1203_cast_fp16)[name = string("linear_206_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_1207_cast_fp16 = silu(x = linear_206_cast_fp16)[name = string("input_1207_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1916371]> module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292939328))), indices_nonzero_data = tensor<uint6, [1916371]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293463680))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294901056))))[name = string("module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_207_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_1207_cast_fp16)[name = string("linear_207_cast_fp16")];
+            fp16 var_3952_to_fp16 = const()[name = string("op_3952_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 128, 1024]> var_3953_cast_fp16 = mul(x = linear_207_cast_fp16, y = var_3952_to_fp16)[name = string("op_3953_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_1213_cast_fp16 = add(x = input_1201_cast_fp16, y = var_3953_cast_fp16)[name = string("input_1213_cast_fp16")];
+            tensor<int32, [1]> input_1215_axes_0 = const()[name = string("input_1215_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_22_norm_out_weight_to_fp16 = const()[name = string("module_layers_22_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294917504)))];
+            tensor<fp16, [1024]> module_layers_22_norm_out_bias_to_fp16 = const()[name = string("module_layers_22_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294919616)))];
+            tensor<fp16, [1, 128, 1024]> input_1215_cast_fp16 = layer_norm(axes = input_1215_axes_0, beta = module_layers_22_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_out_weight_to_fp16, x = input_1213_cast_fp16)[name = string("input_1215_cast_fp16")];
+            tensor<int32, [1]> input_1217_axes_0 = const()[name = string("input_1217_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_23_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_23_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294921728)))];
+            tensor<fp16, [1024]> module_layers_23_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_23_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294923840)))];
+            tensor<fp16, [1, 128, 1024]> input_1217_cast_fp16 = layer_norm(axes = input_1217_axes_0, beta = module_layers_23_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_feed_forward1_weight_to_fp16, x = input_1215_cast_fp16)[name = string("input_1217_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2102731]> module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294925952))), indices_nonzero_data = tensor<uint6, [2102731]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295450304))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297027456))))[name = string("module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 4096]> linear_208_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_1217_cast_fp16)[name = string("linear_208_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_1221_cast_fp16 = silu(x = linear_208_cast_fp16)[name = string("input_1221_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1967096]> module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297093056))), indices_nonzero_data = tensor<uint6, [1967096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297617408))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299092800))))[name = string("module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_209_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_1221_cast_fp16)[name = string("linear_209_cast_fp16")];
+            fp16 var_3981_to_fp16 = const()[name = string("op_3981_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 128, 1024]> var_3982_cast_fp16 = mul(x = linear_209_cast_fp16, y = var_3981_to_fp16)[name = string("op_3982_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_1227_cast_fp16 = add(x = input_1215_cast_fp16, y = var_3982_cast_fp16)[name = string("input_1227_cast_fp16")];
+            tensor<int32, [1]> query_axes_0 = const()[name = string("query_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_23_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_23_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299109248)))];
+            tensor<fp16, [1024]> module_layers_23_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_23_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299111360)))];
+            tensor<fp16, [1, 128, 1024]> query_cast_fp16 = layer_norm(axes = query_axes_0, beta = module_layers_23_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_self_att_weight_to_fp16, x = input_1227_cast_fp16)[name = string("query_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [526594]> module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299113472))), indices_nonzero_data = tensor<uint6, [526594]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299244608))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299639680))))[name = string("module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_210_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_cast_fp16)[name = string("linear_210_cast_fp16")];
+            tensor<int32, [4]> var_3998 = const()[name = string("op_3998"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> q_139_cast_fp16 = reshape(shape = var_3998, x = linear_210_cast_fp16)[name = string("q_139_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524871]> module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299656128))), indices_nonzero_data = tensor<uint6, [524871]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299787264))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300180992))))[name = string("module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_211_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_cast_fp16)[name = string("linear_211_cast_fp16")];
+            tensor<int32, [4]> var_4002 = const()[name = string("op_4002"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> k_93_cast_fp16 = reshape(shape = var_4002, x = linear_211_cast_fp16)[name = string("k_93_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525256]> module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300197440))), indices_nonzero_data = tensor<uint6, [525256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300328576))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300722624))))[name = string("module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_212_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_cast_fp16)[name = string("linear_212_cast_fp16")];
+            tensor<int32, [4]> var_4006 = const()[name = string("op_4006"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 128, 8, 128]> v_cast_fp16 = reshape(shape = var_4006, x = linear_212_cast_fp16)[name = string("v_cast_fp16")];
+            tensor<int32, [4]> value_perm_0 = const()[name = string("value_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_23_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_23_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300739072)))];
+            tensor<fp16, [1, 128, 8, 128]> var_4018_cast_fp16 = add(x = q_139_cast_fp16, y = module_layers_23_self_attn_pos_bias_u_to_fp16)[name = string("op_4018_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_23_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_23_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300741184)))];
+            tensor<fp16, [1, 128, 8, 128]> var_4020_cast_fp16 = add(x = q_139_cast_fp16, y = module_layers_23_self_attn_pos_bias_v_to_fp16)[name = string("op_4020_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_perm_0 = const()[name = string("q_with_bias_v_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_513_transpose_x_0 = const()[name = string("x_513_transpose_x_0"), val = bool(false)];
+            bool x_513_transpose_y_0 = const()[name = string("x_513_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 255]> var_4022_to_fp16 = const()[name = string("op_4022_to_fp16"), val = tensor<fp16, [1, 8, 128, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(319434816)))];
+            tensor<fp16, [1, 8, 128, 128]> q_with_bias_v_cast_fp16 = transpose(perm = q_with_bias_v_perm_0, x = var_4020_cast_fp16)[name = string("transpose_151")];
+            tensor<fp16, [1, 8, 128, 255]> x_513_cast_fp16 = matmul(transpose_x = x_513_transpose_x_0, transpose_y = x_513_transpose_y_0, x = q_with_bias_v_cast_fp16, y = var_4022_to_fp16)[name = string("x_513_cast_fp16")];
+            tensor<int32, [8]> x_515_pad_0 = const()[name = string("x_515_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_515_mode_0 = const()[name = string("x_515_mode_0"), val = string("constant")];
+            fp16 const_244_to_fp16 = const()[name = string("const_244_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 128, 256]> x_515_cast_fp16 = pad(constant_val = const_244_to_fp16, mode = x_515_mode_0, pad = x_515_pad_0, x = x_513_cast_fp16)[name = string("x_515_cast_fp16")];
+            tensor<int32, [4]> var_4030 = const()[name = string("op_4030"), val = tensor<int32, [4]>([1, 8, -1, 128])];
+            tensor<fp16, [1, 8, 256, 128]> x_517_cast_fp16 = reshape(shape = var_4030, x = x_515_cast_fp16)[name = string("x_517_cast_fp16")];
+            tensor<int32, [4]> var_4034_begin_0 = const()[name = string("op_4034_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_4034_end_0 = const()[name = string("op_4034_end_0"), val = tensor<int32, [4]>([1, 8, 256, 128])];
+            tensor<bool, [4]> var_4034_end_mask_0 = const()[name = string("op_4034_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 255, 128]> var_4034_cast_fp16 = slice_by_index(begin = var_4034_begin_0, end = var_4034_end_0, end_mask = var_4034_end_mask_0, x = x_517_cast_fp16)[name = string("op_4034_cast_fp16")];
+            tensor<int32, [4]> var_4035 = const()[name = string("op_4035"), val = tensor<int32, [4]>([1, 8, 128, 255])];
+            tensor<fp16, [1, 8, 128, 255]> matrix_bd_93_cast_fp16 = reshape(shape = var_4035, x = var_4034_cast_fp16)[name = string("matrix_bd_93_cast_fp16")];
+            bool matrix_ac_transpose_x_0 = const()[name = string("matrix_ac_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_transpose_y_0 = const()[name = string("matrix_ac_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_142_perm_0 = const()[name = string("transpose_142_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_143_perm_0 = const()[name = string("transpose_143_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 128]> transpose_143 = transpose(perm = transpose_143_perm_0, x = k_93_cast_fp16)[name = string("transpose_149")];
+            tensor<fp16, [1, 8, 128, 128]> transpose_142 = transpose(perm = transpose_142_perm_0, x = var_4018_cast_fp16)[name = string("transpose_150")];
+            tensor<fp16, [1, 8, 128, 128]> matrix_ac_cast_fp16 = matmul(transpose_x = matrix_ac_transpose_x_0, transpose_y = matrix_ac_transpose_y_0, x = transpose_142, y = transpose_143)[name = string("matrix_ac_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_begin_0 = const()[name = string("matrix_bd_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_end_0 = const()[name = string("matrix_bd_end_0"), val = tensor<int32, [4]>([1, 8, 128, 128])];
+            tensor<bool, [4]> matrix_bd_end_mask_0 = const()[name = string("matrix_bd_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 128, 128]> matrix_bd_cast_fp16 = slice_by_index(begin = matrix_bd_begin_0, end = matrix_bd_end_0, end_mask = matrix_bd_end_mask_0, x = matrix_bd_93_cast_fp16)[name = string("matrix_bd_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_4044_cast_fp16 = add(x = matrix_ac_cast_fp16, y = matrix_bd_cast_fp16)[name = string("op_4044_cast_fp16")];
+            fp16 _inversed_scores_93_y_0_to_fp16 = const()[name = string("_inversed_scores_93_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 128, 128]> _inversed_scores_93_cast_fp16 = mul(x = var_4044_cast_fp16, y = _inversed_scores_93_y_0_to_fp16)[name = string("_inversed_scores_93_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> scores_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_93_cast_fp16, cond = mask_3)[name = string("scores_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> var_4050_cast_fp16 = softmax(axis = var_30, x = scores_cast_fp16)[name = string("op_4050_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 128]> input_1229_cast_fp16 = select(a = var_11_to_fp16, b = var_4050_cast_fp16, cond = mask_3)[name = string("input_1229_cast_fp16")];
+            bool x_519_transpose_x_0 = const()[name = string("x_519_transpose_x_0"), val = bool(false)];
+            bool x_519_transpose_y_0 = const()[name = string("x_519_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 128]> value_cast_fp16 = transpose(perm = value_perm_0, x = v_cast_fp16)[name = string("transpose_148")];
+            tensor<fp16, [1, 8, 128, 128]> x_519_cast_fp16 = matmul(transpose_x = x_519_transpose_x_0, transpose_y = x_519_transpose_y_0, x = input_1229_cast_fp16, y = value_cast_fp16)[name = string("x_519_cast_fp16")];
+            tensor<int32, [4]> var_4054_perm_0 = const()[name = string("op_4054_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_4055 = const()[name = string("op_4055"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 128, 8, 128]> var_4054_cast_fp16 = transpose(perm = var_4054_perm_0, x = x_519_cast_fp16)[name = string("transpose_147")];
+            tensor<fp16, [1, 128, 1024]> input_1231_cast_fp16 = reshape(shape = var_4055, x = var_4054_cast_fp16)[name = string("input_1231_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [527827]> module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301003456))), indices_nonzero_data = tensor<uint6, [527827]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301134592))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301530560))))[name = string("module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_214_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_1231_cast_fp16)[name = string("linear_214_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_1235_cast_fp16 = add(x = input_1227_cast_fp16, y = linear_214_cast_fp16)[name = string("input_1235_cast_fp16")];
+            tensor<int32, [1]> x_523_axes_0 = const()[name = string("x_523_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_23_norm_conv_weight_to_fp16 = const()[name = string("module_layers_23_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301547008)))];
+            tensor<fp16, [1024]> module_layers_23_norm_conv_bias_to_fp16 = const()[name = string("module_layers_23_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301549120)))];
+            tensor<fp16, [1, 128, 1024]> x_523_cast_fp16 = layer_norm(axes = x_523_axes_0, beta = module_layers_23_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_conv_weight_to_fp16, x = input_1235_cast_fp16)[name = string("x_523_cast_fp16")];
+            tensor<int32, [3]> input_1237_perm_0 = const()[name = string("input_1237_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_1239_pad_type_0 = const()[name = string("input_1239_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_1239_strides_0 = const()[name = string("input_1239_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1239_pad_0 = const()[name = string("input_1239_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1239_dilations_0 = const()[name = string("input_1239_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_1239_groups_0 = const()[name = string("input_1239_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1051649]> module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301551232))), indices_nonzero_data = tensor<uint6, [1051649]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301813440))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302602304))))[name = string("module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 128]> input_1237_cast_fp16 = transpose(perm = input_1237_perm_0, x = x_523_cast_fp16)[name = string("transpose_146")];
+            tensor<fp16, [1, 2048, 128]> input_1239_cast_fp16 = conv(dilations = input_1239_dilations_0, groups = input_1239_groups_0, pad = input_1239_pad_0, pad_type = input_1239_pad_type_0, strides = input_1239_strides_0, weight = module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_1237_cast_fp16)[name = string("input_1239_cast_fp16")];
+            int32 x_525_split_num_splits_0 = const()[name = string("x_525_split_num_splits_0"), val = int32(2)];
+            int32 x_525_split_axis_0 = const()[name = string("x_525_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 128]> x_525_split_cast_fp16_0, tensor<fp16, [1, 1024, 128]> x_525_split_cast_fp16_1 = split(axis = x_525_split_axis_0, num_splits = x_525_split_num_splits_0, x = input_1239_cast_fp16)[name = string("x_525_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_525_split_1_sigmoid_cast_fp16 = sigmoid(x = x_525_split_cast_fp16_1)[name = string("x_525_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> x_525_cast_fp16 = mul(x = x_525_split_cast_fp16_0, y = x_525_split_1_sigmoid_cast_fp16)[name = string("x_525_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_1241_cast_fp16 = select(a = var_11_to_fp16, b = x_525_cast_fp16, cond = var_328)[name = string("input_1241_cast_fp16")];
+            tensor<int32, [6]> input_1243_pad_0 = const()[name = string("input_1243_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_1243_mode_0 = const()[name = string("input_1243_mode_0"), val = string("constant")];
+            fp16 const_247_to_fp16 = const()[name = string("const_247_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 136]> input_1243_cast_fp16 = pad(constant_val = const_247_to_fp16, mode = input_1243_mode_0, pad = input_1243_pad_0, x = input_1241_cast_fp16)[name = string("input_1243_cast_fp16")];
+            string input_1245_pad_type_0 = const()[name = string("input_1245_pad_type_0"), val = string("valid")];
+            int32 input_1245_groups_0 = const()[name = string("input_1245_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_1245_strides_0 = const()[name = string("input_1245_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1245_pad_0 = const()[name = string("input_1245_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1245_dilations_0 = const()[name = string("input_1245_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_294_to_fp16 = const()[name = string("const_294_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302635136)))];
+            tensor<fp16, [1024]> const_295_to_fp16 = const()[name = string("const_295_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302653632)))];
+            tensor<fp16, [1, 1024, 128]> input_1247_cast_fp16 = conv(bias = const_295_to_fp16, dilations = input_1245_dilations_0, groups = input_1245_groups_0, pad = input_1245_pad_0, pad_type = input_1245_pad_type_0, strides = input_1245_strides_0, weight = const_294_to_fp16, x = input_1243_cast_fp16)[name = string("input_1247_cast_fp16")];
+            tensor<fp16, [1, 1024, 128]> input_1249_cast_fp16 = silu(x = input_1247_cast_fp16)[name = string("input_1249_cast_fp16")];
+            string x_527_pad_type_0 = const()[name = string("x_527_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_527_strides_0 = const()[name = string("x_527_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_527_pad_0 = const()[name = string("x_527_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_527_dilations_0 = const()[name = string("x_527_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_527_groups_0 = const()[name = string("x_527_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519222]> module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302655744))), indices_nonzero_data = tensor<uint6, [519222]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302786880))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303176384))))[name = string("module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 128]> x_527_cast_fp16 = conv(dilations = x_527_dilations_0, groups = x_527_groups_0, pad = x_527_pad_0, pad_type = x_527_pad_type_0, strides = x_527_strides_0, weight = module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_1249_cast_fp16)[name = string("x_527_cast_fp16")];
+            tensor<int32, [3]> input_1251_perm_0 = const()[name = string("input_1251_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 128, 1024]> input_1251_cast_fp16 = transpose(perm = input_1251_perm_0, x = x_527_cast_fp16)[name = string("transpose_145")];
+            tensor<fp16, [1, 128, 1024]> input_1253_cast_fp16 = add(x = input_1235_cast_fp16, y = input_1251_cast_fp16)[name = string("input_1253_cast_fp16")];
+            tensor<int32, [1]> input_1255_axes_0 = const()[name = string("input_1255_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_23_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_23_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303192832)))];
+            tensor<fp16, [1024]> module_layers_23_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_23_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303194944)))];
+            tensor<fp16, [1, 128, 1024]> input_1255_cast_fp16 = layer_norm(axes = input_1255_axes_0, beta = module_layers_23_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_feed_forward2_weight_to_fp16, x = input_1253_cast_fp16)[name = string("input_1255_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2106054]> module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303197056))), indices_nonzero_data = tensor<uint6, [2106054]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303721408))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305301056))))[name = string("module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 4096]> linear_215_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_1255_cast_fp16)[name = string("linear_215_cast_fp16")];
+            tensor<fp16, [1, 128, 4096]> input_1259_cast_fp16 = silu(x = linear_215_cast_fp16)[name = string("input_1259_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2012639]> module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305366656))), indices_nonzero_data = tensor<uint6, [2012639]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305891008))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307400576))))[name = string("module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 128, 1024]> linear_216_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_1259_cast_fp16)[name = string("linear_216_cast_fp16")];
+            fp16 var_4115_to_fp16 = const()[name = string("op_4115_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 128, 1024]> var_4116_cast_fp16 = mul(x = linear_216_cast_fp16, y = var_4115_to_fp16)[name = string("op_4116_cast_fp16")];
+            tensor<fp16, [1, 128, 1024]> input_cast_fp16 = add(x = input_1253_cast_fp16, y = var_4116_cast_fp16)[name = string("input_cast_fp16")];
+            tensor<int32, [1]> audio_signal_axes_0 = const()[name = string("audio_signal_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_23_norm_out_weight_to_fp16 = const()[name = string("module_layers_23_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307417024)))];
+            tensor<fp16, [1024]> module_layers_23_norm_out_bias_to_fp16 = const()[name = string("module_layers_23_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307419136)))];
+            tensor<fp16, [1, 128, 1024]> audio_signal_cast_fp16 = layer_norm(axes = audio_signal_axes_0, beta = module_layers_23_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_out_weight_to_fp16, x = input_cast_fp16)[name = string("audio_signal_cast_fp16")];
+            tensor<int32, [3]> obj_1_perm_0 = const()[name = string("obj_1_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string obj_1_cast_fp16_to_fp32_dtype_0 = const()[name = string("obj_1_cast_fp16_to_fp32_dtype_0"), val = string("fp32")];
+            tensor<fp16, [1, 1024, 128]> obj_1_cast_fp16 = transpose(perm = obj_1_perm_0, x = audio_signal_cast_fp16)[name = string("transpose_144")];
+            tensor<fp32, [1, 1024, 128]> encoder = cast(dtype = obj_1_cast_fp16_to_fp32_dtype_0, x = obj_1_cast_fp16)[name = string("cast_0")];
+        } -> (encoder, encoder_length);
+    func enc_15000ms<ios18>(tensor<fp32, [1, 128, 1501]> mel, tensor<int32, [1]> mel_length) {
+            int32 var_30 = const()[name = string("op_30"), val = int32(-1)];
+            tensor<int32, [3]> x_1_perm_0 = const()[name = string("x_1_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string mel_to_fp16_dtype_0 = const()[name = string("mel_to_fp16_dtype_0"), val = string("fp16")];
+            string var_86_to_fp16_dtype_0 = const()[name = string("op_86_to_fp16_dtype_0"), val = string("fp16")];
+            fp16 var_87_promoted_to_fp16 = const()[name = string("op_87_promoted_to_fp16"), val = fp16(-0x1p+0)];
+            tensor<fp16, [1]> mel_length_to_fp16 = cast(dtype = var_86_to_fp16_dtype_0, x = mel_length)[name = string("cast_3")];
+            tensor<fp16, [1]> var_88_cast_fp16 = add(x = mel_length_to_fp16, y = var_87_promoted_to_fp16)[name = string("op_88_cast_fp16")];
+            fp16 _inversed_90_y_0_to_fp16 = const()[name = string("_inversed_90_y_0_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1]> _inversed_90_cast_fp16 = mul(x = var_88_cast_fp16, y = _inversed_90_y_0_to_fp16)[name = string("_inversed_90_cast_fp16")];
+            fp16 var_91_to_fp16 = const()[name = string("op_91_to_fp16"), val = fp16(0x1p+0)];
+            tensor<fp16, [1]> lengths_1_cast_fp16 = add(x = _inversed_90_cast_fp16, y = var_91_to_fp16)[name = string("lengths_1_cast_fp16")];
+            tensor<fp16, [1]> lengths_3_cast_fp16 = floor(x = lengths_1_cast_fp16)[name = string("lengths_3_cast_fp16")];
+            fp16 var_95_promoted_to_fp16 = const()[name = string("op_95_promoted_to_fp16"), val = fp16(-0x1p+0)];
+            tensor<fp16, [1]> var_96_cast_fp16 = add(x = lengths_3_cast_fp16, y = var_95_promoted_to_fp16)[name = string("op_96_cast_fp16")];
+            fp16 _inversed_98_y_0_to_fp16 = const()[name = string("_inversed_98_y_0_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1]> _inversed_98_cast_fp16 = mul(x = var_96_cast_fp16, y = _inversed_98_y_0_to_fp16)[name = string("_inversed_98_cast_fp16")];
+            fp16 var_99_to_fp16 = const()[name = string("op_99_to_fp16"), val = fp16(0x1p+0)];
+            tensor<fp16, [1]> lengths_7_cast_fp16 = add(x = _inversed_98_cast_fp16, y = var_99_to_fp16)[name = string("lengths_7_cast_fp16")];
+            tensor<fp16, [1]> lengths_9_cast_fp16 = floor(x = lengths_7_cast_fp16)[name = string("lengths_9_cast_fp16")];
+            fp16 var_103_promoted_to_fp16 = const()[name = string("op_103_promoted_to_fp16"), val = fp16(-0x1p+0)];
+            tensor<fp16, [1]> var_104_cast_fp16 = add(x = lengths_9_cast_fp16, y = var_103_promoted_to_fp16)[name = string("op_104_cast_fp16")];
+            fp16 _inversed_106_y_0_to_fp16 = const()[name = string("_inversed_106_y_0_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1]> _inversed_106_cast_fp16 = mul(x = var_104_cast_fp16, y = _inversed_106_y_0_to_fp16)[name = string("_inversed_106_cast_fp16")];
+            fp16 var_107_to_fp16 = const()[name = string("op_107_to_fp16"), val = fp16(0x1p+0)];
+            tensor<fp16, [1]> lengths_13_cast_fp16 = add(x = _inversed_106_cast_fp16, y = var_107_to_fp16)[name = string("lengths_13_cast_fp16")];
+            tensor<fp16, [1]> lengths_cast_fp16 = floor(x = lengths_13_cast_fp16)[name = string("lengths_cast_fp16")];
+            tensor<int32, [1]> input_1_axes_0 = const()[name = string("input_1_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1, 128, 1501]> mel_to_fp16 = cast(dtype = mel_to_fp16_dtype_0, x = mel)[name = string("cast_2")];
+            tensor<fp16, [1, 1501, 128]> x_1_cast_fp16 = transpose(perm = x_1_perm_0, x = mel_to_fp16)[name = string("transpose_315")];
+            tensor<fp16, [1, 1, 1501, 128]> input_1_cast_fp16 = expand_dims(axes = input_1_axes_0, x = x_1_cast_fp16)[name = string("input_1_cast_fp16")];
+            string input_3_pad_type_0 = const()[name = string("input_3_pad_type_0"), val = string("custom")];
+            tensor<int32, [4]> input_3_pad_0 = const()[name = string("input_3_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
+            tensor<int32, [2]> input_3_strides_0 = const()[name = string("input_3_strides_0"), val = tensor<int32, [2]>([2, 2])];
+            tensor<int32, [2]> input_3_dilations_0 = const()[name = string("input_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
+            int32 input_3_groups_0 = const()[name = string("input_3_groups_0"), val = int32(1)];
+            tensor<fp16, [256, 1, 3, 3]> module_pre_encode_conv_0_weight_to_fp16 = const()[name = string("module_pre_encode_conv_0_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
+            tensor<fp16, [256]> module_pre_encode_conv_0_bias_to_fp16 = const()[name = string("module_pre_encode_conv_0_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4736)))];
+            tensor<fp16, [1, 256, 751, 64]> input_3_cast_fp16 = conv(bias = module_pre_encode_conv_0_bias_to_fp16, dilations = input_3_dilations_0, groups = input_3_groups_0, pad = input_3_pad_0, pad_type = input_3_pad_type_0, strides = input_3_strides_0, weight = module_pre_encode_conv_0_weight_to_fp16, x = input_1_cast_fp16)[name = string("input_3_cast_fp16")];
+            tensor<fp16, [1, 256, 751, 64]> input_5_cast_fp16 = relu(x = input_3_cast_fp16)[name = string("input_5_cast_fp16")];
+            string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("custom")];
+            tensor<int32, [4]> input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
+            tensor<int32, [2]> input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor<int32, [2]>([2, 2])];
+            int32 input_7_groups_0 = const()[name = string("input_7_groups_0"), val = int32(256)];
+            tensor<int32, [2]> input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<fp16, [256, 1, 3, 3]> module_pre_encode_conv_2_weight_to_fp16 = const()[name = string("module_pre_encode_conv_2_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5312)))];
+            tensor<fp16, [256]> module_pre_encode_conv_2_bias_to_fp16 = const()[name = string("module_pre_encode_conv_2_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9984)))];
+            tensor<fp16, [1, 256, 376, 32]> input_7_cast_fp16 = conv(bias = module_pre_encode_conv_2_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = module_pre_encode_conv_2_weight_to_fp16, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")];
+            string input_9_pad_type_0 = const()[name = string("input_9_pad_type_0"), val = string("valid")];
+            tensor<int32, [2]> input_9_strides_0 = const()[name = string("input_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [4]> input_9_pad_0 = const()[name = string("input_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [2]> input_9_dilations_0 = const()[name = string("input_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
+            int32 input_9_groups_0 = const()[name = string("input_9_groups_0"), val = int32(1)];
+            tensor<fp16, [256, 256, 1, 1]> module_pre_encode_conv_3_weight_to_fp16 = const()[name = string("module_pre_encode_conv_3_weight_to_fp16"), val = tensor<fp16, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10560)))];
+            tensor<fp16, [256]> module_pre_encode_conv_3_bias_to_fp16 = const()[name = string("module_pre_encode_conv_3_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141696)))];
+            tensor<fp16, [1, 256, 376, 32]> input_9_cast_fp16 = conv(bias = module_pre_encode_conv_3_bias_to_fp16, dilations = input_9_dilations_0, groups = input_9_groups_0, pad = input_9_pad_0, pad_type = input_9_pad_type_0, strides = input_9_strides_0, weight = module_pre_encode_conv_3_weight_to_fp16, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")];
+            tensor<fp16, [1, 256, 376, 32]> input_11_cast_fp16 = relu(x = input_9_cast_fp16)[name = string("input_11_cast_fp16")];
+            string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("custom")];
+            tensor<int32, [4]> input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
+            tensor<int32, [2]> input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor<int32, [2]>([2, 2])];
+            int32 input_13_groups_0 = const()[name = string("input_13_groups_0"), val = int32(256)];
+            tensor<int32, [2]> input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<fp16, [256, 1, 3, 3]> module_pre_encode_conv_5_weight_to_fp16 = const()[name = string("module_pre_encode_conv_5_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142272)))];
+            tensor<fp16, [256]> module_pre_encode_conv_5_bias_to_fp16 = const()[name = string("module_pre_encode_conv_5_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146944)))];
+            tensor<fp16, [1, 256, 188, 16]> input_13_cast_fp16 = conv(bias = module_pre_encode_conv_5_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = module_pre_encode_conv_5_weight_to_fp16, x = input_11_cast_fp16)[name = string("input_13_cast_fp16")];
+            string input_15_pad_type_0 = const()[name = string("input_15_pad_type_0"), val = string("valid")];
+            tensor<int32, [2]> input_15_strides_0 = const()[name = string("input_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [4]> input_15_pad_0 = const()[name = string("input_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [2]> input_15_dilations_0 = const()[name = string("input_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
+            int32 input_15_groups_0 = const()[name = string("input_15_groups_0"), val = int32(1)];
+            tensor<fp16, [256, 256, 1, 1]> module_pre_encode_conv_6_weight_to_fp16 = const()[name = string("module_pre_encode_conv_6_weight_to_fp16"), val = tensor<fp16, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147520)))];
+            tensor<fp16, [256]> module_pre_encode_conv_6_bias_to_fp16 = const()[name = string("module_pre_encode_conv_6_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278656)))];
+            tensor<fp16, [1, 256, 188, 16]> input_15_cast_fp16 = conv(bias = module_pre_encode_conv_6_bias_to_fp16, dilations = input_15_dilations_0, groups = input_15_groups_0, pad = input_15_pad_0, pad_type = input_15_pad_type_0, strides = input_15_strides_0, weight = module_pre_encode_conv_6_weight_to_fp16, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")];
+            tensor<fp16, [1, 256, 188, 16]> x_3_cast_fp16 = relu(x = input_15_cast_fp16)[name = string("x_3_cast_fp16")];
+            tensor<int32, [4]> var_157_perm_0 = const()[name = string("op_157_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_158 = const()[name = string("op_158"), val = tensor<int32, [3]>([1, 188, -1])];
+            tensor<fp16, [1, 188, 256, 16]> var_157_cast_fp16 = transpose(perm = var_157_perm_0, x = x_3_cast_fp16)[name = string("transpose_314")];
+            tensor<fp16, [1, 188, 4096]> input_17_cast_fp16 = reshape(shape = var_158, x = var_157_cast_fp16)[name = string("input_17_cast_fp16")];
+            tensor<fp16, [1024, 4096]> module_pre_encode_out_weight_to_fp16 = const()[name = string("module_pre_encode_out_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279232)))];
+            tensor<fp16, [1024]> module_pre_encode_out_bias_to_fp16 = const()[name = string("module_pre_encode_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8667904)))];
+            tensor<fp16, [1, 188, 1024]> linear_0_cast_fp16 = linear(bias = module_pre_encode_out_bias_to_fp16, weight = module_pre_encode_out_weight_to_fp16, x = input_17_cast_fp16)[name = string("linear_0_cast_fp16")];
+            string padding_length_dtype_0 = const()[name = string("padding_length_dtype_0"), val = string("int32")];
+            tensor<int32, [1, 188]> expand_dims_0 = const()[name = string("expand_dims_0"), val = tensor<int32, [1, 188]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(319957120)))];
+            tensor<int32, [1]> var_196_axes_0 = const()[name = string("op_196_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<int32, [1]> encoder_length = cast(dtype = padding_length_dtype_0, x = lengths_cast_fp16)[name = string("cast_1")];
+            tensor<int32, [1, 1]> var_196 = expand_dims(axes = var_196_axes_0, x = encoder_length)[name = string("op_196")];
+            tensor<bool, [1, 188]> pad_mask_1 = less(x = expand_dims_0, y = var_196)[name = string("pad_mask_1")];
+            tensor<int32, [1]> var_198_axes_0 = const()[name = string("op_198_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<bool, [1, 1, 188]> var_198 = expand_dims(axes = var_198_axes_0, x = pad_mask_1)[name = string("op_198")];
+            tensor<int32, [3]> var_199 = const()[name = string("op_199"), val = tensor<int32, [3]>([1, 188, 1])];
+            tensor<bool, [1, 188, 188]> pad_mask_for_att_mask_1 = tile(reps = var_199, x = var_198)[name = string("pad_mask_for_att_mask_1")];
+            tensor<int32, [3]> var_201_perm_0 = const()[name = string("op_201_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<bool, [1, 188, 188]> var_201 = transpose(perm = var_201_perm_0, x = pad_mask_for_att_mask_1)[name = string("transpose_313")];
+            tensor<bool, [1, 188, 188]> pad_mask_for_att_mask = logical_and(x = pad_mask_for_att_mask_1, y = var_201)[name = string("pad_mask_for_att_mask")];
+            tensor<bool, [1, 188, 188]> const_7 = const()[name = string("const_7"), val = tensor<bool, [1, 188, 188]>([[[true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true]]])];
+            tensor<bool, [1, 188, 188]> att_mask = logical_and(x = pad_mask_for_att_mask, y = const_7)[name = string("att_mask")];
+            tensor<bool, [1, 188, 188]> mask_1 = logical_not(x = att_mask)[name = string("mask_1")];
+            tensor<bool, [1, 188]> pad_mask = logical_not(x = pad_mask_1)[name = string("pad_mask")];
+            tensor<int32, [1]> input_21_axes_0 = const()[name = string("input_21_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_0_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_0_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8670336)))];
+            tensor<fp16, [1024]> module_layers_0_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_0_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8672448)))];
+            fp16 var_9_to_fp16 = const()[name = string("op_9_to_fp16"), val = fp16(0x1.5p-17)];
+            tensor<fp16, [1, 188, 1024]> input_21_cast_fp16 = layer_norm(axes = input_21_axes_0, beta = module_layers_0_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_feed_forward1_weight_to_fp16, x = linear_0_cast_fp16)[name = string("input_21_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2072390]> module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8674560))), indices_nonzero_data = tensor<uint6, [2072390]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9198912))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10753280))))[name = string("module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [4096]> linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10818880)))];
+            tensor<fp16, [1, 188, 4096]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_21_cast_fp16)[name = string("linear_1_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_25_cast_fp16 = silu(x = linear_1_cast_fp16)[name = string("input_25_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2038634]> module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10827136))), indices_nonzero_data = tensor<uint6, [2038634]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11351488))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12880576))))[name = string("module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1024]> linear_2_bias_0_to_fp16 = const()[name = string("linear_2_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12897024)))];
+            tensor<fp16, [1, 188, 1024]> linear_2_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_25_cast_fp16)[name = string("linear_2_cast_fp16")];
+            fp16 var_232_to_fp16 = const()[name = string("op_232_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_233_cast_fp16 = mul(x = linear_2_cast_fp16, y = var_232_to_fp16)[name = string("op_233_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_31_cast_fp16 = add(x = linear_0_cast_fp16, y = var_233_cast_fp16)[name = string("input_31_cast_fp16")];
+            tensor<int32, [1]> query_1_axes_0 = const()[name = string("query_1_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_0_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_0_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12899136)))];
+            tensor<fp16, [1024]> module_layers_0_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_0_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12901248)))];
+            tensor<fp16, [1, 188, 1024]> query_1_cast_fp16 = layer_norm(axes = query_1_axes_0, beta = module_layers_0_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_self_att_weight_to_fp16, x = input_31_cast_fp16)[name = string("query_1_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522455]> module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12903360))), indices_nonzero_data = tensor<uint6, [522455]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13034496))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13426432))))[name = string("module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_3_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_1_cast_fp16)[name = string("linear_3_cast_fp16")];
+            tensor<int32, [4]> var_249 = const()[name = string("op_249"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_1_cast_fp16 = reshape(shape = var_249, x = linear_3_cast_fp16)[name = string("q_1_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [515024]> module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13442880))), indices_nonzero_data = tensor<uint6, [515024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13574016))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13960384))))[name = string("module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_4_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_1_cast_fp16)[name = string("linear_4_cast_fp16")];
+            tensor<int32, [4]> var_253 = const()[name = string("op_253"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_1_cast_fp16 = reshape(shape = var_253, x = linear_4_cast_fp16)[name = string("k_1_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [505889]> module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13976832))), indices_nonzero_data = tensor<uint6, [505889]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14107968))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14487488))))[name = string("module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_5_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_1_cast_fp16)[name = string("linear_5_cast_fp16")];
+            tensor<int32, [4]> var_257 = const()[name = string("op_257"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_1_cast_fp16 = reshape(shape = var_257, x = linear_5_cast_fp16)[name = string("v_1_cast_fp16")];
+            tensor<int32, [4]> value_3_perm_0 = const()[name = string("value_3_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_0_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_0_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14503936)))];
+            tensor<fp16, [1, 188, 8, 128]> var_269_cast_fp16 = add(x = q_1_cast_fp16, y = module_layers_0_self_attn_pos_bias_u_to_fp16)[name = string("op_269_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_0_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_0_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14506048)))];
+            tensor<fp16, [1, 188, 8, 128]> var_271_cast_fp16 = add(x = q_1_cast_fp16, y = module_layers_0_self_attn_pos_bias_v_to_fp16)[name = string("op_271_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_1_perm_0 = const()[name = string("q_with_bias_v_1_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_7_transpose_x_0 = const()[name = string("x_7_transpose_x_0"), val = bool(false)];
+            bool x_7_transpose_y_0 = const()[name = string("x_7_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_273_to_fp16 = const()[name = string("op_273_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(319957952)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_1_cast_fp16 = transpose(perm = q_with_bias_v_1_perm_0, x = var_271_cast_fp16)[name = string("transpose_312")];
+            tensor<fp16, [1, 8, 188, 375]> x_7_cast_fp16 = matmul(transpose_x = x_7_transpose_x_0, transpose_y = x_7_transpose_y_0, x = q_with_bias_v_1_cast_fp16, y = var_273_to_fp16)[name = string("x_7_cast_fp16")];
+            tensor<int32, [8]> x_9_pad_0 = const()[name = string("x_9_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_9_mode_0 = const()[name = string("x_9_mode_0"), val = string("constant")];
+            fp16 const_14_to_fp16 = const()[name = string("const_14_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_9_cast_fp16 = pad(constant_val = const_14_to_fp16, mode = x_9_mode_0, pad = x_9_pad_0, x = x_7_cast_fp16)[name = string("x_9_cast_fp16")];
+            tensor<int32, [4]> var_281 = const()[name = string("op_281"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_11_cast_fp16 = reshape(shape = var_281, x = x_9_cast_fp16)[name = string("x_11_cast_fp16")];
+            tensor<int32, [4]> var_285_begin_0 = const()[name = string("op_285_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_285_end_0 = const()[name = string("op_285_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_285_end_mask_0 = const()[name = string("op_285_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_285_cast_fp16 = slice_by_index(begin = var_285_begin_0, end = var_285_end_0, end_mask = var_285_end_mask_0, x = x_11_cast_fp16)[name = string("op_285_cast_fp16")];
+            tensor<int32, [4]> var_286 = const()[name = string("op_286"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_1_cast_fp16 = reshape(shape = var_286, x = var_285_cast_fp16)[name = string("matrix_bd_1_cast_fp16")];
+            bool matrix_ac_1_transpose_x_0 = const()[name = string("matrix_ac_1_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_1_transpose_y_0 = const()[name = string("matrix_ac_1_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_96_perm_0 = const()[name = string("transpose_96_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_97_perm_0 = const()[name = string("transpose_97_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_97 = transpose(perm = transpose_97_perm_0, x = k_1_cast_fp16)[name = string("transpose_310")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_96 = transpose(perm = transpose_96_perm_0, x = var_269_cast_fp16)[name = string("transpose_311")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_1_cast_fp16 = matmul(transpose_x = matrix_ac_1_transpose_x_0, transpose_y = matrix_ac_1_transpose_y_0, x = transpose_96, y = transpose_97)[name = string("matrix_ac_1_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_3_begin_0 = const()[name = string("matrix_bd_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_3_end_0 = const()[name = string("matrix_bd_3_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_3_end_mask_0 = const()[name = string("matrix_bd_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_3_cast_fp16 = slice_by_index(begin = matrix_bd_3_begin_0, end = matrix_bd_3_end_0, end_mask = matrix_bd_3_end_mask_0, x = matrix_bd_1_cast_fp16)[name = string("matrix_bd_3_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_295_cast_fp16 = add(x = matrix_ac_1_cast_fp16, y = matrix_bd_3_cast_fp16)[name = string("op_295_cast_fp16")];
+            fp16 _inversed_scores_1_y_0_to_fp16 = const()[name = string("_inversed_scores_1_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_1_cast_fp16 = mul(x = var_295_cast_fp16, y = _inversed_scores_1_y_0_to_fp16)[name = string("_inversed_scores_1_cast_fp16")];
+            tensor<int32, [1]> mask_3_axes_0 = const()[name = string("mask_3_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<bool, [1, 1, 188, 188]> mask_3 = expand_dims(axes = mask_3_axes_0, x = mask_1)[name = string("mask_3")];
+            fp16 var_12_to_fp16 = const()[name = string("op_12_to_fp16"), val = fp16(-0x1.388p+13)];
+            tensor<fp16, [1, 8, 188, 188]> scores_3_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_1_cast_fp16, cond = mask_3)[name = string("scores_3_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_301_cast_fp16 = softmax(axis = var_30, x = scores_3_cast_fp16)[name = string("op_301_cast_fp16")];
+            fp16 var_11_to_fp16 = const()[name = string("op_11_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 188]> input_33_cast_fp16 = select(a = var_11_to_fp16, b = var_301_cast_fp16, cond = mask_3)[name = string("input_33_cast_fp16")];
+            bool x_13_transpose_x_0 = const()[name = string("x_13_transpose_x_0"), val = bool(false)];
+            bool x_13_transpose_y_0 = const()[name = string("x_13_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_3_cast_fp16 = transpose(perm = value_3_perm_0, x = v_1_cast_fp16)[name = string("transpose_309")];
+            tensor<fp16, [1, 8, 188, 128]> x_13_cast_fp16 = matmul(transpose_x = x_13_transpose_x_0, transpose_y = x_13_transpose_y_0, x = input_33_cast_fp16, y = value_3_cast_fp16)[name = string("x_13_cast_fp16")];
+            tensor<int32, [4]> var_305_perm_0 = const()[name = string("op_305_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_306 = const()[name = string("op_306"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_305_cast_fp16 = transpose(perm = var_305_perm_0, x = x_13_cast_fp16)[name = string("transpose_308")];
+            tensor<fp16, [1, 188, 1024]> input_35_cast_fp16 = reshape(shape = var_306, x = var_305_cast_fp16)[name = string("input_35_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [517310]> module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14768320))), indices_nonzero_data = tensor<uint6, [517310]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14899456))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15287552))))[name = string("module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_7_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_35_cast_fp16)[name = string("linear_7_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_39_cast_fp16 = add(x = input_31_cast_fp16, y = linear_7_cast_fp16)[name = string("input_39_cast_fp16")];
+            tensor<int32, [1]> x_17_axes_0 = const()[name = string("x_17_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_0_norm_conv_weight_to_fp16 = const()[name = string("module_layers_0_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15304000)))];
+            tensor<fp16, [1024]> module_layers_0_norm_conv_bias_to_fp16 = const()[name = string("module_layers_0_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15306112)))];
+            tensor<fp16, [1, 188, 1024]> x_17_cast_fp16 = layer_norm(axes = x_17_axes_0, beta = module_layers_0_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_conv_weight_to_fp16, x = input_39_cast_fp16)[name = string("x_17_cast_fp16")];
+            tensor<int32, [3]> input_41_perm_0 = const()[name = string("input_41_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_43_pad_type_0 = const()[name = string("input_43_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_43_strides_0 = const()[name = string("input_43_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_43_pad_0 = const()[name = string("input_43_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_43_dilations_0 = const()[name = string("input_43_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_43_groups_0 = const()[name = string("input_43_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1033148]> module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15308224))), indices_nonzero_data = tensor<uint6, [1033148]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15570432))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16345408))))[name = string("module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_41_cast_fp16 = transpose(perm = input_41_perm_0, x = x_17_cast_fp16)[name = string("transpose_307")];
+            tensor<fp16, [1, 2048, 188]> input_43_cast_fp16 = conv(dilations = input_43_dilations_0, groups = input_43_groups_0, pad = input_43_pad_0, pad_type = input_43_pad_type_0, strides = input_43_strides_0, weight = module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_41_cast_fp16)[name = string("input_43_cast_fp16")];
+            int32 x_19_split_num_splits_0 = const()[name = string("x_19_split_num_splits_0"), val = int32(2)];
+            int32 x_19_split_axis_0 = const()[name = string("x_19_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_19_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_19_split_cast_fp16_1 = split(axis = x_19_split_axis_0, num_splits = x_19_split_num_splits_0, x = input_43_cast_fp16)[name = string("x_19_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_19_split_1_sigmoid_cast_fp16 = sigmoid(x = x_19_split_cast_fp16_1)[name = string("x_19_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_19_cast_fp16 = mul(x = x_19_split_cast_fp16_0, y = x_19_split_1_sigmoid_cast_fp16)[name = string("x_19_cast_fp16")];
+            tensor<int32, [1]> var_328_axes_0 = const()[name = string("op_328_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<bool, [1, 1, 188]> var_328 = expand_dims(axes = var_328_axes_0, x = pad_mask)[name = string("op_328")];
+            tensor<fp16, [1, 1024, 188]> input_45_cast_fp16 = select(a = var_11_to_fp16, b = x_19_cast_fp16, cond = var_328)[name = string("input_45_cast_fp16")];
+            tensor<int32, [6]> input_47_pad_0 = const()[name = string("input_47_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_47_mode_0 = const()[name = string("input_47_mode_0"), val = string("constant")];
+            fp16 const_17_to_fp16 = const()[name = string("const_17_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_47_cast_fp16 = pad(constant_val = const_17_to_fp16, mode = input_47_mode_0, pad = input_47_pad_0, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")];
+            string input_49_pad_type_0 = const()[name = string("input_49_pad_type_0"), val = string("valid")];
+            int32 input_49_groups_0 = const()[name = string("input_49_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_49_strides_0 = const()[name = string("input_49_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_49_pad_0 = const()[name = string("input_49_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_49_dilations_0 = const()[name = string("input_49_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_248_to_fp16 = const()[name = string("const_248_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16378240)))];
+            tensor<fp16, [1024]> const_249_to_fp16 = const()[name = string("const_249_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16396736)))];
+            tensor<fp16, [1, 1024, 188]> input_51_cast_fp16 = conv(bias = const_249_to_fp16, dilations = input_49_dilations_0, groups = input_49_groups_0, pad = input_49_pad_0, pad_type = input_49_pad_type_0, strides = input_49_strides_0, weight = const_248_to_fp16, x = input_47_cast_fp16)[name = string("input_51_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_53_cast_fp16 = silu(x = input_51_cast_fp16)[name = string("input_53_cast_fp16")];
+            string x_21_pad_type_0 = const()[name = string("x_21_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_21_strides_0 = const()[name = string("x_21_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_21_pad_0 = const()[name = string("x_21_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_21_dilations_0 = const()[name = string("x_21_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_21_groups_0 = const()[name = string("x_21_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [512831]> module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16398848))), indices_nonzero_data = tensor<uint6, [512831]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16529984))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16914688))))[name = string("module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_21_cast_fp16 = conv(dilations = x_21_dilations_0, groups = x_21_groups_0, pad = x_21_pad_0, pad_type = x_21_pad_type_0, strides = x_21_strides_0, weight = module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_53_cast_fp16)[name = string("x_21_cast_fp16")];
+            tensor<int32, [3]> input_55_perm_0 = const()[name = string("input_55_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_55_cast_fp16 = transpose(perm = input_55_perm_0, x = x_21_cast_fp16)[name = string("transpose_306")];
+            tensor<fp16, [1, 188, 1024]> input_57_cast_fp16 = add(x = input_39_cast_fp16, y = input_55_cast_fp16)[name = string("input_57_cast_fp16")];
+            tensor<int32, [1]> input_59_axes_0 = const()[name = string("input_59_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_0_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_0_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16931136)))];
+            tensor<fp16, [1024]> module_layers_0_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_0_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16933248)))];
+            tensor<fp16, [1, 188, 1024]> input_59_cast_fp16 = layer_norm(axes = input_59_axes_0, beta = module_layers_0_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_feed_forward2_weight_to_fp16, x = input_57_cast_fp16)[name = string("input_59_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2071218]> module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16935360))), indices_nonzero_data = tensor<uint6, [2071218]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17459712))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19013248))))[name = string("module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_8_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_59_cast_fp16)[name = string("linear_8_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_63_cast_fp16 = silu(x = linear_8_cast_fp16)[name = string("input_63_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2031949]> module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19078848))), indices_nonzero_data = tensor<uint6, [2031949]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19603200))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21127232))))[name = string("module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_9_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_63_cast_fp16)[name = string("linear_9_cast_fp16")];
+            fp16 var_366_to_fp16 = const()[name = string("op_366_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_367_cast_fp16 = mul(x = linear_9_cast_fp16, y = var_366_to_fp16)[name = string("op_367_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_69_cast_fp16 = add(x = input_57_cast_fp16, y = var_367_cast_fp16)[name = string("input_69_cast_fp16")];
+            tensor<int32, [1]> input_71_axes_0 = const()[name = string("input_71_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_0_norm_out_weight_to_fp16 = const()[name = string("module_layers_0_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21143680)))];
+            tensor<fp16, [1024]> module_layers_0_norm_out_bias_to_fp16 = const()[name = string("module_layers_0_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21145792)))];
+            tensor<fp16, [1, 188, 1024]> input_71_cast_fp16 = layer_norm(axes = input_71_axes_0, beta = module_layers_0_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_out_weight_to_fp16, x = input_69_cast_fp16)[name = string("input_71_cast_fp16")];
+            tensor<int32, [1]> input_73_axes_0 = const()[name = string("input_73_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_1_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_1_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21147904)))];
+            tensor<fp16, [1024]> module_layers_1_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_1_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21150016)))];
+            tensor<fp16, [1, 188, 1024]> input_73_cast_fp16 = layer_norm(axes = input_73_axes_0, beta = module_layers_1_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_feed_forward1_weight_to_fp16, x = input_71_cast_fp16)[name = string("input_73_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2039600]> module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21152128))), indices_nonzero_data = tensor<uint6, [2039600]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21676480))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23206272))))[name = string("module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_10_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_73_cast_fp16)[name = string("linear_10_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_77_cast_fp16 = silu(x = linear_10_cast_fp16)[name = string("input_77_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1944084]> module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23271872))), indices_nonzero_data = tensor<uint6, [1944084]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23796224))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25254400))))[name = string("module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_11_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_77_cast_fp16)[name = string("linear_11_cast_fp16")];
+            fp16 var_395_to_fp16 = const()[name = string("op_395_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_396_cast_fp16 = mul(x = linear_11_cast_fp16, y = var_395_to_fp16)[name = string("op_396_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_83_cast_fp16 = add(x = input_71_cast_fp16, y = var_396_cast_fp16)[name = string("input_83_cast_fp16")];
+            tensor<int32, [1]> query_3_axes_0 = const()[name = string("query_3_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_1_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_1_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25270848)))];
+            tensor<fp16, [1024]> module_layers_1_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_1_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25272960)))];
+            tensor<fp16, [1, 188, 1024]> query_3_cast_fp16 = layer_norm(axes = query_3_axes_0, beta = module_layers_1_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_self_att_weight_to_fp16, x = input_83_cast_fp16)[name = string("query_3_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523158]> module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25275072))), indices_nonzero_data = tensor<uint6, [523158]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25406208))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25798656))))[name = string("module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_12_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_3_cast_fp16)[name = string("linear_12_cast_fp16")];
+            tensor<int32, [4]> var_412 = const()[name = string("op_412"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_7_cast_fp16 = reshape(shape = var_412, x = linear_12_cast_fp16)[name = string("q_7_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [515603]> module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25815104))), indices_nonzero_data = tensor<uint6, [515603]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25946240))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26333056))))[name = string("module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_13_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_3_cast_fp16)[name = string("linear_13_cast_fp16")];
+            tensor<int32, [4]> var_416 = const()[name = string("op_416"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_5_cast_fp16 = reshape(shape = var_416, x = linear_13_cast_fp16)[name = string("k_5_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [511483]> module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26349504))), indices_nonzero_data = tensor<uint6, [511483]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26480640))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26864320))))[name = string("module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_14_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_3_cast_fp16)[name = string("linear_14_cast_fp16")];
+            tensor<int32, [4]> var_420 = const()[name = string("op_420"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_3_cast_fp16 = reshape(shape = var_420, x = linear_14_cast_fp16)[name = string("v_3_cast_fp16")];
+            tensor<int32, [4]> value_5_perm_0 = const()[name = string("value_5_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_1_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_1_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26880768)))];
+            tensor<fp16, [1, 188, 8, 128]> var_432_cast_fp16 = add(x = q_7_cast_fp16, y = module_layers_1_self_attn_pos_bias_u_to_fp16)[name = string("op_432_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_1_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_1_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26882880)))];
+            tensor<fp16, [1, 188, 8, 128]> var_434_cast_fp16 = add(x = q_7_cast_fp16, y = module_layers_1_self_attn_pos_bias_v_to_fp16)[name = string("op_434_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_3_perm_0 = const()[name = string("q_with_bias_v_3_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_29_transpose_x_0 = const()[name = string("x_29_transpose_x_0"), val = bool(false)];
+            bool x_29_transpose_y_0 = const()[name = string("x_29_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_436_to_fp16 = const()[name = string("op_436_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320726016)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_3_cast_fp16 = transpose(perm = q_with_bias_v_3_perm_0, x = var_434_cast_fp16)[name = string("transpose_305")];
+            tensor<fp16, [1, 8, 188, 375]> x_29_cast_fp16 = matmul(transpose_x = x_29_transpose_x_0, transpose_y = x_29_transpose_y_0, x = q_with_bias_v_3_cast_fp16, y = var_436_to_fp16)[name = string("x_29_cast_fp16")];
+            tensor<int32, [8]> x_31_pad_0 = const()[name = string("x_31_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_31_mode_0 = const()[name = string("x_31_mode_0"), val = string("constant")];
+            fp16 const_24_to_fp16 = const()[name = string("const_24_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_31_cast_fp16 = pad(constant_val = const_24_to_fp16, mode = x_31_mode_0, pad = x_31_pad_0, x = x_29_cast_fp16)[name = string("x_31_cast_fp16")];
+            tensor<int32, [4]> var_444 = const()[name = string("op_444"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_33_cast_fp16 = reshape(shape = var_444, x = x_31_cast_fp16)[name = string("x_33_cast_fp16")];
+            tensor<int32, [4]> var_448_begin_0 = const()[name = string("op_448_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_448_end_0 = const()[name = string("op_448_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_448_end_mask_0 = const()[name = string("op_448_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_448_cast_fp16 = slice_by_index(begin = var_448_begin_0, end = var_448_end_0, end_mask = var_448_end_mask_0, x = x_33_cast_fp16)[name = string("op_448_cast_fp16")];
+            tensor<int32, [4]> var_449 = const()[name = string("op_449"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_5_cast_fp16 = reshape(shape = var_449, x = var_448_cast_fp16)[name = string("matrix_bd_5_cast_fp16")];
+            bool matrix_ac_3_transpose_x_0 = const()[name = string("matrix_ac_3_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_3_transpose_y_0 = const()[name = string("matrix_ac_3_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_98_perm_0 = const()[name = string("transpose_98_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_99_perm_0 = const()[name = string("transpose_99_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_99 = transpose(perm = transpose_99_perm_0, x = k_5_cast_fp16)[name = string("transpose_303")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_98 = transpose(perm = transpose_98_perm_0, x = var_432_cast_fp16)[name = string("transpose_304")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_3_cast_fp16 = matmul(transpose_x = matrix_ac_3_transpose_x_0, transpose_y = matrix_ac_3_transpose_y_0, x = transpose_98, y = transpose_99)[name = string("matrix_ac_3_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_7_begin_0 = const()[name = string("matrix_bd_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_7_end_0 = const()[name = string("matrix_bd_7_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_7_end_mask_0 = const()[name = string("matrix_bd_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_7_cast_fp16 = slice_by_index(begin = matrix_bd_7_begin_0, end = matrix_bd_7_end_0, end_mask = matrix_bd_7_end_mask_0, x = matrix_bd_5_cast_fp16)[name = string("matrix_bd_7_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_458_cast_fp16 = add(x = matrix_ac_3_cast_fp16, y = matrix_bd_7_cast_fp16)[name = string("op_458_cast_fp16")];
+            fp16 _inversed_scores_5_y_0_to_fp16 = const()[name = string("_inversed_scores_5_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_5_cast_fp16 = mul(x = var_458_cast_fp16, y = _inversed_scores_5_y_0_to_fp16)[name = string("_inversed_scores_5_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_7_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_5_cast_fp16, cond = mask_3)[name = string("scores_7_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_464_cast_fp16 = softmax(axis = var_30, x = scores_7_cast_fp16)[name = string("op_464_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_85_cast_fp16 = select(a = var_11_to_fp16, b = var_464_cast_fp16, cond = mask_3)[name = string("input_85_cast_fp16")];
+            bool x_35_transpose_x_0 = const()[name = string("x_35_transpose_x_0"), val = bool(false)];
+            bool x_35_transpose_y_0 = const()[name = string("x_35_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_5_cast_fp16 = transpose(perm = value_5_perm_0, x = v_3_cast_fp16)[name = string("transpose_302")];
+            tensor<fp16, [1, 8, 188, 128]> x_35_cast_fp16 = matmul(transpose_x = x_35_transpose_x_0, transpose_y = x_35_transpose_y_0, x = input_85_cast_fp16, y = value_5_cast_fp16)[name = string("x_35_cast_fp16")];
+            tensor<int32, [4]> var_468_perm_0 = const()[name = string("op_468_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_469 = const()[name = string("op_469"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_468_cast_fp16 = transpose(perm = var_468_perm_0, x = x_35_cast_fp16)[name = string("transpose_301")];
+            tensor<fp16, [1, 188, 1024]> input_87_cast_fp16 = reshape(shape = var_469, x = var_468_cast_fp16)[name = string("input_87_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523005]> module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27145152))), indices_nonzero_data = tensor<uint6, [523005]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27276288))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27668608))))[name = string("module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_16_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_87_cast_fp16)[name = string("linear_16_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_91_cast_fp16 = add(x = input_83_cast_fp16, y = linear_16_cast_fp16)[name = string("input_91_cast_fp16")];
+            tensor<int32, [1]> x_39_axes_0 = const()[name = string("x_39_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_1_norm_conv_weight_to_fp16 = const()[name = string("module_layers_1_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27685056)))];
+            tensor<fp16, [1024]> module_layers_1_norm_conv_bias_to_fp16 = const()[name = string("module_layers_1_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27687168)))];
+            tensor<fp16, [1, 188, 1024]> x_39_cast_fp16 = layer_norm(axes = x_39_axes_0, beta = module_layers_1_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_conv_weight_to_fp16, x = input_91_cast_fp16)[name = string("x_39_cast_fp16")];
+            tensor<int32, [3]> input_93_perm_0 = const()[name = string("input_93_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_95_pad_type_0 = const()[name = string("input_95_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_95_strides_0 = const()[name = string("input_95_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_95_pad_0 = const()[name = string("input_95_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_95_dilations_0 = const()[name = string("input_95_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_95_groups_0 = const()[name = string("input_95_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1039540]> module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27689280))), indices_nonzero_data = tensor<uint6, [1039540]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27951488))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28731264))))[name = string("module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_93_cast_fp16 = transpose(perm = input_93_perm_0, x = x_39_cast_fp16)[name = string("transpose_300")];
+            tensor<fp16, [1, 2048, 188]> input_95_cast_fp16 = conv(dilations = input_95_dilations_0, groups = input_95_groups_0, pad = input_95_pad_0, pad_type = input_95_pad_type_0, strides = input_95_strides_0, weight = module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_93_cast_fp16)[name = string("input_95_cast_fp16")];
+            int32 x_41_split_num_splits_0 = const()[name = string("x_41_split_num_splits_0"), val = int32(2)];
+            int32 x_41_split_axis_0 = const()[name = string("x_41_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_41_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_41_split_cast_fp16_1 = split(axis = x_41_split_axis_0, num_splits = x_41_split_num_splits_0, x = input_95_cast_fp16)[name = string("x_41_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_41_split_1_sigmoid_cast_fp16 = sigmoid(x = x_41_split_cast_fp16_1)[name = string("x_41_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_41_cast_fp16 = mul(x = x_41_split_cast_fp16_0, y = x_41_split_1_sigmoid_cast_fp16)[name = string("x_41_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_97_cast_fp16 = select(a = var_11_to_fp16, b = x_41_cast_fp16, cond = var_328)[name = string("input_97_cast_fp16")];
+            tensor<int32, [6]> input_99_pad_0 = const()[name = string("input_99_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_99_mode_0 = const()[name = string("input_99_mode_0"), val = string("constant")];
+            fp16 const_27_to_fp16 = const()[name = string("const_27_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_99_cast_fp16 = pad(constant_val = const_27_to_fp16, mode = input_99_mode_0, pad = input_99_pad_0, x = input_97_cast_fp16)[name = string("input_99_cast_fp16")];
+            string input_101_pad_type_0 = const()[name = string("input_101_pad_type_0"), val = string("valid")];
+            int32 input_101_groups_0 = const()[name = string("input_101_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_101_strides_0 = const()[name = string("input_101_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_101_pad_0 = const()[name = string("input_101_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_101_dilations_0 = const()[name = string("input_101_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_250_to_fp16 = const()[name = string("const_250_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28764096)))];
+            tensor<fp16, [1024]> const_251_to_fp16 = const()[name = string("const_251_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28782592)))];
+            tensor<fp16, [1, 1024, 188]> input_103_cast_fp16 = conv(bias = const_251_to_fp16, dilations = input_101_dilations_0, groups = input_101_groups_0, pad = input_101_pad_0, pad_type = input_101_pad_type_0, strides = input_101_strides_0, weight = const_250_to_fp16, x = input_99_cast_fp16)[name = string("input_103_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_105_cast_fp16 = silu(x = input_103_cast_fp16)[name = string("input_105_cast_fp16")];
+            string x_43_pad_type_0 = const()[name = string("x_43_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_43_strides_0 = const()[name = string("x_43_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_43_pad_0 = const()[name = string("x_43_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_43_dilations_0 = const()[name = string("x_43_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_43_groups_0 = const()[name = string("x_43_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [515869]> module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28784704))), indices_nonzero_data = tensor<uint6, [515869]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28915840))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29302848))))[name = string("module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_43_cast_fp16 = conv(dilations = x_43_dilations_0, groups = x_43_groups_0, pad = x_43_pad_0, pad_type = x_43_pad_type_0, strides = x_43_strides_0, weight = module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_105_cast_fp16)[name = string("x_43_cast_fp16")];
+            tensor<int32, [3]> input_107_perm_0 = const()[name = string("input_107_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_107_cast_fp16 = transpose(perm = input_107_perm_0, x = x_43_cast_fp16)[name = string("transpose_299")];
+            tensor<fp16, [1, 188, 1024]> input_109_cast_fp16 = add(x = input_91_cast_fp16, y = input_107_cast_fp16)[name = string("input_109_cast_fp16")];
+            tensor<int32, [1]> input_111_axes_0 = const()[name = string("input_111_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_1_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_1_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29319296)))];
+            tensor<fp16, [1024]> module_layers_1_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_1_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29321408)))];
+            tensor<fp16, [1, 188, 1024]> input_111_cast_fp16 = layer_norm(axes = input_111_axes_0, beta = module_layers_1_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_feed_forward2_weight_to_fp16, x = input_109_cast_fp16)[name = string("input_111_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2057265]> module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29323520))), indices_nonzero_data = tensor<uint6, [2057265]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29847872))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31390912))))[name = string("module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_17_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_111_cast_fp16)[name = string("linear_17_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_115_cast_fp16 = silu(x = linear_17_cast_fp16)[name = string("input_115_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1937635]> module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31456512))), indices_nonzero_data = tensor<uint6, [1937635]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31980864))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33434176))))[name = string("module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_18_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_115_cast_fp16)[name = string("linear_18_cast_fp16")];
+            fp16 var_529_to_fp16 = const()[name = string("op_529_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_530_cast_fp16 = mul(x = linear_18_cast_fp16, y = var_529_to_fp16)[name = string("op_530_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_121_cast_fp16 = add(x = input_109_cast_fp16, y = var_530_cast_fp16)[name = string("input_121_cast_fp16")];
+            tensor<int32, [1]> input_123_axes_0 = const()[name = string("input_123_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_1_norm_out_weight_to_fp16 = const()[name = string("module_layers_1_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33450624)))];
+            tensor<fp16, [1024]> module_layers_1_norm_out_bias_to_fp16 = const()[name = string("module_layers_1_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33452736)))];
+            tensor<fp16, [1, 188, 1024]> input_123_cast_fp16 = layer_norm(axes = input_123_axes_0, beta = module_layers_1_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_out_weight_to_fp16, x = input_121_cast_fp16)[name = string("input_123_cast_fp16")];
+            tensor<int32, [1]> input_125_axes_0 = const()[name = string("input_125_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_2_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_2_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33454848)))];
+            tensor<fp16, [1024]> module_layers_2_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_2_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33456960)))];
+            tensor<fp16, [1, 188, 1024]> input_125_cast_fp16 = layer_norm(axes = input_125_axes_0, beta = module_layers_2_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_feed_forward1_weight_to_fp16, x = input_123_cast_fp16)[name = string("input_125_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2079554]> module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33459072))), indices_nonzero_data = tensor<uint6, [2079554]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33983424))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35543168))))[name = string("module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_125_cast_fp16)[name = string("linear_19_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_129_cast_fp16 = silu(x = linear_19_cast_fp16)[name = string("input_129_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1998259]> module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35608768))), indices_nonzero_data = tensor<uint6, [1998259]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36133120))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37631936))))[name = string("module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_20_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_129_cast_fp16)[name = string("linear_20_cast_fp16")];
+            fp16 var_558_to_fp16 = const()[name = string("op_558_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_559_cast_fp16 = mul(x = linear_20_cast_fp16, y = var_558_to_fp16)[name = string("op_559_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_135_cast_fp16 = add(x = input_123_cast_fp16, y = var_559_cast_fp16)[name = string("input_135_cast_fp16")];
+            tensor<int32, [1]> query_5_axes_0 = const()[name = string("query_5_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_2_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_2_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37648384)))];
+            tensor<fp16, [1024]> module_layers_2_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_2_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37650496)))];
+            tensor<fp16, [1, 188, 1024]> query_5_cast_fp16 = layer_norm(axes = query_5_axes_0, beta = module_layers_2_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_self_att_weight_to_fp16, x = input_135_cast_fp16)[name = string("query_5_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522858]> module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37652608))), indices_nonzero_data = tensor<uint6, [522858]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37783744))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38176000))))[name = string("module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_21_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_5_cast_fp16)[name = string("linear_21_cast_fp16")];
+            tensor<int32, [4]> var_575 = const()[name = string("op_575"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_13_cast_fp16 = reshape(shape = var_575, x = linear_21_cast_fp16)[name = string("q_13_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521195]> module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38192448))), indices_nonzero_data = tensor<uint6, [521195]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38323584))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38714560))))[name = string("module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_22_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_5_cast_fp16)[name = string("linear_22_cast_fp16")];
+            tensor<int32, [4]> var_579 = const()[name = string("op_579"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_9_cast_fp16 = reshape(shape = var_579, x = linear_22_cast_fp16)[name = string("k_9_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521755]> module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38731008))), indices_nonzero_data = tensor<uint6, [521755]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38862144))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39253568))))[name = string("module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_23_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_5_cast_fp16)[name = string("linear_23_cast_fp16")];
+            tensor<int32, [4]> var_583 = const()[name = string("op_583"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_5_cast_fp16 = reshape(shape = var_583, x = linear_23_cast_fp16)[name = string("v_5_cast_fp16")];
+            tensor<int32, [4]> value_7_perm_0 = const()[name = string("value_7_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_2_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_2_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39270016)))];
+            tensor<fp16, [1, 188, 8, 128]> var_595_cast_fp16 = add(x = q_13_cast_fp16, y = module_layers_2_self_attn_pos_bias_u_to_fp16)[name = string("op_595_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_2_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_2_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39272128)))];
+            tensor<fp16, [1, 188, 8, 128]> var_597_cast_fp16 = add(x = q_13_cast_fp16, y = module_layers_2_self_attn_pos_bias_v_to_fp16)[name = string("op_597_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_5_perm_0 = const()[name = string("q_with_bias_v_5_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_51_transpose_x_0 = const()[name = string("x_51_transpose_x_0"), val = bool(false)];
+            bool x_51_transpose_y_0 = const()[name = string("x_51_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_599_to_fp16 = const()[name = string("op_599_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(321494080)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_5_cast_fp16 = transpose(perm = q_with_bias_v_5_perm_0, x = var_597_cast_fp16)[name = string("transpose_298")];
+            tensor<fp16, [1, 8, 188, 375]> x_51_cast_fp16 = matmul(transpose_x = x_51_transpose_x_0, transpose_y = x_51_transpose_y_0, x = q_with_bias_v_5_cast_fp16, y = var_599_to_fp16)[name = string("x_51_cast_fp16")];
+            tensor<int32, [8]> x_53_pad_0 = const()[name = string("x_53_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_53_mode_0 = const()[name = string("x_53_mode_0"), val = string("constant")];
+            fp16 const_34_to_fp16 = const()[name = string("const_34_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_53_cast_fp16 = pad(constant_val = const_34_to_fp16, mode = x_53_mode_0, pad = x_53_pad_0, x = x_51_cast_fp16)[name = string("x_53_cast_fp16")];
+            tensor<int32, [4]> var_607 = const()[name = string("op_607"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_55_cast_fp16 = reshape(shape = var_607, x = x_53_cast_fp16)[name = string("x_55_cast_fp16")];
+            tensor<int32, [4]> var_611_begin_0 = const()[name = string("op_611_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_611_end_0 = const()[name = string("op_611_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_611_end_mask_0 = const()[name = string("op_611_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_611_cast_fp16 = slice_by_index(begin = var_611_begin_0, end = var_611_end_0, end_mask = var_611_end_mask_0, x = x_55_cast_fp16)[name = string("op_611_cast_fp16")];
+            tensor<int32, [4]> var_612 = const()[name = string("op_612"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_9_cast_fp16 = reshape(shape = var_612, x = var_611_cast_fp16)[name = string("matrix_bd_9_cast_fp16")];
+            bool matrix_ac_5_transpose_x_0 = const()[name = string("matrix_ac_5_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_5_transpose_y_0 = const()[name = string("matrix_ac_5_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_100_perm_0 = const()[name = string("transpose_100_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_101_perm_0 = const()[name = string("transpose_101_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_101 = transpose(perm = transpose_101_perm_0, x = k_9_cast_fp16)[name = string("transpose_296")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_100 = transpose(perm = transpose_100_perm_0, x = var_595_cast_fp16)[name = string("transpose_297")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_5_cast_fp16 = matmul(transpose_x = matrix_ac_5_transpose_x_0, transpose_y = matrix_ac_5_transpose_y_0, x = transpose_100, y = transpose_101)[name = string("matrix_ac_5_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_11_begin_0 = const()[name = string("matrix_bd_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_11_end_0 = const()[name = string("matrix_bd_11_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_11_end_mask_0 = const()[name = string("matrix_bd_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_11_cast_fp16 = slice_by_index(begin = matrix_bd_11_begin_0, end = matrix_bd_11_end_0, end_mask = matrix_bd_11_end_mask_0, x = matrix_bd_9_cast_fp16)[name = string("matrix_bd_11_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_621_cast_fp16 = add(x = matrix_ac_5_cast_fp16, y = matrix_bd_11_cast_fp16)[name = string("op_621_cast_fp16")];
+            fp16 _inversed_scores_9_y_0_to_fp16 = const()[name = string("_inversed_scores_9_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_9_cast_fp16 = mul(x = var_621_cast_fp16, y = _inversed_scores_9_y_0_to_fp16)[name = string("_inversed_scores_9_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_11_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_9_cast_fp16, cond = mask_3)[name = string("scores_11_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_627_cast_fp16 = softmax(axis = var_30, x = scores_11_cast_fp16)[name = string("op_627_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_137_cast_fp16 = select(a = var_11_to_fp16, b = var_627_cast_fp16, cond = mask_3)[name = string("input_137_cast_fp16")];
+            bool x_57_transpose_x_0 = const()[name = string("x_57_transpose_x_0"), val = bool(false)];
+            bool x_57_transpose_y_0 = const()[name = string("x_57_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_7_cast_fp16 = transpose(perm = value_7_perm_0, x = v_5_cast_fp16)[name = string("transpose_295")];
+            tensor<fp16, [1, 8, 188, 128]> x_57_cast_fp16 = matmul(transpose_x = x_57_transpose_x_0, transpose_y = x_57_transpose_y_0, x = input_137_cast_fp16, y = value_7_cast_fp16)[name = string("x_57_cast_fp16")];
+            tensor<int32, [4]> var_631_perm_0 = const()[name = string("op_631_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_632 = const()[name = string("op_632"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_631_cast_fp16 = transpose(perm = var_631_perm_0, x = x_57_cast_fp16)[name = string("transpose_294")];
+            tensor<fp16, [1, 188, 1024]> input_139_cast_fp16 = reshape(shape = var_632, x = var_631_cast_fp16)[name = string("input_139_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525056]> module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39534400))), indices_nonzero_data = tensor<uint6, [525056]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39665536))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40059392))))[name = string("module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_25_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_139_cast_fp16)[name = string("linear_25_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_143_cast_fp16 = add(x = input_135_cast_fp16, y = linear_25_cast_fp16)[name = string("input_143_cast_fp16")];
+            tensor<int32, [1]> x_61_axes_0 = const()[name = string("x_61_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_2_norm_conv_weight_to_fp16 = const()[name = string("module_layers_2_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40075840)))];
+            tensor<fp16, [1024]> module_layers_2_norm_conv_bias_to_fp16 = const()[name = string("module_layers_2_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40077952)))];
+            tensor<fp16, [1, 188, 1024]> x_61_cast_fp16 = layer_norm(axes = x_61_axes_0, beta = module_layers_2_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_conv_weight_to_fp16, x = input_143_cast_fp16)[name = string("x_61_cast_fp16")];
+            tensor<int32, [3]> input_145_perm_0 = const()[name = string("input_145_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_147_pad_type_0 = const()[name = string("input_147_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_147_strides_0 = const()[name = string("input_147_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_147_pad_0 = const()[name = string("input_147_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_147_dilations_0 = const()[name = string("input_147_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_147_groups_0 = const()[name = string("input_147_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1042766]> module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40080064))), indices_nonzero_data = tensor<uint6, [1042766]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40342272))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41124416))))[name = string("module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_145_cast_fp16 = transpose(perm = input_145_perm_0, x = x_61_cast_fp16)[name = string("transpose_293")];
+            tensor<fp16, [1, 2048, 188]> input_147_cast_fp16 = conv(dilations = input_147_dilations_0, groups = input_147_groups_0, pad = input_147_pad_0, pad_type = input_147_pad_type_0, strides = input_147_strides_0, weight = module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_145_cast_fp16)[name = string("input_147_cast_fp16")];
+            int32 x_63_split_num_splits_0 = const()[name = string("x_63_split_num_splits_0"), val = int32(2)];
+            int32 x_63_split_axis_0 = const()[name = string("x_63_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_63_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_63_split_cast_fp16_1 = split(axis = x_63_split_axis_0, num_splits = x_63_split_num_splits_0, x = input_147_cast_fp16)[name = string("x_63_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_63_split_1_sigmoid_cast_fp16 = sigmoid(x = x_63_split_cast_fp16_1)[name = string("x_63_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_63_cast_fp16 = mul(x = x_63_split_cast_fp16_0, y = x_63_split_1_sigmoid_cast_fp16)[name = string("x_63_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_149_cast_fp16 = select(a = var_11_to_fp16, b = x_63_cast_fp16, cond = var_328)[name = string("input_149_cast_fp16")];
+            tensor<int32, [6]> input_151_pad_0 = const()[name = string("input_151_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_151_mode_0 = const()[name = string("input_151_mode_0"), val = string("constant")];
+            fp16 const_37_to_fp16 = const()[name = string("const_37_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_151_cast_fp16 = pad(constant_val = const_37_to_fp16, mode = input_151_mode_0, pad = input_151_pad_0, x = input_149_cast_fp16)[name = string("input_151_cast_fp16")];
+            string input_153_pad_type_0 = const()[name = string("input_153_pad_type_0"), val = string("valid")];
+            int32 input_153_groups_0 = const()[name = string("input_153_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_153_strides_0 = const()[name = string("input_153_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_153_pad_0 = const()[name = string("input_153_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_153_dilations_0 = const()[name = string("input_153_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_252_to_fp16 = const()[name = string("const_252_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41157248)))];
+            tensor<fp16, [1024]> const_253_to_fp16 = const()[name = string("const_253_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41175744)))];
+            tensor<fp16, [1, 1024, 188]> input_155_cast_fp16 = conv(bias = const_253_to_fp16, dilations = input_153_dilations_0, groups = input_153_groups_0, pad = input_153_pad_0, pad_type = input_153_pad_type_0, strides = input_153_strides_0, weight = const_252_to_fp16, x = input_151_cast_fp16)[name = string("input_155_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_157_cast_fp16 = silu(x = input_155_cast_fp16)[name = string("input_157_cast_fp16")];
+            string x_65_pad_type_0 = const()[name = string("x_65_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_65_strides_0 = const()[name = string("x_65_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_65_pad_0 = const()[name = string("x_65_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_65_dilations_0 = const()[name = string("x_65_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_65_groups_0 = const()[name = string("x_65_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519385]> module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41177856))), indices_nonzero_data = tensor<uint6, [519385]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41308992))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41698624))))[name = string("module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_65_cast_fp16 = conv(dilations = x_65_dilations_0, groups = x_65_groups_0, pad = x_65_pad_0, pad_type = x_65_pad_type_0, strides = x_65_strides_0, weight = module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_157_cast_fp16)[name = string("x_65_cast_fp16")];
+            tensor<int32, [3]> input_159_perm_0 = const()[name = string("input_159_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_159_cast_fp16 = transpose(perm = input_159_perm_0, x = x_65_cast_fp16)[name = string("transpose_292")];
+            tensor<fp16, [1, 188, 1024]> input_161_cast_fp16 = add(x = input_143_cast_fp16, y = input_159_cast_fp16)[name = string("input_161_cast_fp16")];
+            tensor<int32, [1]> input_163_axes_0 = const()[name = string("input_163_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_2_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_2_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41715072)))];
+            tensor<fp16, [1024]> module_layers_2_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_2_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41717184)))];
+            tensor<fp16, [1, 188, 1024]> input_163_cast_fp16 = layer_norm(axes = input_163_axes_0, beta = module_layers_2_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_feed_forward2_weight_to_fp16, x = input_161_cast_fp16)[name = string("input_163_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2065499]> module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41719296))), indices_nonzero_data = tensor<uint6, [2065499]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42243648))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43792896))))[name = string("module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_26_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_163_cast_fp16)[name = string("linear_26_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_167_cast_fp16 = silu(x = linear_26_cast_fp16)[name = string("input_167_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1965997]> module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43858496))), indices_nonzero_data = tensor<uint6, [1965997]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44382848))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45857472))))[name = string("module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_27_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_167_cast_fp16)[name = string("linear_27_cast_fp16")];
+            fp16 var_692_to_fp16 = const()[name = string("op_692_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_693_cast_fp16 = mul(x = linear_27_cast_fp16, y = var_692_to_fp16)[name = string("op_693_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_173_cast_fp16 = add(x = input_161_cast_fp16, y = var_693_cast_fp16)[name = string("input_173_cast_fp16")];
+            tensor<int32, [1]> input_175_axes_0 = const()[name = string("input_175_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_2_norm_out_weight_to_fp16 = const()[name = string("module_layers_2_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45873920)))];
+            tensor<fp16, [1024]> module_layers_2_norm_out_bias_to_fp16 = const()[name = string("module_layers_2_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45876032)))];
+            tensor<fp16, [1, 188, 1024]> input_175_cast_fp16 = layer_norm(axes = input_175_axes_0, beta = module_layers_2_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_out_weight_to_fp16, x = input_173_cast_fp16)[name = string("input_175_cast_fp16")];
+            tensor<int32, [1]> input_177_axes_0 = const()[name = string("input_177_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_3_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_3_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45878144)))];
+            tensor<fp16, [1024]> module_layers_3_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_3_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45880256)))];
+            tensor<fp16, [1, 188, 1024]> input_177_cast_fp16 = layer_norm(axes = input_177_axes_0, beta = module_layers_3_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_feed_forward1_weight_to_fp16, x = input_175_cast_fp16)[name = string("input_177_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2075730]> module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45882368))), indices_nonzero_data = tensor<uint6, [2075730]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46406720))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47963584))))[name = string("module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_28_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_177_cast_fp16)[name = string("linear_28_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_181_cast_fp16 = silu(x = linear_28_cast_fp16)[name = string("input_181_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1953322]> module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48029184))), indices_nonzero_data = tensor<uint6, [1953322]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48553536))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50018624))))[name = string("module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_29_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_181_cast_fp16)[name = string("linear_29_cast_fp16")];
+            fp16 var_721_to_fp16 = const()[name = string("op_721_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_722_cast_fp16 = mul(x = linear_29_cast_fp16, y = var_721_to_fp16)[name = string("op_722_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_187_cast_fp16 = add(x = input_175_cast_fp16, y = var_722_cast_fp16)[name = string("input_187_cast_fp16")];
+            tensor<int32, [1]> query_7_axes_0 = const()[name = string("query_7_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_3_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_3_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50035072)))];
+            tensor<fp16, [1024]> module_layers_3_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_3_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50037184)))];
+            tensor<fp16, [1, 188, 1024]> query_7_cast_fp16 = layer_norm(axes = query_7_axes_0, beta = module_layers_3_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_self_att_weight_to_fp16, x = input_187_cast_fp16)[name = string("query_7_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522877]> module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50039296))), indices_nonzero_data = tensor<uint6, [522877]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50170432))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50562688))))[name = string("module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_30_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_7_cast_fp16)[name = string("linear_30_cast_fp16")];
+            tensor<int32, [4]> var_738 = const()[name = string("op_738"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_19_cast_fp16 = reshape(shape = var_738, x = linear_30_cast_fp16)[name = string("q_19_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520403]> module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50579136))), indices_nonzero_data = tensor<uint6, [520403]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50710272))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51100672))))[name = string("module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_31_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_7_cast_fp16)[name = string("linear_31_cast_fp16")];
+            tensor<int32, [4]> var_742 = const()[name = string("op_742"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_13_cast_fp16 = reshape(shape = var_742, x = linear_31_cast_fp16)[name = string("k_13_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521127]> module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51117120))), indices_nonzero_data = tensor<uint6, [521127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51248256))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51639168))))[name = string("module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_32_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_7_cast_fp16)[name = string("linear_32_cast_fp16")];
+            tensor<int32, [4]> var_746 = const()[name = string("op_746"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_7_cast_fp16 = reshape(shape = var_746, x = linear_32_cast_fp16)[name = string("v_7_cast_fp16")];
+            tensor<int32, [4]> value_9_perm_0 = const()[name = string("value_9_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_3_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_3_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51655616)))];
+            tensor<fp16, [1, 188, 8, 128]> var_758_cast_fp16 = add(x = q_19_cast_fp16, y = module_layers_3_self_attn_pos_bias_u_to_fp16)[name = string("op_758_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_3_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_3_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51657728)))];
+            tensor<fp16, [1, 188, 8, 128]> var_760_cast_fp16 = add(x = q_19_cast_fp16, y = module_layers_3_self_attn_pos_bias_v_to_fp16)[name = string("op_760_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_7_perm_0 = const()[name = string("q_with_bias_v_7_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_73_transpose_x_0 = const()[name = string("x_73_transpose_x_0"), val = bool(false)];
+            bool x_73_transpose_y_0 = const()[name = string("x_73_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_762_to_fp16 = const()[name = string("op_762_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(322262144)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_7_cast_fp16 = transpose(perm = q_with_bias_v_7_perm_0, x = var_760_cast_fp16)[name = string("transpose_291")];
+            tensor<fp16, [1, 8, 188, 375]> x_73_cast_fp16 = matmul(transpose_x = x_73_transpose_x_0, transpose_y = x_73_transpose_y_0, x = q_with_bias_v_7_cast_fp16, y = var_762_to_fp16)[name = string("x_73_cast_fp16")];
+            tensor<int32, [8]> x_75_pad_0 = const()[name = string("x_75_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_75_mode_0 = const()[name = string("x_75_mode_0"), val = string("constant")];
+            fp16 const_44_to_fp16 = const()[name = string("const_44_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_75_cast_fp16 = pad(constant_val = const_44_to_fp16, mode = x_75_mode_0, pad = x_75_pad_0, x = x_73_cast_fp16)[name = string("x_75_cast_fp16")];
+            tensor<int32, [4]> var_770 = const()[name = string("op_770"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_77_cast_fp16 = reshape(shape = var_770, x = x_75_cast_fp16)[name = string("x_77_cast_fp16")];
+            tensor<int32, [4]> var_774_begin_0 = const()[name = string("op_774_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_774_end_0 = const()[name = string("op_774_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_774_end_mask_0 = const()[name = string("op_774_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_774_cast_fp16 = slice_by_index(begin = var_774_begin_0, end = var_774_end_0, end_mask = var_774_end_mask_0, x = x_77_cast_fp16)[name = string("op_774_cast_fp16")];
+            tensor<int32, [4]> var_775 = const()[name = string("op_775"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_13_cast_fp16 = reshape(shape = var_775, x = var_774_cast_fp16)[name = string("matrix_bd_13_cast_fp16")];
+            bool matrix_ac_7_transpose_x_0 = const()[name = string("matrix_ac_7_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_7_transpose_y_0 = const()[name = string("matrix_ac_7_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_102_perm_0 = const()[name = string("transpose_102_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_103_perm_0 = const()[name = string("transpose_103_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_103 = transpose(perm = transpose_103_perm_0, x = k_13_cast_fp16)[name = string("transpose_289")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_102 = transpose(perm = transpose_102_perm_0, x = var_758_cast_fp16)[name = string("transpose_290")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_7_cast_fp16 = matmul(transpose_x = matrix_ac_7_transpose_x_0, transpose_y = matrix_ac_7_transpose_y_0, x = transpose_102, y = transpose_103)[name = string("matrix_ac_7_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_15_begin_0 = const()[name = string("matrix_bd_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_15_end_0 = const()[name = string("matrix_bd_15_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_15_end_mask_0 = const()[name = string("matrix_bd_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_15_cast_fp16 = slice_by_index(begin = matrix_bd_15_begin_0, end = matrix_bd_15_end_0, end_mask = matrix_bd_15_end_mask_0, x = matrix_bd_13_cast_fp16)[name = string("matrix_bd_15_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_784_cast_fp16 = add(x = matrix_ac_7_cast_fp16, y = matrix_bd_15_cast_fp16)[name = string("op_784_cast_fp16")];
+            fp16 _inversed_scores_13_y_0_to_fp16 = const()[name = string("_inversed_scores_13_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_13_cast_fp16 = mul(x = var_784_cast_fp16, y = _inversed_scores_13_y_0_to_fp16)[name = string("_inversed_scores_13_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_15_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_13_cast_fp16, cond = mask_3)[name = string("scores_15_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_790_cast_fp16 = softmax(axis = var_30, x = scores_15_cast_fp16)[name = string("op_790_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_189_cast_fp16 = select(a = var_11_to_fp16, b = var_790_cast_fp16, cond = mask_3)[name = string("input_189_cast_fp16")];
+            bool x_79_transpose_x_0 = const()[name = string("x_79_transpose_x_0"), val = bool(false)];
+            bool x_79_transpose_y_0 = const()[name = string("x_79_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_9_cast_fp16 = transpose(perm = value_9_perm_0, x = v_7_cast_fp16)[name = string("transpose_288")];
+            tensor<fp16, [1, 8, 188, 128]> x_79_cast_fp16 = matmul(transpose_x = x_79_transpose_x_0, transpose_y = x_79_transpose_y_0, x = input_189_cast_fp16, y = value_9_cast_fp16)[name = string("x_79_cast_fp16")];
+            tensor<int32, [4]> var_794_perm_0 = const()[name = string("op_794_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_795 = const()[name = string("op_795"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_794_cast_fp16 = transpose(perm = var_794_perm_0, x = x_79_cast_fp16)[name = string("transpose_287")];
+            tensor<fp16, [1, 188, 1024]> input_191_cast_fp16 = reshape(shape = var_795, x = var_794_cast_fp16)[name = string("input_191_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522589]> module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51920000))), indices_nonzero_data = tensor<uint6, [522589]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52051136))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52443200))))[name = string("module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_34_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_191_cast_fp16)[name = string("linear_34_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_195_cast_fp16 = add(x = input_187_cast_fp16, y = linear_34_cast_fp16)[name = string("input_195_cast_fp16")];
+            tensor<int32, [1]> x_83_axes_0 = const()[name = string("x_83_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_3_norm_conv_weight_to_fp16 = const()[name = string("module_layers_3_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52459648)))];
+            tensor<fp16, [1024]> module_layers_3_norm_conv_bias_to_fp16 = const()[name = string("module_layers_3_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52461760)))];
+            tensor<fp16, [1, 188, 1024]> x_83_cast_fp16 = layer_norm(axes = x_83_axes_0, beta = module_layers_3_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_conv_weight_to_fp16, x = input_195_cast_fp16)[name = string("x_83_cast_fp16")];
+            tensor<int32, [3]> input_197_perm_0 = const()[name = string("input_197_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_199_pad_type_0 = const()[name = string("input_199_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_199_strides_0 = const()[name = string("input_199_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_199_pad_0 = const()[name = string("input_199_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_199_dilations_0 = const()[name = string("input_199_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_199_groups_0 = const()[name = string("input_199_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1044665]> module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52463872))), indices_nonzero_data = tensor<uint6, [1044665]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52726080))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53509696))))[name = string("module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_197_cast_fp16 = transpose(perm = input_197_perm_0, x = x_83_cast_fp16)[name = string("transpose_286")];
+            tensor<fp16, [1, 2048, 188]> input_199_cast_fp16 = conv(dilations = input_199_dilations_0, groups = input_199_groups_0, pad = input_199_pad_0, pad_type = input_199_pad_type_0, strides = input_199_strides_0, weight = module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_197_cast_fp16)[name = string("input_199_cast_fp16")];
+            int32 x_85_split_num_splits_0 = const()[name = string("x_85_split_num_splits_0"), val = int32(2)];
+            int32 x_85_split_axis_0 = const()[name = string("x_85_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_85_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_85_split_cast_fp16_1 = split(axis = x_85_split_axis_0, num_splits = x_85_split_num_splits_0, x = input_199_cast_fp16)[name = string("x_85_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_85_split_1_sigmoid_cast_fp16 = sigmoid(x = x_85_split_cast_fp16_1)[name = string("x_85_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_85_cast_fp16 = mul(x = x_85_split_cast_fp16_0, y = x_85_split_1_sigmoid_cast_fp16)[name = string("x_85_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_201_cast_fp16 = select(a = var_11_to_fp16, b = x_85_cast_fp16, cond = var_328)[name = string("input_201_cast_fp16")];
+            tensor<int32, [6]> input_203_pad_0 = const()[name = string("input_203_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_203_mode_0 = const()[name = string("input_203_mode_0"), val = string("constant")];
+            fp16 const_47_to_fp16 = const()[name = string("const_47_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_203_cast_fp16 = pad(constant_val = const_47_to_fp16, mode = input_203_mode_0, pad = input_203_pad_0, x = input_201_cast_fp16)[name = string("input_203_cast_fp16")];
+            string input_205_pad_type_0 = const()[name = string("input_205_pad_type_0"), val = string("valid")];
+            int32 input_205_groups_0 = const()[name = string("input_205_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_205_strides_0 = const()[name = string("input_205_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_205_pad_0 = const()[name = string("input_205_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_205_dilations_0 = const()[name = string("input_205_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_254_to_fp16 = const()[name = string("const_254_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53542528)))];
+            tensor<fp16, [1024]> const_255_to_fp16 = const()[name = string("const_255_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53561024)))];
+            tensor<fp16, [1, 1024, 188]> input_207_cast_fp16 = conv(bias = const_255_to_fp16, dilations = input_205_dilations_0, groups = input_205_groups_0, pad = input_205_pad_0, pad_type = input_205_pad_type_0, strides = input_205_strides_0, weight = const_254_to_fp16, x = input_203_cast_fp16)[name = string("input_207_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_209_cast_fp16 = silu(x = input_207_cast_fp16)[name = string("input_209_cast_fp16")];
+            string x_87_pad_type_0 = const()[name = string("x_87_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_87_strides_0 = const()[name = string("x_87_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_87_pad_0 = const()[name = string("x_87_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_87_dilations_0 = const()[name = string("x_87_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_87_groups_0 = const()[name = string("x_87_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519219]> module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53563136))), indices_nonzero_data = tensor<uint6, [519219]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53694272))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54083776))))[name = string("module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_87_cast_fp16 = conv(dilations = x_87_dilations_0, groups = x_87_groups_0, pad = x_87_pad_0, pad_type = x_87_pad_type_0, strides = x_87_strides_0, weight = module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_209_cast_fp16)[name = string("x_87_cast_fp16")];
+            tensor<int32, [3]> input_211_perm_0 = const()[name = string("input_211_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_211_cast_fp16 = transpose(perm = input_211_perm_0, x = x_87_cast_fp16)[name = string("transpose_285")];
+            tensor<fp16, [1, 188, 1024]> input_213_cast_fp16 = add(x = input_195_cast_fp16, y = input_211_cast_fp16)[name = string("input_213_cast_fp16")];
+            tensor<int32, [1]> input_215_axes_0 = const()[name = string("input_215_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_3_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_3_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54100224)))];
+            tensor<fp16, [1024]> module_layers_3_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_3_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54102336)))];
+            tensor<fp16, [1, 188, 1024]> input_215_cast_fp16 = layer_norm(axes = input_215_axes_0, beta = module_layers_3_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_feed_forward2_weight_to_fp16, x = input_213_cast_fp16)[name = string("input_215_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2059316]> module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54104448))), indices_nonzero_data = tensor<uint6, [2059316]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54628800))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56173376))))[name = string("module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_35_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_215_cast_fp16)[name = string("linear_35_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_219_cast_fp16 = silu(x = linear_35_cast_fp16)[name = string("input_219_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1938642]> module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56238976))), indices_nonzero_data = tensor<uint6, [1938642]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56763328))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58217408))))[name = string("module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_36_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_219_cast_fp16)[name = string("linear_36_cast_fp16")];
+            fp16 var_855_to_fp16 = const()[name = string("op_855_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_856_cast_fp16 = mul(x = linear_36_cast_fp16, y = var_855_to_fp16)[name = string("op_856_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_225_cast_fp16 = add(x = input_213_cast_fp16, y = var_856_cast_fp16)[name = string("input_225_cast_fp16")];
+            tensor<int32, [1]> input_227_axes_0 = const()[name = string("input_227_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_3_norm_out_weight_to_fp16 = const()[name = string("module_layers_3_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58233856)))];
+            tensor<fp16, [1024]> module_layers_3_norm_out_bias_to_fp16 = const()[name = string("module_layers_3_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58235968)))];
+            tensor<fp16, [1, 188, 1024]> input_227_cast_fp16 = layer_norm(axes = input_227_axes_0, beta = module_layers_3_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_out_weight_to_fp16, x = input_225_cast_fp16)[name = string("input_227_cast_fp16")];
+            tensor<int32, [1]> input_229_axes_0 = const()[name = string("input_229_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_4_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_4_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58238080)))];
+            tensor<fp16, [1024]> module_layers_4_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_4_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58240192)))];
+            tensor<fp16, [1, 188, 1024]> input_229_cast_fp16 = layer_norm(axes = input_229_axes_0, beta = module_layers_4_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_feed_forward1_weight_to_fp16, x = input_227_cast_fp16)[name = string("input_229_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2071628]> module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58242304))), indices_nonzero_data = tensor<uint6, [2071628]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58766656))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60320448))))[name = string("module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_37_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_229_cast_fp16)[name = string("linear_37_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_233_cast_fp16 = silu(x = linear_37_cast_fp16)[name = string("input_233_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1939129]> module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60386048))), indices_nonzero_data = tensor<uint6, [1939129]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60910400))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62364864))))[name = string("module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_38_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_233_cast_fp16)[name = string("linear_38_cast_fp16")];
+            fp16 var_884_to_fp16 = const()[name = string("op_884_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_885_cast_fp16 = mul(x = linear_38_cast_fp16, y = var_884_to_fp16)[name = string("op_885_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_239_cast_fp16 = add(x = input_227_cast_fp16, y = var_885_cast_fp16)[name = string("input_239_cast_fp16")];
+            tensor<int32, [1]> query_9_axes_0 = const()[name = string("query_9_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_4_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_4_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62381312)))];
+            tensor<fp16, [1024]> module_layers_4_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_4_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62383424)))];
+            tensor<fp16, [1, 188, 1024]> query_9_cast_fp16 = layer_norm(axes = query_9_axes_0, beta = module_layers_4_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_self_att_weight_to_fp16, x = input_239_cast_fp16)[name = string("query_9_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523421]> module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62385536))), indices_nonzero_data = tensor<uint6, [523421]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62516672))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62909312))))[name = string("module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_39_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_9_cast_fp16)[name = string("linear_39_cast_fp16")];
+            tensor<int32, [4]> var_901 = const()[name = string("op_901"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_25_cast_fp16 = reshape(shape = var_901, x = linear_39_cast_fp16)[name = string("q_25_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521081]> module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62925760))), indices_nonzero_data = tensor<uint6, [521081]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63056896))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63447808))))[name = string("module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_40_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_9_cast_fp16)[name = string("linear_40_cast_fp16")];
+            tensor<int32, [4]> var_905 = const()[name = string("op_905"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_17_cast_fp16 = reshape(shape = var_905, x = linear_40_cast_fp16)[name = string("k_17_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520224]> module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63464256))), indices_nonzero_data = tensor<uint6, [520224]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63595392))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63985664))))[name = string("module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_41_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_9_cast_fp16)[name = string("linear_41_cast_fp16")];
+            tensor<int32, [4]> var_909 = const()[name = string("op_909"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_9_cast_fp16 = reshape(shape = var_909, x = linear_41_cast_fp16)[name = string("v_9_cast_fp16")];
+            tensor<int32, [4]> value_11_perm_0 = const()[name = string("value_11_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_4_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_4_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64002112)))];
+            tensor<fp16, [1, 188, 8, 128]> var_921_cast_fp16 = add(x = q_25_cast_fp16, y = module_layers_4_self_attn_pos_bias_u_to_fp16)[name = string("op_921_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_4_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_4_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64004224)))];
+            tensor<fp16, [1, 188, 8, 128]> var_923_cast_fp16 = add(x = q_25_cast_fp16, y = module_layers_4_self_attn_pos_bias_v_to_fp16)[name = string("op_923_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_9_perm_0 = const()[name = string("q_with_bias_v_9_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_95_transpose_x_0 = const()[name = string("x_95_transpose_x_0"), val = bool(false)];
+            bool x_95_transpose_y_0 = const()[name = string("x_95_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_925_to_fp16 = const()[name = string("op_925_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(323030208)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_9_cast_fp16 = transpose(perm = q_with_bias_v_9_perm_0, x = var_923_cast_fp16)[name = string("transpose_284")];
+            tensor<fp16, [1, 8, 188, 375]> x_95_cast_fp16 = matmul(transpose_x = x_95_transpose_x_0, transpose_y = x_95_transpose_y_0, x = q_with_bias_v_9_cast_fp16, y = var_925_to_fp16)[name = string("x_95_cast_fp16")];
+            tensor<int32, [8]> x_97_pad_0 = const()[name = string("x_97_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_97_mode_0 = const()[name = string("x_97_mode_0"), val = string("constant")];
+            fp16 const_54_to_fp16 = const()[name = string("const_54_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_97_cast_fp16 = pad(constant_val = const_54_to_fp16, mode = x_97_mode_0, pad = x_97_pad_0, x = x_95_cast_fp16)[name = string("x_97_cast_fp16")];
+            tensor<int32, [4]> var_933 = const()[name = string("op_933"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_99_cast_fp16 = reshape(shape = var_933, x = x_97_cast_fp16)[name = string("x_99_cast_fp16")];
+            tensor<int32, [4]> var_937_begin_0 = const()[name = string("op_937_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_937_end_0 = const()[name = string("op_937_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_937_end_mask_0 = const()[name = string("op_937_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_937_cast_fp16 = slice_by_index(begin = var_937_begin_0, end = var_937_end_0, end_mask = var_937_end_mask_0, x = x_99_cast_fp16)[name = string("op_937_cast_fp16")];
+            tensor<int32, [4]> var_938 = const()[name = string("op_938"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_17_cast_fp16 = reshape(shape = var_938, x = var_937_cast_fp16)[name = string("matrix_bd_17_cast_fp16")];
+            bool matrix_ac_9_transpose_x_0 = const()[name = string("matrix_ac_9_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_9_transpose_y_0 = const()[name = string("matrix_ac_9_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_104_perm_0 = const()[name = string("transpose_104_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_105_perm_0 = const()[name = string("transpose_105_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_105 = transpose(perm = transpose_105_perm_0, x = k_17_cast_fp16)[name = string("transpose_282")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_104 = transpose(perm = transpose_104_perm_0, x = var_921_cast_fp16)[name = string("transpose_283")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_9_cast_fp16 = matmul(transpose_x = matrix_ac_9_transpose_x_0, transpose_y = matrix_ac_9_transpose_y_0, x = transpose_104, y = transpose_105)[name = string("matrix_ac_9_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_19_begin_0 = const()[name = string("matrix_bd_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_19_end_0 = const()[name = string("matrix_bd_19_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_19_end_mask_0 = const()[name = string("matrix_bd_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_19_cast_fp16 = slice_by_index(begin = matrix_bd_19_begin_0, end = matrix_bd_19_end_0, end_mask = matrix_bd_19_end_mask_0, x = matrix_bd_17_cast_fp16)[name = string("matrix_bd_19_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_947_cast_fp16 = add(x = matrix_ac_9_cast_fp16, y = matrix_bd_19_cast_fp16)[name = string("op_947_cast_fp16")];
+            fp16 _inversed_scores_17_y_0_to_fp16 = const()[name = string("_inversed_scores_17_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_17_cast_fp16 = mul(x = var_947_cast_fp16, y = _inversed_scores_17_y_0_to_fp16)[name = string("_inversed_scores_17_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_19_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_17_cast_fp16, cond = mask_3)[name = string("scores_19_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_953_cast_fp16 = softmax(axis = var_30, x = scores_19_cast_fp16)[name = string("op_953_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_241_cast_fp16 = select(a = var_11_to_fp16, b = var_953_cast_fp16, cond = mask_3)[name = string("input_241_cast_fp16")];
+            bool x_101_transpose_x_0 = const()[name = string("x_101_transpose_x_0"), val = bool(false)];
+            bool x_101_transpose_y_0 = const()[name = string("x_101_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_11_cast_fp16 = transpose(perm = value_11_perm_0, x = v_9_cast_fp16)[name = string("transpose_281")];
+            tensor<fp16, [1, 8, 188, 128]> x_101_cast_fp16 = matmul(transpose_x = x_101_transpose_x_0, transpose_y = x_101_transpose_y_0, x = input_241_cast_fp16, y = value_11_cast_fp16)[name = string("x_101_cast_fp16")];
+            tensor<int32, [4]> var_957_perm_0 = const()[name = string("op_957_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_958 = const()[name = string("op_958"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_957_cast_fp16 = transpose(perm = var_957_perm_0, x = x_101_cast_fp16)[name = string("transpose_280")];
+            tensor<fp16, [1, 188, 1024]> input_243_cast_fp16 = reshape(shape = var_958, x = var_957_cast_fp16)[name = string("input_243_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523771]> module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64266496))), indices_nonzero_data = tensor<uint6, [523771]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64397632))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64790528))))[name = string("module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_43_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_243_cast_fp16)[name = string("linear_43_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_247_cast_fp16 = add(x = input_239_cast_fp16, y = linear_43_cast_fp16)[name = string("input_247_cast_fp16")];
+            tensor<int32, [1]> x_105_axes_0 = const()[name = string("x_105_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_4_norm_conv_weight_to_fp16 = const()[name = string("module_layers_4_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64806976)))];
+            tensor<fp16, [1024]> module_layers_4_norm_conv_bias_to_fp16 = const()[name = string("module_layers_4_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64809088)))];
+            tensor<fp16, [1, 188, 1024]> x_105_cast_fp16 = layer_norm(axes = x_105_axes_0, beta = module_layers_4_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_conv_weight_to_fp16, x = input_247_cast_fp16)[name = string("x_105_cast_fp16")];
+            tensor<int32, [3]> input_249_perm_0 = const()[name = string("input_249_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_251_pad_type_0 = const()[name = string("input_251_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_251_strides_0 = const()[name = string("input_251_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_251_pad_0 = const()[name = string("input_251_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_251_dilations_0 = const()[name = string("input_251_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_251_groups_0 = const()[name = string("input_251_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1046993]> module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64811200))), indices_nonzero_data = tensor<uint6, [1046993]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65073408))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65858752))))[name = string("module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_249_cast_fp16 = transpose(perm = input_249_perm_0, x = x_105_cast_fp16)[name = string("transpose_279")];
+            tensor<fp16, [1, 2048, 188]> input_251_cast_fp16 = conv(dilations = input_251_dilations_0, groups = input_251_groups_0, pad = input_251_pad_0, pad_type = input_251_pad_type_0, strides = input_251_strides_0, weight = module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_249_cast_fp16)[name = string("input_251_cast_fp16")];
+            int32 x_107_split_num_splits_0 = const()[name = string("x_107_split_num_splits_0"), val = int32(2)];
+            int32 x_107_split_axis_0 = const()[name = string("x_107_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_107_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_107_split_cast_fp16_1 = split(axis = x_107_split_axis_0, num_splits = x_107_split_num_splits_0, x = input_251_cast_fp16)[name = string("x_107_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_107_split_1_sigmoid_cast_fp16 = sigmoid(x = x_107_split_cast_fp16_1)[name = string("x_107_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_107_cast_fp16 = mul(x = x_107_split_cast_fp16_0, y = x_107_split_1_sigmoid_cast_fp16)[name = string("x_107_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_253_cast_fp16 = select(a = var_11_to_fp16, b = x_107_cast_fp16, cond = var_328)[name = string("input_253_cast_fp16")];
+            tensor<int32, [6]> input_255_pad_0 = const()[name = string("input_255_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_255_mode_0 = const()[name = string("input_255_mode_0"), val = string("constant")];
+            fp16 const_57_to_fp16 = const()[name = string("const_57_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_255_cast_fp16 = pad(constant_val = const_57_to_fp16, mode = input_255_mode_0, pad = input_255_pad_0, x = input_253_cast_fp16)[name = string("input_255_cast_fp16")];
+            string input_257_pad_type_0 = const()[name = string("input_257_pad_type_0"), val = string("valid")];
+            int32 input_257_groups_0 = const()[name = string("input_257_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_257_strides_0 = const()[name = string("input_257_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_257_pad_0 = const()[name = string("input_257_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_257_dilations_0 = const()[name = string("input_257_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_256_to_fp16 = const()[name = string("const_256_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65891584)))];
+            tensor<fp16, [1024]> const_257_to_fp16 = const()[name = string("const_257_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65910080)))];
+            tensor<fp16, [1, 1024, 188]> input_259_cast_fp16 = conv(bias = const_257_to_fp16, dilations = input_257_dilations_0, groups = input_257_groups_0, pad = input_257_pad_0, pad_type = input_257_pad_type_0, strides = input_257_strides_0, weight = const_256_to_fp16, x = input_255_cast_fp16)[name = string("input_259_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_261_cast_fp16 = silu(x = input_259_cast_fp16)[name = string("input_261_cast_fp16")];
+            string x_109_pad_type_0 = const()[name = string("x_109_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_109_strides_0 = const()[name = string("x_109_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_109_pad_0 = const()[name = string("x_109_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_109_dilations_0 = const()[name = string("x_109_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_109_groups_0 = const()[name = string("x_109_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520275]> module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65912192))), indices_nonzero_data = tensor<uint6, [520275]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66043328))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66433600))))[name = string("module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_109_cast_fp16 = conv(dilations = x_109_dilations_0, groups = x_109_groups_0, pad = x_109_pad_0, pad_type = x_109_pad_type_0, strides = x_109_strides_0, weight = module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_261_cast_fp16)[name = string("x_109_cast_fp16")];
+            tensor<int32, [3]> input_263_perm_0 = const()[name = string("input_263_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_263_cast_fp16 = transpose(perm = input_263_perm_0, x = x_109_cast_fp16)[name = string("transpose_278")];
+            tensor<fp16, [1, 188, 1024]> input_265_cast_fp16 = add(x = input_247_cast_fp16, y = input_263_cast_fp16)[name = string("input_265_cast_fp16")];
+            tensor<int32, [1]> input_267_axes_0 = const()[name = string("input_267_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_4_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_4_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66450048)))];
+            tensor<fp16, [1024]> module_layers_4_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_4_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66452160)))];
+            tensor<fp16, [1, 188, 1024]> input_267_cast_fp16 = layer_norm(axes = input_267_axes_0, beta = module_layers_4_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_feed_forward2_weight_to_fp16, x = input_265_cast_fp16)[name = string("input_267_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2060977]> module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66454272))), indices_nonzero_data = tensor<uint6, [2060977]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66978624))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68524480))))[name = string("module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_44_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_267_cast_fp16)[name = string("linear_44_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_271_cast_fp16 = silu(x = linear_44_cast_fp16)[name = string("input_271_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1931171]> module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68590080))), indices_nonzero_data = tensor<uint6, [1931171]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69114432))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70562880))))[name = string("module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_45_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_271_cast_fp16)[name = string("linear_45_cast_fp16")];
+            fp16 var_1018_to_fp16 = const()[name = string("op_1018_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_1019_cast_fp16 = mul(x = linear_45_cast_fp16, y = var_1018_to_fp16)[name = string("op_1019_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_277_cast_fp16 = add(x = input_265_cast_fp16, y = var_1019_cast_fp16)[name = string("input_277_cast_fp16")];
+            tensor<int32, [1]> input_279_axes_0 = const()[name = string("input_279_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_4_norm_out_weight_to_fp16 = const()[name = string("module_layers_4_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70579328)))];
+            tensor<fp16, [1024]> module_layers_4_norm_out_bias_to_fp16 = const()[name = string("module_layers_4_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70581440)))];
+            tensor<fp16, [1, 188, 1024]> input_279_cast_fp16 = layer_norm(axes = input_279_axes_0, beta = module_layers_4_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_out_weight_to_fp16, x = input_277_cast_fp16)[name = string("input_279_cast_fp16")];
+            tensor<int32, [1]> input_281_axes_0 = const()[name = string("input_281_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_5_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_5_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70583552)))];
+            tensor<fp16, [1024]> module_layers_5_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_5_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70585664)))];
+            tensor<fp16, [1, 188, 1024]> input_281_cast_fp16 = layer_norm(axes = input_281_axes_0, beta = module_layers_5_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_feed_forward1_weight_to_fp16, x = input_279_cast_fp16)[name = string("input_281_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2075089]> module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70587776))), indices_nonzero_data = tensor<uint6, [2075089]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71112128))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72668544))))[name = string("module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_46_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_281_cast_fp16)[name = string("linear_46_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_285_cast_fp16 = silu(x = linear_46_cast_fp16)[name = string("input_285_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1956387]> module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72734144))), indices_nonzero_data = tensor<uint6, [1956387]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73258496))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74725888))))[name = string("module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_47_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_285_cast_fp16)[name = string("linear_47_cast_fp16")];
+            fp16 var_1047_to_fp16 = const()[name = string("op_1047_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_1048_cast_fp16 = mul(x = linear_47_cast_fp16, y = var_1047_to_fp16)[name = string("op_1048_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_291_cast_fp16 = add(x = input_279_cast_fp16, y = var_1048_cast_fp16)[name = string("input_291_cast_fp16")];
+            tensor<int32, [1]> query_11_axes_0 = const()[name = string("query_11_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_5_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_5_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74742336)))];
+            tensor<fp16, [1024]> module_layers_5_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_5_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74744448)))];
+            tensor<fp16, [1, 188, 1024]> query_11_cast_fp16 = layer_norm(axes = query_11_axes_0, beta = module_layers_5_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_self_att_weight_to_fp16, x = input_291_cast_fp16)[name = string("query_11_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521972]> module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74746560))), indices_nonzero_data = tensor<uint6, [521972]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74877696))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75269248))))[name = string("module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_48_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_11_cast_fp16)[name = string("linear_48_cast_fp16")];
+            tensor<int32, [4]> var_1064 = const()[name = string("op_1064"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_31_cast_fp16 = reshape(shape = var_1064, x = linear_48_cast_fp16)[name = string("q_31_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522028]> module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75285696))), indices_nonzero_data = tensor<uint6, [522028]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75416832))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75808448))))[name = string("module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_49_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_11_cast_fp16)[name = string("linear_49_cast_fp16")];
+            tensor<int32, [4]> var_1068 = const()[name = string("op_1068"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_21_cast_fp16 = reshape(shape = var_1068, x = linear_49_cast_fp16)[name = string("k_21_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519496]> module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75824896))), indices_nonzero_data = tensor<uint6, [519496]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75956032))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76345728))))[name = string("module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_50_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_11_cast_fp16)[name = string("linear_50_cast_fp16")];
+            tensor<int32, [4]> var_1072 = const()[name = string("op_1072"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_11_cast_fp16 = reshape(shape = var_1072, x = linear_50_cast_fp16)[name = string("v_11_cast_fp16")];
+            tensor<int32, [4]> value_13_perm_0 = const()[name = string("value_13_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_5_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_5_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76362176)))];
+            tensor<fp16, [1, 188, 8, 128]> var_1084_cast_fp16 = add(x = q_31_cast_fp16, y = module_layers_5_self_attn_pos_bias_u_to_fp16)[name = string("op_1084_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_5_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_5_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76364288)))];
+            tensor<fp16, [1, 188, 8, 128]> var_1086_cast_fp16 = add(x = q_31_cast_fp16, y = module_layers_5_self_attn_pos_bias_v_to_fp16)[name = string("op_1086_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_11_perm_0 = const()[name = string("q_with_bias_v_11_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_117_transpose_x_0 = const()[name = string("x_117_transpose_x_0"), val = bool(false)];
+            bool x_117_transpose_y_0 = const()[name = string("x_117_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_1088_to_fp16 = const()[name = string("op_1088_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(323798272)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_11_cast_fp16 = transpose(perm = q_with_bias_v_11_perm_0, x = var_1086_cast_fp16)[name = string("transpose_277")];
+            tensor<fp16, [1, 8, 188, 375]> x_117_cast_fp16 = matmul(transpose_x = x_117_transpose_x_0, transpose_y = x_117_transpose_y_0, x = q_with_bias_v_11_cast_fp16, y = var_1088_to_fp16)[name = string("x_117_cast_fp16")];
+            tensor<int32, [8]> x_119_pad_0 = const()[name = string("x_119_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_119_mode_0 = const()[name = string("x_119_mode_0"), val = string("constant")];
+            fp16 const_64_to_fp16 = const()[name = string("const_64_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_119_cast_fp16 = pad(constant_val = const_64_to_fp16, mode = x_119_mode_0, pad = x_119_pad_0, x = x_117_cast_fp16)[name = string("x_119_cast_fp16")];
+            tensor<int32, [4]> var_1096 = const()[name = string("op_1096"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_121_cast_fp16 = reshape(shape = var_1096, x = x_119_cast_fp16)[name = string("x_121_cast_fp16")];
+            tensor<int32, [4]> var_1100_begin_0 = const()[name = string("op_1100_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1100_end_0 = const()[name = string("op_1100_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_1100_end_mask_0 = const()[name = string("op_1100_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_1100_cast_fp16 = slice_by_index(begin = var_1100_begin_0, end = var_1100_end_0, end_mask = var_1100_end_mask_0, x = x_121_cast_fp16)[name = string("op_1100_cast_fp16")];
+            tensor<int32, [4]> var_1101 = const()[name = string("op_1101"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_21_cast_fp16 = reshape(shape = var_1101, x = var_1100_cast_fp16)[name = string("matrix_bd_21_cast_fp16")];
+            bool matrix_ac_11_transpose_x_0 = const()[name = string("matrix_ac_11_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_11_transpose_y_0 = const()[name = string("matrix_ac_11_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_106_perm_0 = const()[name = string("transpose_106_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_107_perm_0 = const()[name = string("transpose_107_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_107 = transpose(perm = transpose_107_perm_0, x = k_21_cast_fp16)[name = string("transpose_275")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_106 = transpose(perm = transpose_106_perm_0, x = var_1084_cast_fp16)[name = string("transpose_276")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_11_cast_fp16 = matmul(transpose_x = matrix_ac_11_transpose_x_0, transpose_y = matrix_ac_11_transpose_y_0, x = transpose_106, y = transpose_107)[name = string("matrix_ac_11_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_23_begin_0 = const()[name = string("matrix_bd_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_23_end_0 = const()[name = string("matrix_bd_23_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_23_end_mask_0 = const()[name = string("matrix_bd_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_23_cast_fp16 = slice_by_index(begin = matrix_bd_23_begin_0, end = matrix_bd_23_end_0, end_mask = matrix_bd_23_end_mask_0, x = matrix_bd_21_cast_fp16)[name = string("matrix_bd_23_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_1110_cast_fp16 = add(x = matrix_ac_11_cast_fp16, y = matrix_bd_23_cast_fp16)[name = string("op_1110_cast_fp16")];
+            fp16 _inversed_scores_21_y_0_to_fp16 = const()[name = string("_inversed_scores_21_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_21_cast_fp16 = mul(x = var_1110_cast_fp16, y = _inversed_scores_21_y_0_to_fp16)[name = string("_inversed_scores_21_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_23_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_21_cast_fp16, cond = mask_3)[name = string("scores_23_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_1116_cast_fp16 = softmax(axis = var_30, x = scores_23_cast_fp16)[name = string("op_1116_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_293_cast_fp16 = select(a = var_11_to_fp16, b = var_1116_cast_fp16, cond = mask_3)[name = string("input_293_cast_fp16")];
+            bool x_123_transpose_x_0 = const()[name = string("x_123_transpose_x_0"), val = bool(false)];
+            bool x_123_transpose_y_0 = const()[name = string("x_123_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_13_cast_fp16 = transpose(perm = value_13_perm_0, x = v_11_cast_fp16)[name = string("transpose_274")];
+            tensor<fp16, [1, 8, 188, 128]> x_123_cast_fp16 = matmul(transpose_x = x_123_transpose_x_0, transpose_y = x_123_transpose_y_0, x = input_293_cast_fp16, y = value_13_cast_fp16)[name = string("x_123_cast_fp16")];
+            tensor<int32, [4]> var_1120_perm_0 = const()[name = string("op_1120_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1121 = const()[name = string("op_1121"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_1120_cast_fp16 = transpose(perm = var_1120_perm_0, x = x_123_cast_fp16)[name = string("transpose_273")];
+            tensor<fp16, [1, 188, 1024]> input_295_cast_fp16 = reshape(shape = var_1121, x = var_1120_cast_fp16)[name = string("input_295_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523851]> module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76626560))), indices_nonzero_data = tensor<uint6, [523851]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76757696))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77150656))))[name = string("module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_52_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_295_cast_fp16)[name = string("linear_52_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_299_cast_fp16 = add(x = input_291_cast_fp16, y = linear_52_cast_fp16)[name = string("input_299_cast_fp16")];
+            tensor<int32, [1]> x_127_axes_0 = const()[name = string("x_127_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_5_norm_conv_weight_to_fp16 = const()[name = string("module_layers_5_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77167104)))];
+            tensor<fp16, [1024]> module_layers_5_norm_conv_bias_to_fp16 = const()[name = string("module_layers_5_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77169216)))];
+            tensor<fp16, [1, 188, 1024]> x_127_cast_fp16 = layer_norm(axes = x_127_axes_0, beta = module_layers_5_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_conv_weight_to_fp16, x = input_299_cast_fp16)[name = string("x_127_cast_fp16")];
+            tensor<int32, [3]> input_301_perm_0 = const()[name = string("input_301_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_303_pad_type_0 = const()[name = string("input_303_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_303_strides_0 = const()[name = string("input_303_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_303_pad_0 = const()[name = string("input_303_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_303_dilations_0 = const()[name = string("input_303_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_303_groups_0 = const()[name = string("input_303_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047810]> module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77171328))), indices_nonzero_data = tensor<uint6, [1047810]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77433536))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78219520))))[name = string("module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_301_cast_fp16 = transpose(perm = input_301_perm_0, x = x_127_cast_fp16)[name = string("transpose_272")];
+            tensor<fp16, [1, 2048, 188]> input_303_cast_fp16 = conv(dilations = input_303_dilations_0, groups = input_303_groups_0, pad = input_303_pad_0, pad_type = input_303_pad_type_0, strides = input_303_strides_0, weight = module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_301_cast_fp16)[name = string("input_303_cast_fp16")];
+            int32 x_129_split_num_splits_0 = const()[name = string("x_129_split_num_splits_0"), val = int32(2)];
+            int32 x_129_split_axis_0 = const()[name = string("x_129_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_129_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_129_split_cast_fp16_1 = split(axis = x_129_split_axis_0, num_splits = x_129_split_num_splits_0, x = input_303_cast_fp16)[name = string("x_129_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_129_split_1_sigmoid_cast_fp16 = sigmoid(x = x_129_split_cast_fp16_1)[name = string("x_129_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_129_cast_fp16 = mul(x = x_129_split_cast_fp16_0, y = x_129_split_1_sigmoid_cast_fp16)[name = string("x_129_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_305_cast_fp16 = select(a = var_11_to_fp16, b = x_129_cast_fp16, cond = var_328)[name = string("input_305_cast_fp16")];
+            tensor<int32, [6]> input_307_pad_0 = const()[name = string("input_307_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_307_mode_0 = const()[name = string("input_307_mode_0"), val = string("constant")];
+            fp16 const_67_to_fp16 = const()[name = string("const_67_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_307_cast_fp16 = pad(constant_val = const_67_to_fp16, mode = input_307_mode_0, pad = input_307_pad_0, x = input_305_cast_fp16)[name = string("input_307_cast_fp16")];
+            string input_309_pad_type_0 = const()[name = string("input_309_pad_type_0"), val = string("valid")];
+            int32 input_309_groups_0 = const()[name = string("input_309_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_309_strides_0 = const()[name = string("input_309_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_309_pad_0 = const()[name = string("input_309_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_309_dilations_0 = const()[name = string("input_309_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_258_to_fp16 = const()[name = string("const_258_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78252352)))];
+            tensor<fp16, [1024]> const_259_to_fp16 = const()[name = string("const_259_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78270848)))];
+            tensor<fp16, [1, 1024, 188]> input_311_cast_fp16 = conv(bias = const_259_to_fp16, dilations = input_309_dilations_0, groups = input_309_groups_0, pad = input_309_pad_0, pad_type = input_309_pad_type_0, strides = input_309_strides_0, weight = const_258_to_fp16, x = input_307_cast_fp16)[name = string("input_311_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_313_cast_fp16 = silu(x = input_311_cast_fp16)[name = string("input_313_cast_fp16")];
+            string x_131_pad_type_0 = const()[name = string("x_131_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_131_strides_0 = const()[name = string("x_131_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_131_pad_0 = const()[name = string("x_131_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_131_dilations_0 = const()[name = string("x_131_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_131_groups_0 = const()[name = string("x_131_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520509]> module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78272960))), indices_nonzero_data = tensor<uint6, [520509]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78404096))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78794560))))[name = string("module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_131_cast_fp16 = conv(dilations = x_131_dilations_0, groups = x_131_groups_0, pad = x_131_pad_0, pad_type = x_131_pad_type_0, strides = x_131_strides_0, weight = module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_313_cast_fp16)[name = string("x_131_cast_fp16")];
+            tensor<int32, [3]> input_315_perm_0 = const()[name = string("input_315_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_315_cast_fp16 = transpose(perm = input_315_perm_0, x = x_131_cast_fp16)[name = string("transpose_271")];
+            tensor<fp16, [1, 188, 1024]> input_317_cast_fp16 = add(x = input_299_cast_fp16, y = input_315_cast_fp16)[name = string("input_317_cast_fp16")];
+            tensor<int32, [1]> input_319_axes_0 = const()[name = string("input_319_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_5_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_5_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78811008)))];
+            tensor<fp16, [1024]> module_layers_5_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_5_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78813120)))];
+            tensor<fp16, [1, 188, 1024]> input_319_cast_fp16 = layer_norm(axes = input_319_axes_0, beta = module_layers_5_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_feed_forward2_weight_to_fp16, x = input_317_cast_fp16)[name = string("input_319_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2063618]> module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78815232))), indices_nonzero_data = tensor<uint6, [2063618]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79339584))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80887424))))[name = string("module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_53_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_319_cast_fp16)[name = string("linear_53_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_323_cast_fp16 = silu(x = linear_53_cast_fp16)[name = string("input_323_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1985406]> module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80953024))), indices_nonzero_data = tensor<uint6, [1985406]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81477376))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82966528))))[name = string("module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_54_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_323_cast_fp16)[name = string("linear_54_cast_fp16")];
+            fp16 var_1181_to_fp16 = const()[name = string("op_1181_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_1182_cast_fp16 = mul(x = linear_54_cast_fp16, y = var_1181_to_fp16)[name = string("op_1182_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_329_cast_fp16 = add(x = input_317_cast_fp16, y = var_1182_cast_fp16)[name = string("input_329_cast_fp16")];
+            tensor<int32, [1]> input_331_axes_0 = const()[name = string("input_331_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_5_norm_out_weight_to_fp16 = const()[name = string("module_layers_5_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82982976)))];
+            tensor<fp16, [1024]> module_layers_5_norm_out_bias_to_fp16 = const()[name = string("module_layers_5_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82985088)))];
+            tensor<fp16, [1, 188, 1024]> input_331_cast_fp16 = layer_norm(axes = input_331_axes_0, beta = module_layers_5_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_out_weight_to_fp16, x = input_329_cast_fp16)[name = string("input_331_cast_fp16")];
+            tensor<int32, [1]> input_333_axes_0 = const()[name = string("input_333_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_6_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_6_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82987200)))];
+            tensor<fp16, [1024]> module_layers_6_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_6_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82989312)))];
+            tensor<fp16, [1, 188, 1024]> input_333_cast_fp16 = layer_norm(axes = input_333_axes_0, beta = module_layers_6_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_feed_forward1_weight_to_fp16, x = input_331_cast_fp16)[name = string("input_333_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2073214]> module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82991424))), indices_nonzero_data = tensor<uint6, [2073214]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83515776))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85070784))))[name = string("module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_333_cast_fp16)[name = string("linear_55_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_337_cast_fp16 = silu(x = linear_55_cast_fp16)[name = string("input_337_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1956749]> module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85136384))), indices_nonzero_data = tensor<uint6, [1956749]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85660736))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87128384))))[name = string("module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_56_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_337_cast_fp16)[name = string("linear_56_cast_fp16")];
+            fp16 var_1210_to_fp16 = const()[name = string("op_1210_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_1211_cast_fp16 = mul(x = linear_56_cast_fp16, y = var_1210_to_fp16)[name = string("op_1211_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_343_cast_fp16 = add(x = input_331_cast_fp16, y = var_1211_cast_fp16)[name = string("input_343_cast_fp16")];
+            tensor<int32, [1]> query_13_axes_0 = const()[name = string("query_13_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_6_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_6_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87144832)))];
+            tensor<fp16, [1024]> module_layers_6_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_6_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87146944)))];
+            tensor<fp16, [1, 188, 1024]> query_13_cast_fp16 = layer_norm(axes = query_13_axes_0, beta = module_layers_6_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_self_att_weight_to_fp16, x = input_343_cast_fp16)[name = string("query_13_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522973]> module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87149056))), indices_nonzero_data = tensor<uint6, [522973]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87280192))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87672512))))[name = string("module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_57_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_13_cast_fp16)[name = string("linear_57_cast_fp16")];
+            tensor<int32, [4]> var_1227 = const()[name = string("op_1227"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_37_cast_fp16 = reshape(shape = var_1227, x = linear_57_cast_fp16)[name = string("q_37_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521918]> module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87688960))), indices_nonzero_data = tensor<uint6, [521918]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87820096))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88211648))))[name = string("module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_58_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_13_cast_fp16)[name = string("linear_58_cast_fp16")];
+            tensor<int32, [4]> var_1231 = const()[name = string("op_1231"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_25_cast_fp16 = reshape(shape = var_1231, x = linear_58_cast_fp16)[name = string("k_25_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520459]> module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88228096))), indices_nonzero_data = tensor<uint6, [520459]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88359232))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88749696))))[name = string("module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_59_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_13_cast_fp16)[name = string("linear_59_cast_fp16")];
+            tensor<int32, [4]> var_1235 = const()[name = string("op_1235"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_13_cast_fp16 = reshape(shape = var_1235, x = linear_59_cast_fp16)[name = string("v_13_cast_fp16")];
+            tensor<int32, [4]> value_15_perm_0 = const()[name = string("value_15_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_6_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_6_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88766144)))];
+            tensor<fp16, [1, 188, 8, 128]> var_1247_cast_fp16 = add(x = q_37_cast_fp16, y = module_layers_6_self_attn_pos_bias_u_to_fp16)[name = string("op_1247_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_6_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_6_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88768256)))];
+            tensor<fp16, [1, 188, 8, 128]> var_1249_cast_fp16 = add(x = q_37_cast_fp16, y = module_layers_6_self_attn_pos_bias_v_to_fp16)[name = string("op_1249_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_13_perm_0 = const()[name = string("q_with_bias_v_13_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_139_transpose_x_0 = const()[name = string("x_139_transpose_x_0"), val = bool(false)];
+            bool x_139_transpose_y_0 = const()[name = string("x_139_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_1251_to_fp16 = const()[name = string("op_1251_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(324566336)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_13_cast_fp16 = transpose(perm = q_with_bias_v_13_perm_0, x = var_1249_cast_fp16)[name = string("transpose_270")];
+            tensor<fp16, [1, 8, 188, 375]> x_139_cast_fp16 = matmul(transpose_x = x_139_transpose_x_0, transpose_y = x_139_transpose_y_0, x = q_with_bias_v_13_cast_fp16, y = var_1251_to_fp16)[name = string("x_139_cast_fp16")];
+            tensor<int32, [8]> x_141_pad_0 = const()[name = string("x_141_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_141_mode_0 = const()[name = string("x_141_mode_0"), val = string("constant")];
+            fp16 const_74_to_fp16 = const()[name = string("const_74_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_141_cast_fp16 = pad(constant_val = const_74_to_fp16, mode = x_141_mode_0, pad = x_141_pad_0, x = x_139_cast_fp16)[name = string("x_141_cast_fp16")];
+            tensor<int32, [4]> var_1259 = const()[name = string("op_1259"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_143_cast_fp16 = reshape(shape = var_1259, x = x_141_cast_fp16)[name = string("x_143_cast_fp16")];
+            tensor<int32, [4]> var_1263_begin_0 = const()[name = string("op_1263_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1263_end_0 = const()[name = string("op_1263_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_1263_end_mask_0 = const()[name = string("op_1263_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_1263_cast_fp16 = slice_by_index(begin = var_1263_begin_0, end = var_1263_end_0, end_mask = var_1263_end_mask_0, x = x_143_cast_fp16)[name = string("op_1263_cast_fp16")];
+            tensor<int32, [4]> var_1264 = const()[name = string("op_1264"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_25_cast_fp16 = reshape(shape = var_1264, x = var_1263_cast_fp16)[name = string("matrix_bd_25_cast_fp16")];
+            bool matrix_ac_13_transpose_x_0 = const()[name = string("matrix_ac_13_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_13_transpose_y_0 = const()[name = string("matrix_ac_13_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_108_perm_0 = const()[name = string("transpose_108_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_109_perm_0 = const()[name = string("transpose_109_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_109 = transpose(perm = transpose_109_perm_0, x = k_25_cast_fp16)[name = string("transpose_268")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_108 = transpose(perm = transpose_108_perm_0, x = var_1247_cast_fp16)[name = string("transpose_269")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_13_cast_fp16 = matmul(transpose_x = matrix_ac_13_transpose_x_0, transpose_y = matrix_ac_13_transpose_y_0, x = transpose_108, y = transpose_109)[name = string("matrix_ac_13_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_27_begin_0 = const()[name = string("matrix_bd_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_27_end_0 = const()[name = string("matrix_bd_27_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_27_end_mask_0 = const()[name = string("matrix_bd_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_27_cast_fp16 = slice_by_index(begin = matrix_bd_27_begin_0, end = matrix_bd_27_end_0, end_mask = matrix_bd_27_end_mask_0, x = matrix_bd_25_cast_fp16)[name = string("matrix_bd_27_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_1273_cast_fp16 = add(x = matrix_ac_13_cast_fp16, y = matrix_bd_27_cast_fp16)[name = string("op_1273_cast_fp16")];
+            fp16 _inversed_scores_25_y_0_to_fp16 = const()[name = string("_inversed_scores_25_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_25_cast_fp16 = mul(x = var_1273_cast_fp16, y = _inversed_scores_25_y_0_to_fp16)[name = string("_inversed_scores_25_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_27_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_25_cast_fp16, cond = mask_3)[name = string("scores_27_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_1279_cast_fp16 = softmax(axis = var_30, x = scores_27_cast_fp16)[name = string("op_1279_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_345_cast_fp16 = select(a = var_11_to_fp16, b = var_1279_cast_fp16, cond = mask_3)[name = string("input_345_cast_fp16")];
+            bool x_145_transpose_x_0 = const()[name = string("x_145_transpose_x_0"), val = bool(false)];
+            bool x_145_transpose_y_0 = const()[name = string("x_145_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_15_cast_fp16 = transpose(perm = value_15_perm_0, x = v_13_cast_fp16)[name = string("transpose_267")];
+            tensor<fp16, [1, 8, 188, 128]> x_145_cast_fp16 = matmul(transpose_x = x_145_transpose_x_0, transpose_y = x_145_transpose_y_0, x = input_345_cast_fp16, y = value_15_cast_fp16)[name = string("x_145_cast_fp16")];
+            tensor<int32, [4]> var_1283_perm_0 = const()[name = string("op_1283_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1284 = const()[name = string("op_1284"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_1283_cast_fp16 = transpose(perm = var_1283_perm_0, x = x_145_cast_fp16)[name = string("transpose_266")];
+            tensor<fp16, [1, 188, 1024]> input_347_cast_fp16 = reshape(shape = var_1284, x = var_1283_cast_fp16)[name = string("input_347_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524874]> module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89030528))), indices_nonzero_data = tensor<uint6, [524874]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89161664))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89555392))))[name = string("module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_61_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_347_cast_fp16)[name = string("linear_61_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_351_cast_fp16 = add(x = input_343_cast_fp16, y = linear_61_cast_fp16)[name = string("input_351_cast_fp16")];
+            tensor<int32, [1]> x_149_axes_0 = const()[name = string("x_149_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_6_norm_conv_weight_to_fp16 = const()[name = string("module_layers_6_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89571840)))];
+            tensor<fp16, [1024]> module_layers_6_norm_conv_bias_to_fp16 = const()[name = string("module_layers_6_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89573952)))];
+            tensor<fp16, [1, 188, 1024]> x_149_cast_fp16 = layer_norm(axes = x_149_axes_0, beta = module_layers_6_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_conv_weight_to_fp16, x = input_351_cast_fp16)[name = string("x_149_cast_fp16")];
+            tensor<int32, [3]> input_353_perm_0 = const()[name = string("input_353_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_355_pad_type_0 = const()[name = string("input_355_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_355_strides_0 = const()[name = string("input_355_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_355_pad_0 = const()[name = string("input_355_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_355_dilations_0 = const()[name = string("input_355_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_355_groups_0 = const()[name = string("input_355_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1048536]> module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89576064))), indices_nonzero_data = tensor<uint6, [1048536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89838272))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90624768))))[name = string("module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_353_cast_fp16 = transpose(perm = input_353_perm_0, x = x_149_cast_fp16)[name = string("transpose_265")];
+            tensor<fp16, [1, 2048, 188]> input_355_cast_fp16 = conv(dilations = input_355_dilations_0, groups = input_355_groups_0, pad = input_355_pad_0, pad_type = input_355_pad_type_0, strides = input_355_strides_0, weight = module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_353_cast_fp16)[name = string("input_355_cast_fp16")];
+            int32 x_151_split_num_splits_0 = const()[name = string("x_151_split_num_splits_0"), val = int32(2)];
+            int32 x_151_split_axis_0 = const()[name = string("x_151_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_151_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_151_split_cast_fp16_1 = split(axis = x_151_split_axis_0, num_splits = x_151_split_num_splits_0, x = input_355_cast_fp16)[name = string("x_151_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_151_split_1_sigmoid_cast_fp16 = sigmoid(x = x_151_split_cast_fp16_1)[name = string("x_151_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_151_cast_fp16 = mul(x = x_151_split_cast_fp16_0, y = x_151_split_1_sigmoid_cast_fp16)[name = string("x_151_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_357_cast_fp16 = select(a = var_11_to_fp16, b = x_151_cast_fp16, cond = var_328)[name = string("input_357_cast_fp16")];
+            tensor<int32, [6]> input_359_pad_0 = const()[name = string("input_359_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_359_mode_0 = const()[name = string("input_359_mode_0"), val = string("constant")];
+            fp16 const_77_to_fp16 = const()[name = string("const_77_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_359_cast_fp16 = pad(constant_val = const_77_to_fp16, mode = input_359_mode_0, pad = input_359_pad_0, x = input_357_cast_fp16)[name = string("input_359_cast_fp16")];
+            string input_361_pad_type_0 = const()[name = string("input_361_pad_type_0"), val = string("valid")];
+            int32 input_361_groups_0 = const()[name = string("input_361_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_361_strides_0 = const()[name = string("input_361_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_361_pad_0 = const()[name = string("input_361_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_361_dilations_0 = const()[name = string("input_361_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_260_to_fp16 = const()[name = string("const_260_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90657600)))];
+            tensor<fp16, [1024]> const_261_to_fp16 = const()[name = string("const_261_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90676096)))];
+            tensor<fp16, [1, 1024, 188]> input_363_cast_fp16 = conv(bias = const_261_to_fp16, dilations = input_361_dilations_0, groups = input_361_groups_0, pad = input_361_pad_0, pad_type = input_361_pad_type_0, strides = input_361_strides_0, weight = const_260_to_fp16, x = input_359_cast_fp16)[name = string("input_363_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_365_cast_fp16 = silu(x = input_363_cast_fp16)[name = string("input_365_cast_fp16")];
+            string x_153_pad_type_0 = const()[name = string("x_153_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_153_strides_0 = const()[name = string("x_153_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_153_pad_0 = const()[name = string("x_153_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_153_dilations_0 = const()[name = string("x_153_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_153_groups_0 = const()[name = string("x_153_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519645]> module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90678208))), indices_nonzero_data = tensor<uint6, [519645]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90809344))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91199168))))[name = string("module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_153_cast_fp16 = conv(dilations = x_153_dilations_0, groups = x_153_groups_0, pad = x_153_pad_0, pad_type = x_153_pad_type_0, strides = x_153_strides_0, weight = module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_365_cast_fp16)[name = string("x_153_cast_fp16")];
+            tensor<int32, [3]> input_367_perm_0 = const()[name = string("input_367_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_367_cast_fp16 = transpose(perm = input_367_perm_0, x = x_153_cast_fp16)[name = string("transpose_264")];
+            tensor<fp16, [1, 188, 1024]> input_369_cast_fp16 = add(x = input_351_cast_fp16, y = input_367_cast_fp16)[name = string("input_369_cast_fp16")];
+            tensor<int32, [1]> input_371_axes_0 = const()[name = string("input_371_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_6_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_6_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91215616)))];
+            tensor<fp16, [1024]> module_layers_6_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_6_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91217728)))];
+            tensor<fp16, [1, 188, 1024]> input_371_cast_fp16 = layer_norm(axes = input_371_axes_0, beta = module_layers_6_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_feed_forward2_weight_to_fp16, x = input_369_cast_fp16)[name = string("input_371_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2063870]> module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91219840))), indices_nonzero_data = tensor<uint6, [2063870]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91744192))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93292160))))[name = string("module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_62_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_371_cast_fp16)[name = string("linear_62_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_375_cast_fp16 = silu(x = linear_62_cast_fp16)[name = string("input_375_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2019355]> module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93357760))), indices_nonzero_data = tensor<uint6, [2019355]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93882112))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95396736))))[name = string("module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_63_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_375_cast_fp16)[name = string("linear_63_cast_fp16")];
+            fp16 var_1344_to_fp16 = const()[name = string("op_1344_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_1345_cast_fp16 = mul(x = linear_63_cast_fp16, y = var_1344_to_fp16)[name = string("op_1345_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_381_cast_fp16 = add(x = input_369_cast_fp16, y = var_1345_cast_fp16)[name = string("input_381_cast_fp16")];
+            tensor<int32, [1]> input_383_axes_0 = const()[name = string("input_383_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_6_norm_out_weight_to_fp16 = const()[name = string("module_layers_6_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95413184)))];
+            tensor<fp16, [1024]> module_layers_6_norm_out_bias_to_fp16 = const()[name = string("module_layers_6_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95415296)))];
+            tensor<fp16, [1, 188, 1024]> input_383_cast_fp16 = layer_norm(axes = input_383_axes_0, beta = module_layers_6_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_out_weight_to_fp16, x = input_381_cast_fp16)[name = string("input_383_cast_fp16")];
+            tensor<int32, [1]> input_385_axes_0 = const()[name = string("input_385_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_7_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_7_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95417408)))];
+            tensor<fp16, [1024]> module_layers_7_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_7_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95419520)))];
+            tensor<fp16, [1, 188, 1024]> input_385_cast_fp16 = layer_norm(axes = input_385_axes_0, beta = module_layers_7_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_feed_forward1_weight_to_fp16, x = input_383_cast_fp16)[name = string("input_385_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2072252]> module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95421632))), indices_nonzero_data = tensor<uint6, [2072252]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95945984))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97500288))))[name = string("module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_64_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_385_cast_fp16)[name = string("linear_64_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_389_cast_fp16 = silu(x = linear_64_cast_fp16)[name = string("input_389_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1978853]> module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97565888))), indices_nonzero_data = tensor<uint6, [1978853]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98090240))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99574464))))[name = string("module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_65_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_389_cast_fp16)[name = string("linear_65_cast_fp16")];
+            fp16 var_1373_to_fp16 = const()[name = string("op_1373_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_1374_cast_fp16 = mul(x = linear_65_cast_fp16, y = var_1373_to_fp16)[name = string("op_1374_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_395_cast_fp16 = add(x = input_383_cast_fp16, y = var_1374_cast_fp16)[name = string("input_395_cast_fp16")];
+            tensor<int32, [1]> query_15_axes_0 = const()[name = string("query_15_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_7_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_7_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99590912)))];
+            tensor<fp16, [1024]> module_layers_7_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_7_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99593024)))];
+            tensor<fp16, [1, 188, 1024]> query_15_cast_fp16 = layer_norm(axes = query_15_axes_0, beta = module_layers_7_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_self_att_weight_to_fp16, x = input_395_cast_fp16)[name = string("query_15_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522949]> module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99595136))), indices_nonzero_data = tensor<uint6, [522949]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99726272))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100118592))))[name = string("module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_66_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_15_cast_fp16)[name = string("linear_66_cast_fp16")];
+            tensor<int32, [4]> var_1390 = const()[name = string("op_1390"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_43_cast_fp16 = reshape(shape = var_1390, x = linear_66_cast_fp16)[name = string("q_43_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [515713]> module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100135040))), indices_nonzero_data = tensor<uint6, [515713]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100266176))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100653056))))[name = string("module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_67_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_15_cast_fp16)[name = string("linear_67_cast_fp16")];
+            tensor<int32, [4]> var_1394 = const()[name = string("op_1394"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_29_cast_fp16 = reshape(shape = var_1394, x = linear_67_cast_fp16)[name = string("k_29_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [507764]> module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100669504))), indices_nonzero_data = tensor<uint6, [507764]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100800640))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101181568))))[name = string("module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_68_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_15_cast_fp16)[name = string("linear_68_cast_fp16")];
+            tensor<int32, [4]> var_1398 = const()[name = string("op_1398"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_15_cast_fp16 = reshape(shape = var_1398, x = linear_68_cast_fp16)[name = string("v_15_cast_fp16")];
+            tensor<int32, [4]> value_17_perm_0 = const()[name = string("value_17_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_7_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_7_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101198016)))];
+            tensor<fp16, [1, 188, 8, 128]> var_1410_cast_fp16 = add(x = q_43_cast_fp16, y = module_layers_7_self_attn_pos_bias_u_to_fp16)[name = string("op_1410_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_7_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_7_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101200128)))];
+            tensor<fp16, [1, 188, 8, 128]> var_1412_cast_fp16 = add(x = q_43_cast_fp16, y = module_layers_7_self_attn_pos_bias_v_to_fp16)[name = string("op_1412_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_15_perm_0 = const()[name = string("q_with_bias_v_15_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_161_transpose_x_0 = const()[name = string("x_161_transpose_x_0"), val = bool(false)];
+            bool x_161_transpose_y_0 = const()[name = string("x_161_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_1414_to_fp16 = const()[name = string("op_1414_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(325334400)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_15_cast_fp16 = transpose(perm = q_with_bias_v_15_perm_0, x = var_1412_cast_fp16)[name = string("transpose_263")];
+            tensor<fp16, [1, 8, 188, 375]> x_161_cast_fp16 = matmul(transpose_x = x_161_transpose_x_0, transpose_y = x_161_transpose_y_0, x = q_with_bias_v_15_cast_fp16, y = var_1414_to_fp16)[name = string("x_161_cast_fp16")];
+            tensor<int32, [8]> x_163_pad_0 = const()[name = string("x_163_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_163_mode_0 = const()[name = string("x_163_mode_0"), val = string("constant")];
+            fp16 const_84_to_fp16 = const()[name = string("const_84_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_163_cast_fp16 = pad(constant_val = const_84_to_fp16, mode = x_163_mode_0, pad = x_163_pad_0, x = x_161_cast_fp16)[name = string("x_163_cast_fp16")];
+            tensor<int32, [4]> var_1422 = const()[name = string("op_1422"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_165_cast_fp16 = reshape(shape = var_1422, x = x_163_cast_fp16)[name = string("x_165_cast_fp16")];
+            tensor<int32, [4]> var_1426_begin_0 = const()[name = string("op_1426_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1426_end_0 = const()[name = string("op_1426_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_1426_end_mask_0 = const()[name = string("op_1426_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_1426_cast_fp16 = slice_by_index(begin = var_1426_begin_0, end = var_1426_end_0, end_mask = var_1426_end_mask_0, x = x_165_cast_fp16)[name = string("op_1426_cast_fp16")];
+            tensor<int32, [4]> var_1427 = const()[name = string("op_1427"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_29_cast_fp16 = reshape(shape = var_1427, x = var_1426_cast_fp16)[name = string("matrix_bd_29_cast_fp16")];
+            bool matrix_ac_15_transpose_x_0 = const()[name = string("matrix_ac_15_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_15_transpose_y_0 = const()[name = string("matrix_ac_15_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_110_perm_0 = const()[name = string("transpose_110_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_111_perm_0 = const()[name = string("transpose_111_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_111 = transpose(perm = transpose_111_perm_0, x = k_29_cast_fp16)[name = string("transpose_261")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_110 = transpose(perm = transpose_110_perm_0, x = var_1410_cast_fp16)[name = string("transpose_262")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_15_cast_fp16 = matmul(transpose_x = matrix_ac_15_transpose_x_0, transpose_y = matrix_ac_15_transpose_y_0, x = transpose_110, y = transpose_111)[name = string("matrix_ac_15_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_31_begin_0 = const()[name = string("matrix_bd_31_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_31_end_0 = const()[name = string("matrix_bd_31_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_31_end_mask_0 = const()[name = string("matrix_bd_31_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_31_cast_fp16 = slice_by_index(begin = matrix_bd_31_begin_0, end = matrix_bd_31_end_0, end_mask = matrix_bd_31_end_mask_0, x = matrix_bd_29_cast_fp16)[name = string("matrix_bd_31_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_1436_cast_fp16 = add(x = matrix_ac_15_cast_fp16, y = matrix_bd_31_cast_fp16)[name = string("op_1436_cast_fp16")];
+            fp16 _inversed_scores_29_y_0_to_fp16 = const()[name = string("_inversed_scores_29_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_29_cast_fp16 = mul(x = var_1436_cast_fp16, y = _inversed_scores_29_y_0_to_fp16)[name = string("_inversed_scores_29_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_31_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_29_cast_fp16, cond = mask_3)[name = string("scores_31_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_1442_cast_fp16 = softmax(axis = var_30, x = scores_31_cast_fp16)[name = string("op_1442_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_397_cast_fp16 = select(a = var_11_to_fp16, b = var_1442_cast_fp16, cond = mask_3)[name = string("input_397_cast_fp16")];
+            bool x_167_transpose_x_0 = const()[name = string("x_167_transpose_x_0"), val = bool(false)];
+            bool x_167_transpose_y_0 = const()[name = string("x_167_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_17_cast_fp16 = transpose(perm = value_17_perm_0, x = v_15_cast_fp16)[name = string("transpose_260")];
+            tensor<fp16, [1, 8, 188, 128]> x_167_cast_fp16 = matmul(transpose_x = x_167_transpose_x_0, transpose_y = x_167_transpose_y_0, x = input_397_cast_fp16, y = value_17_cast_fp16)[name = string("x_167_cast_fp16")];
+            tensor<int32, [4]> var_1446_perm_0 = const()[name = string("op_1446_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1447 = const()[name = string("op_1447"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_1446_cast_fp16 = transpose(perm = var_1446_perm_0, x = x_167_cast_fp16)[name = string("transpose_259")];
+            tensor<fp16, [1, 188, 1024]> input_399_cast_fp16 = reshape(shape = var_1447, x = var_1446_cast_fp16)[name = string("input_399_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525262]> module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101462400))), indices_nonzero_data = tensor<uint6, [525262]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101593536))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101987584))))[name = string("module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_70_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_399_cast_fp16)[name = string("linear_70_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_403_cast_fp16 = add(x = input_395_cast_fp16, y = linear_70_cast_fp16)[name = string("input_403_cast_fp16")];
+            tensor<int32, [1]> x_171_axes_0 = const()[name = string("x_171_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_7_norm_conv_weight_to_fp16 = const()[name = string("module_layers_7_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102004032)))];
+            tensor<fp16, [1024]> module_layers_7_norm_conv_bias_to_fp16 = const()[name = string("module_layers_7_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102006144)))];
+            tensor<fp16, [1, 188, 1024]> x_171_cast_fp16 = layer_norm(axes = x_171_axes_0, beta = module_layers_7_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_conv_weight_to_fp16, x = input_403_cast_fp16)[name = string("x_171_cast_fp16")];
+            tensor<int32, [3]> input_405_perm_0 = const()[name = string("input_405_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_407_pad_type_0 = const()[name = string("input_407_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_407_strides_0 = const()[name = string("input_407_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_407_pad_0 = const()[name = string("input_407_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_407_dilations_0 = const()[name = string("input_407_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_407_groups_0 = const()[name = string("input_407_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047562]> module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102008256))), indices_nonzero_data = tensor<uint6, [1047562]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102270464))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103056256))))[name = string("module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_405_cast_fp16 = transpose(perm = input_405_perm_0, x = x_171_cast_fp16)[name = string("transpose_258")];
+            tensor<fp16, [1, 2048, 188]> input_407_cast_fp16 = conv(dilations = input_407_dilations_0, groups = input_407_groups_0, pad = input_407_pad_0, pad_type = input_407_pad_type_0, strides = input_407_strides_0, weight = module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_405_cast_fp16)[name = string("input_407_cast_fp16")];
+            int32 x_173_split_num_splits_0 = const()[name = string("x_173_split_num_splits_0"), val = int32(2)];
+            int32 x_173_split_axis_0 = const()[name = string("x_173_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_173_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_173_split_cast_fp16_1 = split(axis = x_173_split_axis_0, num_splits = x_173_split_num_splits_0, x = input_407_cast_fp16)[name = string("x_173_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_173_split_1_sigmoid_cast_fp16 = sigmoid(x = x_173_split_cast_fp16_1)[name = string("x_173_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_173_cast_fp16 = mul(x = x_173_split_cast_fp16_0, y = x_173_split_1_sigmoid_cast_fp16)[name = string("x_173_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_409_cast_fp16 = select(a = var_11_to_fp16, b = x_173_cast_fp16, cond = var_328)[name = string("input_409_cast_fp16")];
+            tensor<int32, [6]> input_411_pad_0 = const()[name = string("input_411_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_411_mode_0 = const()[name = string("input_411_mode_0"), val = string("constant")];
+            fp16 const_87_to_fp16 = const()[name = string("const_87_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_411_cast_fp16 = pad(constant_val = const_87_to_fp16, mode = input_411_mode_0, pad = input_411_pad_0, x = input_409_cast_fp16)[name = string("input_411_cast_fp16")];
+            string input_413_pad_type_0 = const()[name = string("input_413_pad_type_0"), val = string("valid")];
+            int32 input_413_groups_0 = const()[name = string("input_413_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_413_strides_0 = const()[name = string("input_413_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_413_pad_0 = const()[name = string("input_413_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_413_dilations_0 = const()[name = string("input_413_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_262_to_fp16 = const()[name = string("const_262_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103089088)))];
+            tensor<fp16, [1024]> const_263_to_fp16 = const()[name = string("const_263_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103107584)))];
+            tensor<fp16, [1, 1024, 188]> input_415_cast_fp16 = conv(bias = const_263_to_fp16, dilations = input_413_dilations_0, groups = input_413_groups_0, pad = input_413_pad_0, pad_type = input_413_pad_type_0, strides = input_413_strides_0, weight = const_262_to_fp16, x = input_411_cast_fp16)[name = string("input_415_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_417_cast_fp16 = silu(x = input_415_cast_fp16)[name = string("input_417_cast_fp16")];
+            string x_175_pad_type_0 = const()[name = string("x_175_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_175_strides_0 = const()[name = string("x_175_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_175_pad_0 = const()[name = string("x_175_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_175_dilations_0 = const()[name = string("x_175_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_175_groups_0 = const()[name = string("x_175_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519522]> module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103109696))), indices_nonzero_data = tensor<uint6, [519522]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103240832))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103630592))))[name = string("module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_175_cast_fp16 = conv(dilations = x_175_dilations_0, groups = x_175_groups_0, pad = x_175_pad_0, pad_type = x_175_pad_type_0, strides = x_175_strides_0, weight = module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_417_cast_fp16)[name = string("x_175_cast_fp16")];
+            tensor<int32, [3]> input_419_perm_0 = const()[name = string("input_419_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_419_cast_fp16 = transpose(perm = input_419_perm_0, x = x_175_cast_fp16)[name = string("transpose_257")];
+            tensor<fp16, [1, 188, 1024]> input_421_cast_fp16 = add(x = input_403_cast_fp16, y = input_419_cast_fp16)[name = string("input_421_cast_fp16")];
+            tensor<int32, [1]> input_423_axes_0 = const()[name = string("input_423_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_7_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_7_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103647040)))];
+            tensor<fp16, [1024]> module_layers_7_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_7_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103649152)))];
+            tensor<fp16, [1, 188, 1024]> input_423_cast_fp16 = layer_norm(axes = input_423_axes_0, beta = module_layers_7_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_feed_forward2_weight_to_fp16, x = input_421_cast_fp16)[name = string("input_423_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2062964]> module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103651264))), indices_nonzero_data = tensor<uint6, [2062964]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104175616))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105722944))))[name = string("module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_71_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_423_cast_fp16)[name = string("linear_71_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_427_cast_fp16 = silu(x = linear_71_cast_fp16)[name = string("input_427_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2021791]> module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105788544))), indices_nonzero_data = tensor<uint6, [2021791]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106312896))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107829312))))[name = string("module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_72_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_427_cast_fp16)[name = string("linear_72_cast_fp16")];
+            fp16 var_1507_to_fp16 = const()[name = string("op_1507_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_1508_cast_fp16 = mul(x = linear_72_cast_fp16, y = var_1507_to_fp16)[name = string("op_1508_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_433_cast_fp16 = add(x = input_421_cast_fp16, y = var_1508_cast_fp16)[name = string("input_433_cast_fp16")];
+            tensor<int32, [1]> input_435_axes_0 = const()[name = string("input_435_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_7_norm_out_weight_to_fp16 = const()[name = string("module_layers_7_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107845760)))];
+            tensor<fp16, [1024]> module_layers_7_norm_out_bias_to_fp16 = const()[name = string("module_layers_7_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107847872)))];
+            tensor<fp16, [1, 188, 1024]> input_435_cast_fp16 = layer_norm(axes = input_435_axes_0, beta = module_layers_7_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_out_weight_to_fp16, x = input_433_cast_fp16)[name = string("input_435_cast_fp16")];
+            tensor<int32, [1]> input_437_axes_0 = const()[name = string("input_437_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_8_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_8_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107849984)))];
+            tensor<fp16, [1024]> module_layers_8_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_8_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107852096)))];
+            tensor<fp16, [1, 188, 1024]> input_437_cast_fp16 = layer_norm(axes = input_437_axes_0, beta = module_layers_8_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_feed_forward1_weight_to_fp16, x = input_435_cast_fp16)[name = string("input_437_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2070011]> module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107854208))), indices_nonzero_data = tensor<uint6, [2070011]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108378560))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109931136))))[name = string("module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_73_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_437_cast_fp16)[name = string("linear_73_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_441_cast_fp16 = silu(x = linear_73_cast_fp16)[name = string("input_441_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1995292]> module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109996736))), indices_nonzero_data = tensor<uint6, [1995292]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110521088))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112017664))))[name = string("module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_74_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_441_cast_fp16)[name = string("linear_74_cast_fp16")];
+            fp16 var_1536_to_fp16 = const()[name = string("op_1536_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_1537_cast_fp16 = mul(x = linear_74_cast_fp16, y = var_1536_to_fp16)[name = string("op_1537_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_447_cast_fp16 = add(x = input_435_cast_fp16, y = var_1537_cast_fp16)[name = string("input_447_cast_fp16")];
+            tensor<int32, [1]> query_17_axes_0 = const()[name = string("query_17_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_8_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_8_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112034112)))];
+            tensor<fp16, [1024]> module_layers_8_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_8_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112036224)))];
+            tensor<fp16, [1, 188, 1024]> query_17_cast_fp16 = layer_norm(axes = query_17_axes_0, beta = module_layers_8_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_self_att_weight_to_fp16, x = input_447_cast_fp16)[name = string("query_17_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522805]> module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112038336))), indices_nonzero_data = tensor<uint6, [522805]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112169472))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112561664))))[name = string("module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_75_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_17_cast_fp16)[name = string("linear_75_cast_fp16")];
+            tensor<int32, [4]> var_1553 = const()[name = string("op_1553"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_49_cast_fp16 = reshape(shape = var_1553, x = linear_75_cast_fp16)[name = string("q_49_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521269]> module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112578112))), indices_nonzero_data = tensor<uint6, [521269]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112709248))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113100288))))[name = string("module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_76_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_17_cast_fp16)[name = string("linear_76_cast_fp16")];
+            tensor<int32, [4]> var_1557 = const()[name = string("op_1557"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_33_cast_fp16 = reshape(shape = var_1557, x = linear_76_cast_fp16)[name = string("k_33_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [517626]> module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113116736))), indices_nonzero_data = tensor<uint6, [517626]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113247872))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113636160))))[name = string("module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_77_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_17_cast_fp16)[name = string("linear_77_cast_fp16")];
+            tensor<int32, [4]> var_1561 = const()[name = string("op_1561"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_17_cast_fp16 = reshape(shape = var_1561, x = linear_77_cast_fp16)[name = string("v_17_cast_fp16")];
+            tensor<int32, [4]> value_19_perm_0 = const()[name = string("value_19_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_8_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_8_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113652608)))];
+            tensor<fp16, [1, 188, 8, 128]> var_1573_cast_fp16 = add(x = q_49_cast_fp16, y = module_layers_8_self_attn_pos_bias_u_to_fp16)[name = string("op_1573_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_8_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_8_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113654720)))];
+            tensor<fp16, [1, 188, 8, 128]> var_1575_cast_fp16 = add(x = q_49_cast_fp16, y = module_layers_8_self_attn_pos_bias_v_to_fp16)[name = string("op_1575_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_17_perm_0 = const()[name = string("q_with_bias_v_17_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_183_transpose_x_0 = const()[name = string("x_183_transpose_x_0"), val = bool(false)];
+            bool x_183_transpose_y_0 = const()[name = string("x_183_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_1577_to_fp16 = const()[name = string("op_1577_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326102464)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_17_cast_fp16 = transpose(perm = q_with_bias_v_17_perm_0, x = var_1575_cast_fp16)[name = string("transpose_256")];
+            tensor<fp16, [1, 8, 188, 375]> x_183_cast_fp16 = matmul(transpose_x = x_183_transpose_x_0, transpose_y = x_183_transpose_y_0, x = q_with_bias_v_17_cast_fp16, y = var_1577_to_fp16)[name = string("x_183_cast_fp16")];
+            tensor<int32, [8]> x_185_pad_0 = const()[name = string("x_185_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_185_mode_0 = const()[name = string("x_185_mode_0"), val = string("constant")];
+            fp16 const_94_to_fp16 = const()[name = string("const_94_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_185_cast_fp16 = pad(constant_val = const_94_to_fp16, mode = x_185_mode_0, pad = x_185_pad_0, x = x_183_cast_fp16)[name = string("x_185_cast_fp16")];
+            tensor<int32, [4]> var_1585 = const()[name = string("op_1585"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_187_cast_fp16 = reshape(shape = var_1585, x = x_185_cast_fp16)[name = string("x_187_cast_fp16")];
+            tensor<int32, [4]> var_1589_begin_0 = const()[name = string("op_1589_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1589_end_0 = const()[name = string("op_1589_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_1589_end_mask_0 = const()[name = string("op_1589_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_1589_cast_fp16 = slice_by_index(begin = var_1589_begin_0, end = var_1589_end_0, end_mask = var_1589_end_mask_0, x = x_187_cast_fp16)[name = string("op_1589_cast_fp16")];
+            tensor<int32, [4]> var_1590 = const()[name = string("op_1590"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_33_cast_fp16 = reshape(shape = var_1590, x = var_1589_cast_fp16)[name = string("matrix_bd_33_cast_fp16")];
+            bool matrix_ac_17_transpose_x_0 = const()[name = string("matrix_ac_17_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_17_transpose_y_0 = const()[name = string("matrix_ac_17_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_112_perm_0 = const()[name = string("transpose_112_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_113_perm_0 = const()[name = string("transpose_113_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_113 = transpose(perm = transpose_113_perm_0, x = k_33_cast_fp16)[name = string("transpose_254")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_112 = transpose(perm = transpose_112_perm_0, x = var_1573_cast_fp16)[name = string("transpose_255")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_17_cast_fp16 = matmul(transpose_x = matrix_ac_17_transpose_x_0, transpose_y = matrix_ac_17_transpose_y_0, x = transpose_112, y = transpose_113)[name = string("matrix_ac_17_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_35_begin_0 = const()[name = string("matrix_bd_35_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_35_end_0 = const()[name = string("matrix_bd_35_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_35_end_mask_0 = const()[name = string("matrix_bd_35_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_35_cast_fp16 = slice_by_index(begin = matrix_bd_35_begin_0, end = matrix_bd_35_end_0, end_mask = matrix_bd_35_end_mask_0, x = matrix_bd_33_cast_fp16)[name = string("matrix_bd_35_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_1599_cast_fp16 = add(x = matrix_ac_17_cast_fp16, y = matrix_bd_35_cast_fp16)[name = string("op_1599_cast_fp16")];
+            fp16 _inversed_scores_33_y_0_to_fp16 = const()[name = string("_inversed_scores_33_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_33_cast_fp16 = mul(x = var_1599_cast_fp16, y = _inversed_scores_33_y_0_to_fp16)[name = string("_inversed_scores_33_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_35_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_33_cast_fp16, cond = mask_3)[name = string("scores_35_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_1605_cast_fp16 = softmax(axis = var_30, x = scores_35_cast_fp16)[name = string("op_1605_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_449_cast_fp16 = select(a = var_11_to_fp16, b = var_1605_cast_fp16, cond = mask_3)[name = string("input_449_cast_fp16")];
+            bool x_189_transpose_x_0 = const()[name = string("x_189_transpose_x_0"), val = bool(false)];
+            bool x_189_transpose_y_0 = const()[name = string("x_189_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_19_cast_fp16 = transpose(perm = value_19_perm_0, x = v_17_cast_fp16)[name = string("transpose_253")];
+            tensor<fp16, [1, 8, 188, 128]> x_189_cast_fp16 = matmul(transpose_x = x_189_transpose_x_0, transpose_y = x_189_transpose_y_0, x = input_449_cast_fp16, y = value_19_cast_fp16)[name = string("x_189_cast_fp16")];
+            tensor<int32, [4]> var_1609_perm_0 = const()[name = string("op_1609_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1610 = const()[name = string("op_1610"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_1609_cast_fp16 = transpose(perm = var_1609_perm_0, x = x_189_cast_fp16)[name = string("transpose_252")];
+            tensor<fp16, [1, 188, 1024]> input_451_cast_fp16 = reshape(shape = var_1610, x = var_1609_cast_fp16)[name = string("input_451_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522953]> module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113916992))), indices_nonzero_data = tensor<uint6, [522953]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114048128))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114440448))))[name = string("module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_79_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_451_cast_fp16)[name = string("linear_79_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_455_cast_fp16 = add(x = input_447_cast_fp16, y = linear_79_cast_fp16)[name = string("input_455_cast_fp16")];
+            tensor<int32, [1]> x_193_axes_0 = const()[name = string("x_193_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_8_norm_conv_weight_to_fp16 = const()[name = string("module_layers_8_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114456896)))];
+            tensor<fp16, [1024]> module_layers_8_norm_conv_bias_to_fp16 = const()[name = string("module_layers_8_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114459008)))];
+            tensor<fp16, [1, 188, 1024]> x_193_cast_fp16 = layer_norm(axes = x_193_axes_0, beta = module_layers_8_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_conv_weight_to_fp16, x = input_455_cast_fp16)[name = string("x_193_cast_fp16")];
+            tensor<int32, [3]> input_457_perm_0 = const()[name = string("input_457_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_459_pad_type_0 = const()[name = string("input_459_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_459_strides_0 = const()[name = string("input_459_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_459_pad_0 = const()[name = string("input_459_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_459_dilations_0 = const()[name = string("input_459_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_459_groups_0 = const()[name = string("input_459_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1050224]> module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114461120))), indices_nonzero_data = tensor<uint6, [1050224]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114723328))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115511104))))[name = string("module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_457_cast_fp16 = transpose(perm = input_457_perm_0, x = x_193_cast_fp16)[name = string("transpose_251")];
+            tensor<fp16, [1, 2048, 188]> input_459_cast_fp16 = conv(dilations = input_459_dilations_0, groups = input_459_groups_0, pad = input_459_pad_0, pad_type = input_459_pad_type_0, strides = input_459_strides_0, weight = module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_457_cast_fp16)[name = string("input_459_cast_fp16")];
+            int32 x_195_split_num_splits_0 = const()[name = string("x_195_split_num_splits_0"), val = int32(2)];
+            int32 x_195_split_axis_0 = const()[name = string("x_195_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_195_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_195_split_cast_fp16_1 = split(axis = x_195_split_axis_0, num_splits = x_195_split_num_splits_0, x = input_459_cast_fp16)[name = string("x_195_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_195_split_1_sigmoid_cast_fp16 = sigmoid(x = x_195_split_cast_fp16_1)[name = string("x_195_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_195_cast_fp16 = mul(x = x_195_split_cast_fp16_0, y = x_195_split_1_sigmoid_cast_fp16)[name = string("x_195_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_461_cast_fp16 = select(a = var_11_to_fp16, b = x_195_cast_fp16, cond = var_328)[name = string("input_461_cast_fp16")];
+            tensor<int32, [6]> input_463_pad_0 = const()[name = string("input_463_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_463_mode_0 = const()[name = string("input_463_mode_0"), val = string("constant")];
+            fp16 const_97_to_fp16 = const()[name = string("const_97_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_463_cast_fp16 = pad(constant_val = const_97_to_fp16, mode = input_463_mode_0, pad = input_463_pad_0, x = input_461_cast_fp16)[name = string("input_463_cast_fp16")];
+            string input_465_pad_type_0 = const()[name = string("input_465_pad_type_0"), val = string("valid")];
+            int32 input_465_groups_0 = const()[name = string("input_465_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_465_strides_0 = const()[name = string("input_465_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_465_pad_0 = const()[name = string("input_465_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_465_dilations_0 = const()[name = string("input_465_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_264_to_fp16 = const()[name = string("const_264_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115543936)))];
+            tensor<fp16, [1024]> const_265_to_fp16 = const()[name = string("const_265_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115562432)))];
+            tensor<fp16, [1, 1024, 188]> input_467_cast_fp16 = conv(bias = const_265_to_fp16, dilations = input_465_dilations_0, groups = input_465_groups_0, pad = input_465_pad_0, pad_type = input_465_pad_type_0, strides = input_465_strides_0, weight = const_264_to_fp16, x = input_463_cast_fp16)[name = string("input_467_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_469_cast_fp16 = silu(x = input_467_cast_fp16)[name = string("input_469_cast_fp16")];
+            string x_197_pad_type_0 = const()[name = string("x_197_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_197_strides_0 = const()[name = string("x_197_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_197_pad_0 = const()[name = string("x_197_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_197_dilations_0 = const()[name = string("x_197_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_197_groups_0 = const()[name = string("x_197_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521521]> module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115564544))), indices_nonzero_data = tensor<uint6, [521521]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115695680))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116086912))))[name = string("module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_197_cast_fp16 = conv(dilations = x_197_dilations_0, groups = x_197_groups_0, pad = x_197_pad_0, pad_type = x_197_pad_type_0, strides = x_197_strides_0, weight = module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_469_cast_fp16)[name = string("x_197_cast_fp16")];
+            tensor<int32, [3]> input_471_perm_0 = const()[name = string("input_471_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_471_cast_fp16 = transpose(perm = input_471_perm_0, x = x_197_cast_fp16)[name = string("transpose_250")];
+            tensor<fp16, [1, 188, 1024]> input_473_cast_fp16 = add(x = input_455_cast_fp16, y = input_471_cast_fp16)[name = string("input_473_cast_fp16")];
+            tensor<int32, [1]> input_475_axes_0 = const()[name = string("input_475_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_8_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_8_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116103360)))];
+            tensor<fp16, [1024]> module_layers_8_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_8_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116105472)))];
+            tensor<fp16, [1, 188, 1024]> input_475_cast_fp16 = layer_norm(axes = input_475_axes_0, beta = module_layers_8_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_feed_forward2_weight_to_fp16, x = input_473_cast_fp16)[name = string("input_475_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2065024]> module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116107584))), indices_nonzero_data = tensor<uint6, [2065024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116631936))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118180800))))[name = string("module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_80_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_475_cast_fp16)[name = string("linear_80_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_479_cast_fp16 = silu(x = linear_80_cast_fp16)[name = string("input_479_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2041094]> module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118246400))), indices_nonzero_data = tensor<uint6, [2041094]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118770752))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120301696))))[name = string("module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_81_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_479_cast_fp16)[name = string("linear_81_cast_fp16")];
+            fp16 var_1670_to_fp16 = const()[name = string("op_1670_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_1671_cast_fp16 = mul(x = linear_81_cast_fp16, y = var_1670_to_fp16)[name = string("op_1671_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_485_cast_fp16 = add(x = input_473_cast_fp16, y = var_1671_cast_fp16)[name = string("input_485_cast_fp16")];
+            tensor<int32, [1]> input_487_axes_0 = const()[name = string("input_487_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_8_norm_out_weight_to_fp16 = const()[name = string("module_layers_8_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120318144)))];
+            tensor<fp16, [1024]> module_layers_8_norm_out_bias_to_fp16 = const()[name = string("module_layers_8_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120320256)))];
+            tensor<fp16, [1, 188, 1024]> input_487_cast_fp16 = layer_norm(axes = input_487_axes_0, beta = module_layers_8_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_out_weight_to_fp16, x = input_485_cast_fp16)[name = string("input_487_cast_fp16")];
+            tensor<int32, [1]> input_489_axes_0 = const()[name = string("input_489_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_9_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_9_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120322368)))];
+            tensor<fp16, [1024]> module_layers_9_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_9_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120324480)))];
+            tensor<fp16, [1, 188, 1024]> input_489_cast_fp16 = layer_norm(axes = input_489_axes_0, beta = module_layers_9_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_feed_forward1_weight_to_fp16, x = input_487_cast_fp16)[name = string("input_489_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2056248]> module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120326592))), indices_nonzero_data = tensor<uint6, [2056248]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120850944))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122393216))))[name = string("module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_82_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_489_cast_fp16)[name = string("linear_82_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_493_cast_fp16 = silu(x = linear_82_cast_fp16)[name = string("input_493_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1988988]> module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122458816))), indices_nonzero_data = tensor<uint6, [1988988]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122983168))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124475008))))[name = string("module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_83_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_493_cast_fp16)[name = string("linear_83_cast_fp16")];
+            fp16 var_1699_to_fp16 = const()[name = string("op_1699_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_1700_cast_fp16 = mul(x = linear_83_cast_fp16, y = var_1699_to_fp16)[name = string("op_1700_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_499_cast_fp16 = add(x = input_487_cast_fp16, y = var_1700_cast_fp16)[name = string("input_499_cast_fp16")];
+            tensor<int32, [1]> query_19_axes_0 = const()[name = string("query_19_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_9_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_9_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124491456)))];
+            tensor<fp16, [1024]> module_layers_9_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_9_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124493568)))];
+            tensor<fp16, [1, 188, 1024]> query_19_cast_fp16 = layer_norm(axes = query_19_axes_0, beta = module_layers_9_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_self_att_weight_to_fp16, x = input_499_cast_fp16)[name = string("query_19_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524317]> module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124495680))), indices_nonzero_data = tensor<uint6, [524317]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124626816))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125020160))))[name = string("module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_84_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_19_cast_fp16)[name = string("linear_84_cast_fp16")];
+            tensor<int32, [4]> var_1716 = const()[name = string("op_1716"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_55_cast_fp16 = reshape(shape = var_1716, x = linear_84_cast_fp16)[name = string("q_55_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524008]> module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125036608))), indices_nonzero_data = tensor<uint6, [524008]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125167744))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125560832))))[name = string("module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_85_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_19_cast_fp16)[name = string("linear_85_cast_fp16")];
+            tensor<int32, [4]> var_1720 = const()[name = string("op_1720"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_37_cast_fp16 = reshape(shape = var_1720, x = linear_85_cast_fp16)[name = string("k_37_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524728]> module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125577280))), indices_nonzero_data = tensor<uint6, [524728]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125708416))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126102080))))[name = string("module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_86_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_19_cast_fp16)[name = string("linear_86_cast_fp16")];
+            tensor<int32, [4]> var_1724 = const()[name = string("op_1724"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_19_cast_fp16 = reshape(shape = var_1724, x = linear_86_cast_fp16)[name = string("v_19_cast_fp16")];
+            tensor<int32, [4]> value_21_perm_0 = const()[name = string("value_21_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_9_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_9_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126118528)))];
+            tensor<fp16, [1, 188, 8, 128]> var_1736_cast_fp16 = add(x = q_55_cast_fp16, y = module_layers_9_self_attn_pos_bias_u_to_fp16)[name = string("op_1736_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_9_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_9_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126120640)))];
+            tensor<fp16, [1, 188, 8, 128]> var_1738_cast_fp16 = add(x = q_55_cast_fp16, y = module_layers_9_self_attn_pos_bias_v_to_fp16)[name = string("op_1738_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_19_perm_0 = const()[name = string("q_with_bias_v_19_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_205_transpose_x_0 = const()[name = string("x_205_transpose_x_0"), val = bool(false)];
+            bool x_205_transpose_y_0 = const()[name = string("x_205_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_1740_to_fp16 = const()[name = string("op_1740_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326870528)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_19_cast_fp16 = transpose(perm = q_with_bias_v_19_perm_0, x = var_1738_cast_fp16)[name = string("transpose_249")];
+            tensor<fp16, [1, 8, 188, 375]> x_205_cast_fp16 = matmul(transpose_x = x_205_transpose_x_0, transpose_y = x_205_transpose_y_0, x = q_with_bias_v_19_cast_fp16, y = var_1740_to_fp16)[name = string("x_205_cast_fp16")];
+            tensor<int32, [8]> x_207_pad_0 = const()[name = string("x_207_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_207_mode_0 = const()[name = string("x_207_mode_0"), val = string("constant")];
+            fp16 const_104_to_fp16 = const()[name = string("const_104_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_207_cast_fp16 = pad(constant_val = const_104_to_fp16, mode = x_207_mode_0, pad = x_207_pad_0, x = x_205_cast_fp16)[name = string("x_207_cast_fp16")];
+            tensor<int32, [4]> var_1748 = const()[name = string("op_1748"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_209_cast_fp16 = reshape(shape = var_1748, x = x_207_cast_fp16)[name = string("x_209_cast_fp16")];
+            tensor<int32, [4]> var_1752_begin_0 = const()[name = string("op_1752_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1752_end_0 = const()[name = string("op_1752_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_1752_end_mask_0 = const()[name = string("op_1752_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_1752_cast_fp16 = slice_by_index(begin = var_1752_begin_0, end = var_1752_end_0, end_mask = var_1752_end_mask_0, x = x_209_cast_fp16)[name = string("op_1752_cast_fp16")];
+            tensor<int32, [4]> var_1753 = const()[name = string("op_1753"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_37_cast_fp16 = reshape(shape = var_1753, x = var_1752_cast_fp16)[name = string("matrix_bd_37_cast_fp16")];
+            bool matrix_ac_19_transpose_x_0 = const()[name = string("matrix_ac_19_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_19_transpose_y_0 = const()[name = string("matrix_ac_19_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_114_perm_0 = const()[name = string("transpose_114_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_115_perm_0 = const()[name = string("transpose_115_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_115 = transpose(perm = transpose_115_perm_0, x = k_37_cast_fp16)[name = string("transpose_247")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_114 = transpose(perm = transpose_114_perm_0, x = var_1736_cast_fp16)[name = string("transpose_248")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_19_cast_fp16 = matmul(transpose_x = matrix_ac_19_transpose_x_0, transpose_y = matrix_ac_19_transpose_y_0, x = transpose_114, y = transpose_115)[name = string("matrix_ac_19_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_39_begin_0 = const()[name = string("matrix_bd_39_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_39_end_0 = const()[name = string("matrix_bd_39_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_39_end_mask_0 = const()[name = string("matrix_bd_39_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_39_cast_fp16 = slice_by_index(begin = matrix_bd_39_begin_0, end = matrix_bd_39_end_0, end_mask = matrix_bd_39_end_mask_0, x = matrix_bd_37_cast_fp16)[name = string("matrix_bd_39_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_1762_cast_fp16 = add(x = matrix_ac_19_cast_fp16, y = matrix_bd_39_cast_fp16)[name = string("op_1762_cast_fp16")];
+            fp16 _inversed_scores_37_y_0_to_fp16 = const()[name = string("_inversed_scores_37_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_37_cast_fp16 = mul(x = var_1762_cast_fp16, y = _inversed_scores_37_y_0_to_fp16)[name = string("_inversed_scores_37_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_39_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_37_cast_fp16, cond = mask_3)[name = string("scores_39_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_1768_cast_fp16 = softmax(axis = var_30, x = scores_39_cast_fp16)[name = string("op_1768_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_501_cast_fp16 = select(a = var_11_to_fp16, b = var_1768_cast_fp16, cond = mask_3)[name = string("input_501_cast_fp16")];
+            bool x_211_transpose_x_0 = const()[name = string("x_211_transpose_x_0"), val = bool(false)];
+            bool x_211_transpose_y_0 = const()[name = string("x_211_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_21_cast_fp16 = transpose(perm = value_21_perm_0, x = v_19_cast_fp16)[name = string("transpose_246")];
+            tensor<fp16, [1, 8, 188, 128]> x_211_cast_fp16 = matmul(transpose_x = x_211_transpose_x_0, transpose_y = x_211_transpose_y_0, x = input_501_cast_fp16, y = value_21_cast_fp16)[name = string("x_211_cast_fp16")];
+            tensor<int32, [4]> var_1772_perm_0 = const()[name = string("op_1772_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1773 = const()[name = string("op_1773"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_1772_cast_fp16 = transpose(perm = var_1772_perm_0, x = x_211_cast_fp16)[name = string("transpose_245")];
+            tensor<fp16, [1, 188, 1024]> input_503_cast_fp16 = reshape(shape = var_1773, x = var_1772_cast_fp16)[name = string("input_503_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524470]> module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126382912))), indices_nonzero_data = tensor<uint6, [524470]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126514048))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126907520))))[name = string("module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_88_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_503_cast_fp16)[name = string("linear_88_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_507_cast_fp16 = add(x = input_499_cast_fp16, y = linear_88_cast_fp16)[name = string("input_507_cast_fp16")];
+            tensor<int32, [1]> x_215_axes_0 = const()[name = string("x_215_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_9_norm_conv_weight_to_fp16 = const()[name = string("module_layers_9_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126923968)))];
+            tensor<fp16, [1024]> module_layers_9_norm_conv_bias_to_fp16 = const()[name = string("module_layers_9_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126926080)))];
+            tensor<fp16, [1, 188, 1024]> x_215_cast_fp16 = layer_norm(axes = x_215_axes_0, beta = module_layers_9_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_conv_weight_to_fp16, x = input_507_cast_fp16)[name = string("x_215_cast_fp16")];
+            tensor<int32, [3]> input_509_perm_0 = const()[name = string("input_509_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_511_pad_type_0 = const()[name = string("input_511_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_511_strides_0 = const()[name = string("input_511_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_511_pad_0 = const()[name = string("input_511_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_511_dilations_0 = const()[name = string("input_511_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_511_groups_0 = const()[name = string("input_511_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1050959]> module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126928192))), indices_nonzero_data = tensor<uint6, [1050959]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127190400))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127978688))))[name = string("module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_509_cast_fp16 = transpose(perm = input_509_perm_0, x = x_215_cast_fp16)[name = string("transpose_244")];
+            tensor<fp16, [1, 2048, 188]> input_511_cast_fp16 = conv(dilations = input_511_dilations_0, groups = input_511_groups_0, pad = input_511_pad_0, pad_type = input_511_pad_type_0, strides = input_511_strides_0, weight = module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_509_cast_fp16)[name = string("input_511_cast_fp16")];
+            int32 x_217_split_num_splits_0 = const()[name = string("x_217_split_num_splits_0"), val = int32(2)];
+            int32 x_217_split_axis_0 = const()[name = string("x_217_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_217_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_217_split_cast_fp16_1 = split(axis = x_217_split_axis_0, num_splits = x_217_split_num_splits_0, x = input_511_cast_fp16)[name = string("x_217_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_217_split_1_sigmoid_cast_fp16 = sigmoid(x = x_217_split_cast_fp16_1)[name = string("x_217_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_217_cast_fp16 = mul(x = x_217_split_cast_fp16_0, y = x_217_split_1_sigmoid_cast_fp16)[name = string("x_217_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_513_cast_fp16 = select(a = var_11_to_fp16, b = x_217_cast_fp16, cond = var_328)[name = string("input_513_cast_fp16")];
+            tensor<int32, [6]> input_515_pad_0 = const()[name = string("input_515_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_515_mode_0 = const()[name = string("input_515_mode_0"), val = string("constant")];
+            fp16 const_107_to_fp16 = const()[name = string("const_107_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_515_cast_fp16 = pad(constant_val = const_107_to_fp16, mode = input_515_mode_0, pad = input_515_pad_0, x = input_513_cast_fp16)[name = string("input_515_cast_fp16")];
+            string input_517_pad_type_0 = const()[name = string("input_517_pad_type_0"), val = string("valid")];
+            int32 input_517_groups_0 = const()[name = string("input_517_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_517_strides_0 = const()[name = string("input_517_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_517_pad_0 = const()[name = string("input_517_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_517_dilations_0 = const()[name = string("input_517_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_266_to_fp16 = const()[name = string("const_266_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128011520)))];
+            tensor<fp16, [1024]> const_267_to_fp16 = const()[name = string("const_267_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128030016)))];
+            tensor<fp16, [1, 1024, 188]> input_519_cast_fp16 = conv(bias = const_267_to_fp16, dilations = input_517_dilations_0, groups = input_517_groups_0, pad = input_517_pad_0, pad_type = input_517_pad_type_0, strides = input_517_strides_0, weight = const_266_to_fp16, x = input_515_cast_fp16)[name = string("input_519_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_521_cast_fp16 = silu(x = input_519_cast_fp16)[name = string("input_521_cast_fp16")];
+            string x_219_pad_type_0 = const()[name = string("x_219_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_219_strides_0 = const()[name = string("x_219_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_219_pad_0 = const()[name = string("x_219_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_219_dilations_0 = const()[name = string("x_219_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_219_groups_0 = const()[name = string("x_219_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519248]> module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128032128))), indices_nonzero_data = tensor<uint6, [519248]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128163264))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128552768))))[name = string("module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_219_cast_fp16 = conv(dilations = x_219_dilations_0, groups = x_219_groups_0, pad = x_219_pad_0, pad_type = x_219_pad_type_0, strides = x_219_strides_0, weight = module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_521_cast_fp16)[name = string("x_219_cast_fp16")];
+            tensor<int32, [3]> input_523_perm_0 = const()[name = string("input_523_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_523_cast_fp16 = transpose(perm = input_523_perm_0, x = x_219_cast_fp16)[name = string("transpose_243")];
+            tensor<fp16, [1, 188, 1024]> input_525_cast_fp16 = add(x = input_507_cast_fp16, y = input_523_cast_fp16)[name = string("input_525_cast_fp16")];
+            tensor<int32, [1]> input_527_axes_0 = const()[name = string("input_527_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_9_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_9_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128569216)))];
+            tensor<fp16, [1024]> module_layers_9_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_9_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128571328)))];
+            tensor<fp16, [1, 188, 1024]> input_527_cast_fp16 = layer_norm(axes = input_527_axes_0, beta = module_layers_9_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_feed_forward2_weight_to_fp16, x = input_525_cast_fp16)[name = string("input_527_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2059311]> module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128573440))), indices_nonzero_data = tensor<uint6, [2059311]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129097792))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130642368))))[name = string("module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_89_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_527_cast_fp16)[name = string("linear_89_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_531_cast_fp16 = silu(x = linear_89_cast_fp16)[name = string("input_531_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2055301]> module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130707968))), indices_nonzero_data = tensor<uint6, [2055301]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131232320))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132773888))))[name = string("module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_90_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_531_cast_fp16)[name = string("linear_90_cast_fp16")];
+            fp16 var_1833_to_fp16 = const()[name = string("op_1833_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_1834_cast_fp16 = mul(x = linear_90_cast_fp16, y = var_1833_to_fp16)[name = string("op_1834_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_537_cast_fp16 = add(x = input_525_cast_fp16, y = var_1834_cast_fp16)[name = string("input_537_cast_fp16")];
+            tensor<int32, [1]> input_539_axes_0 = const()[name = string("input_539_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_9_norm_out_weight_to_fp16 = const()[name = string("module_layers_9_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132790336)))];
+            tensor<fp16, [1024]> module_layers_9_norm_out_bias_to_fp16 = const()[name = string("module_layers_9_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132792448)))];
+            tensor<fp16, [1, 188, 1024]> input_539_cast_fp16 = layer_norm(axes = input_539_axes_0, beta = module_layers_9_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_out_weight_to_fp16, x = input_537_cast_fp16)[name = string("input_539_cast_fp16")];
+            tensor<int32, [1]> input_541_axes_0 = const()[name = string("input_541_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_10_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_10_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132794560)))];
+            tensor<fp16, [1024]> module_layers_10_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_10_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132796672)))];
+            tensor<fp16, [1, 188, 1024]> input_541_cast_fp16 = layer_norm(axes = input_541_axes_0, beta = module_layers_10_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_feed_forward1_weight_to_fp16, x = input_539_cast_fp16)[name = string("input_541_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2042962]> module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132798784))), indices_nonzero_data = tensor<uint6, [2042962]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133323136))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134855424))))[name = string("module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_91_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_541_cast_fp16)[name = string("linear_91_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_545_cast_fp16 = silu(x = linear_91_cast_fp16)[name = string("input_545_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1990388]> module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134921024))), indices_nonzero_data = tensor<uint6, [1990388]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135445376))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136938240))))[name = string("module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_92_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_545_cast_fp16)[name = string("linear_92_cast_fp16")];
+            fp16 var_1862_to_fp16 = const()[name = string("op_1862_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_1863_cast_fp16 = mul(x = linear_92_cast_fp16, y = var_1862_to_fp16)[name = string("op_1863_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_551_cast_fp16 = add(x = input_539_cast_fp16, y = var_1863_cast_fp16)[name = string("input_551_cast_fp16")];
+            tensor<int32, [1]> query_21_axes_0 = const()[name = string("query_21_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_10_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_10_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136954688)))];
+            tensor<fp16, [1024]> module_layers_10_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_10_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136956800)))];
+            tensor<fp16, [1, 188, 1024]> query_21_cast_fp16 = layer_norm(axes = query_21_axes_0, beta = module_layers_10_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_self_att_weight_to_fp16, x = input_551_cast_fp16)[name = string("query_21_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525519]> module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136958912))), indices_nonzero_data = tensor<uint6, [525519]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137090048))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137484288))))[name = string("module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_93_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_21_cast_fp16)[name = string("linear_93_cast_fp16")];
+            tensor<int32, [4]> var_1879 = const()[name = string("op_1879"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_61_cast_fp16 = reshape(shape = var_1879, x = linear_93_cast_fp16)[name = string("q_61_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523272]> module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137500736))), indices_nonzero_data = tensor<uint6, [523272]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137631872))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138024448))))[name = string("module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_94_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_21_cast_fp16)[name = string("linear_94_cast_fp16")];
+            tensor<int32, [4]> var_1883 = const()[name = string("op_1883"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_41_cast_fp16 = reshape(shape = var_1883, x = linear_94_cast_fp16)[name = string("k_41_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523359]> module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138040896))), indices_nonzero_data = tensor<uint6, [523359]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138172032))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138564672))))[name = string("module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_95_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_21_cast_fp16)[name = string("linear_95_cast_fp16")];
+            tensor<int32, [4]> var_1887 = const()[name = string("op_1887"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_21_cast_fp16 = reshape(shape = var_1887, x = linear_95_cast_fp16)[name = string("v_21_cast_fp16")];
+            tensor<int32, [4]> value_23_perm_0 = const()[name = string("value_23_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_10_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_10_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138581120)))];
+            tensor<fp16, [1, 188, 8, 128]> var_1899_cast_fp16 = add(x = q_61_cast_fp16, y = module_layers_10_self_attn_pos_bias_u_to_fp16)[name = string("op_1899_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_10_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_10_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138583232)))];
+            tensor<fp16, [1, 188, 8, 128]> var_1901_cast_fp16 = add(x = q_61_cast_fp16, y = module_layers_10_self_attn_pos_bias_v_to_fp16)[name = string("op_1901_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_21_perm_0 = const()[name = string("q_with_bias_v_21_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_227_transpose_x_0 = const()[name = string("x_227_transpose_x_0"), val = bool(false)];
+            bool x_227_transpose_y_0 = const()[name = string("x_227_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_1903_to_fp16 = const()[name = string("op_1903_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(327638592)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_21_cast_fp16 = transpose(perm = q_with_bias_v_21_perm_0, x = var_1901_cast_fp16)[name = string("transpose_242")];
+            tensor<fp16, [1, 8, 188, 375]> x_227_cast_fp16 = matmul(transpose_x = x_227_transpose_x_0, transpose_y = x_227_transpose_y_0, x = q_with_bias_v_21_cast_fp16, y = var_1903_to_fp16)[name = string("x_227_cast_fp16")];
+            tensor<int32, [8]> x_229_pad_0 = const()[name = string("x_229_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_229_mode_0 = const()[name = string("x_229_mode_0"), val = string("constant")];
+            fp16 const_114_to_fp16 = const()[name = string("const_114_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_229_cast_fp16 = pad(constant_val = const_114_to_fp16, mode = x_229_mode_0, pad = x_229_pad_0, x = x_227_cast_fp16)[name = string("x_229_cast_fp16")];
+            tensor<int32, [4]> var_1911 = const()[name = string("op_1911"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_231_cast_fp16 = reshape(shape = var_1911, x = x_229_cast_fp16)[name = string("x_231_cast_fp16")];
+            tensor<int32, [4]> var_1915_begin_0 = const()[name = string("op_1915_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1915_end_0 = const()[name = string("op_1915_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_1915_end_mask_0 = const()[name = string("op_1915_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_1915_cast_fp16 = slice_by_index(begin = var_1915_begin_0, end = var_1915_end_0, end_mask = var_1915_end_mask_0, x = x_231_cast_fp16)[name = string("op_1915_cast_fp16")];
+            tensor<int32, [4]> var_1916 = const()[name = string("op_1916"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_41_cast_fp16 = reshape(shape = var_1916, x = var_1915_cast_fp16)[name = string("matrix_bd_41_cast_fp16")];
+            bool matrix_ac_21_transpose_x_0 = const()[name = string("matrix_ac_21_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_21_transpose_y_0 = const()[name = string("matrix_ac_21_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_116_perm_0 = const()[name = string("transpose_116_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_117_perm_0 = const()[name = string("transpose_117_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_117 = transpose(perm = transpose_117_perm_0, x = k_41_cast_fp16)[name = string("transpose_240")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_116 = transpose(perm = transpose_116_perm_0, x = var_1899_cast_fp16)[name = string("transpose_241")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_21_cast_fp16 = matmul(transpose_x = matrix_ac_21_transpose_x_0, transpose_y = matrix_ac_21_transpose_y_0, x = transpose_116, y = transpose_117)[name = string("matrix_ac_21_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_43_begin_0 = const()[name = string("matrix_bd_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_43_end_0 = const()[name = string("matrix_bd_43_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_43_end_mask_0 = const()[name = string("matrix_bd_43_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_43_cast_fp16 = slice_by_index(begin = matrix_bd_43_begin_0, end = matrix_bd_43_end_0, end_mask = matrix_bd_43_end_mask_0, x = matrix_bd_41_cast_fp16)[name = string("matrix_bd_43_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_1925_cast_fp16 = add(x = matrix_ac_21_cast_fp16, y = matrix_bd_43_cast_fp16)[name = string("op_1925_cast_fp16")];
+            fp16 _inversed_scores_41_y_0_to_fp16 = const()[name = string("_inversed_scores_41_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_41_cast_fp16 = mul(x = var_1925_cast_fp16, y = _inversed_scores_41_y_0_to_fp16)[name = string("_inversed_scores_41_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_43_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_41_cast_fp16, cond = mask_3)[name = string("scores_43_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_1931_cast_fp16 = softmax(axis = var_30, x = scores_43_cast_fp16)[name = string("op_1931_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_553_cast_fp16 = select(a = var_11_to_fp16, b = var_1931_cast_fp16, cond = mask_3)[name = string("input_553_cast_fp16")];
+            bool x_233_transpose_x_0 = const()[name = string("x_233_transpose_x_0"), val = bool(false)];
+            bool x_233_transpose_y_0 = const()[name = string("x_233_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_23_cast_fp16 = transpose(perm = value_23_perm_0, x = v_21_cast_fp16)[name = string("transpose_239")];
+            tensor<fp16, [1, 8, 188, 128]> x_233_cast_fp16 = matmul(transpose_x = x_233_transpose_x_0, transpose_y = x_233_transpose_y_0, x = input_553_cast_fp16, y = value_23_cast_fp16)[name = string("x_233_cast_fp16")];
+            tensor<int32, [4]> var_1935_perm_0 = const()[name = string("op_1935_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1936 = const()[name = string("op_1936"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_1935_cast_fp16 = transpose(perm = var_1935_perm_0, x = x_233_cast_fp16)[name = string("transpose_238")];
+            tensor<fp16, [1, 188, 1024]> input_555_cast_fp16 = reshape(shape = var_1936, x = var_1935_cast_fp16)[name = string("input_555_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524561]> module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138845504))), indices_nonzero_data = tensor<uint6, [524561]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138976640))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139370176))))[name = string("module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_97_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_555_cast_fp16)[name = string("linear_97_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_559_cast_fp16 = add(x = input_551_cast_fp16, y = linear_97_cast_fp16)[name = string("input_559_cast_fp16")];
+            tensor<int32, [1]> x_237_axes_0 = const()[name = string("x_237_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_10_norm_conv_weight_to_fp16 = const()[name = string("module_layers_10_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139386624)))];
+            tensor<fp16, [1024]> module_layers_10_norm_conv_bias_to_fp16 = const()[name = string("module_layers_10_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139388736)))];
+            tensor<fp16, [1, 188, 1024]> x_237_cast_fp16 = layer_norm(axes = x_237_axes_0, beta = module_layers_10_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_conv_weight_to_fp16, x = input_559_cast_fp16)[name = string("x_237_cast_fp16")];
+            tensor<int32, [3]> input_561_perm_0 = const()[name = string("input_561_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_563_pad_type_0 = const()[name = string("input_563_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_563_strides_0 = const()[name = string("input_563_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_563_pad_0 = const()[name = string("input_563_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_563_dilations_0 = const()[name = string("input_563_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_563_groups_0 = const()[name = string("input_563_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1050237]> module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139390848))), indices_nonzero_data = tensor<uint6, [1050237]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139653056))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140440832))))[name = string("module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_561_cast_fp16 = transpose(perm = input_561_perm_0, x = x_237_cast_fp16)[name = string("transpose_237")];
+            tensor<fp16, [1, 2048, 188]> input_563_cast_fp16 = conv(dilations = input_563_dilations_0, groups = input_563_groups_0, pad = input_563_pad_0, pad_type = input_563_pad_type_0, strides = input_563_strides_0, weight = module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_561_cast_fp16)[name = string("input_563_cast_fp16")];
+            int32 x_239_split_num_splits_0 = const()[name = string("x_239_split_num_splits_0"), val = int32(2)];
+            int32 x_239_split_axis_0 = const()[name = string("x_239_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_239_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_239_split_cast_fp16_1 = split(axis = x_239_split_axis_0, num_splits = x_239_split_num_splits_0, x = input_563_cast_fp16)[name = string("x_239_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_239_split_1_sigmoid_cast_fp16 = sigmoid(x = x_239_split_cast_fp16_1)[name = string("x_239_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_239_cast_fp16 = mul(x = x_239_split_cast_fp16_0, y = x_239_split_1_sigmoid_cast_fp16)[name = string("x_239_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_565_cast_fp16 = select(a = var_11_to_fp16, b = x_239_cast_fp16, cond = var_328)[name = string("input_565_cast_fp16")];
+            tensor<int32, [6]> input_567_pad_0 = const()[name = string("input_567_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_567_mode_0 = const()[name = string("input_567_mode_0"), val = string("constant")];
+            fp16 const_117_to_fp16 = const()[name = string("const_117_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_567_cast_fp16 = pad(constant_val = const_117_to_fp16, mode = input_567_mode_0, pad = input_567_pad_0, x = input_565_cast_fp16)[name = string("input_567_cast_fp16")];
+            string input_569_pad_type_0 = const()[name = string("input_569_pad_type_0"), val = string("valid")];
+            int32 input_569_groups_0 = const()[name = string("input_569_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_569_strides_0 = const()[name = string("input_569_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_569_pad_0 = const()[name = string("input_569_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_569_dilations_0 = const()[name = string("input_569_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_268_to_fp16 = const()[name = string("const_268_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140473664)))];
+            tensor<fp16, [1024]> const_269_to_fp16 = const()[name = string("const_269_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140492160)))];
+            tensor<fp16, [1, 1024, 188]> input_571_cast_fp16 = conv(bias = const_269_to_fp16, dilations = input_569_dilations_0, groups = input_569_groups_0, pad = input_569_pad_0, pad_type = input_569_pad_type_0, strides = input_569_strides_0, weight = const_268_to_fp16, x = input_567_cast_fp16)[name = string("input_571_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_573_cast_fp16 = silu(x = input_571_cast_fp16)[name = string("input_573_cast_fp16")];
+            string x_241_pad_type_0 = const()[name = string("x_241_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_241_strides_0 = const()[name = string("x_241_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_241_pad_0 = const()[name = string("x_241_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_241_dilations_0 = const()[name = string("x_241_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_241_groups_0 = const()[name = string("x_241_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520978]> module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140494272))), indices_nonzero_data = tensor<uint6, [520978]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140625408))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141016256))))[name = string("module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_241_cast_fp16 = conv(dilations = x_241_dilations_0, groups = x_241_groups_0, pad = x_241_pad_0, pad_type = x_241_pad_type_0, strides = x_241_strides_0, weight = module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_573_cast_fp16)[name = string("x_241_cast_fp16")];
+            tensor<int32, [3]> input_575_perm_0 = const()[name = string("input_575_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_575_cast_fp16 = transpose(perm = input_575_perm_0, x = x_241_cast_fp16)[name = string("transpose_236")];
+            tensor<fp16, [1, 188, 1024]> input_577_cast_fp16 = add(x = input_559_cast_fp16, y = input_575_cast_fp16)[name = string("input_577_cast_fp16")];
+            tensor<int32, [1]> input_579_axes_0 = const()[name = string("input_579_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_10_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_10_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141032704)))];
+            tensor<fp16, [1024]> module_layers_10_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_10_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141034816)))];
+            tensor<fp16, [1, 188, 1024]> input_579_cast_fp16 = layer_norm(axes = input_579_axes_0, beta = module_layers_10_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_feed_forward2_weight_to_fp16, x = input_577_cast_fp16)[name = string("input_579_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2057233]> module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141036928))), indices_nonzero_data = tensor<uint6, [2057233]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141561280))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143104320))))[name = string("module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_98_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_579_cast_fp16)[name = string("linear_98_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_583_cast_fp16 = silu(x = linear_98_cast_fp16)[name = string("input_583_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2031712]> module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143169920))), indices_nonzero_data = tensor<uint6, [2031712]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143694272))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145218176))))[name = string("module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_99_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_583_cast_fp16)[name = string("linear_99_cast_fp16")];
+            fp16 var_1996_to_fp16 = const()[name = string("op_1996_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_1997_cast_fp16 = mul(x = linear_99_cast_fp16, y = var_1996_to_fp16)[name = string("op_1997_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_589_cast_fp16 = add(x = input_577_cast_fp16, y = var_1997_cast_fp16)[name = string("input_589_cast_fp16")];
+            tensor<int32, [1]> input_591_axes_0 = const()[name = string("input_591_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_10_norm_out_weight_to_fp16 = const()[name = string("module_layers_10_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145234624)))];
+            tensor<fp16, [1024]> module_layers_10_norm_out_bias_to_fp16 = const()[name = string("module_layers_10_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145236736)))];
+            tensor<fp16, [1, 188, 1024]> input_591_cast_fp16 = layer_norm(axes = input_591_axes_0, beta = module_layers_10_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_out_weight_to_fp16, x = input_589_cast_fp16)[name = string("input_591_cast_fp16")];
+            tensor<int32, [1]> input_593_axes_0 = const()[name = string("input_593_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_11_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_11_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145238848)))];
+            tensor<fp16, [1024]> module_layers_11_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_11_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145240960)))];
+            tensor<fp16, [1, 188, 1024]> input_593_cast_fp16 = layer_norm(axes = input_593_axes_0, beta = module_layers_11_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_feed_forward1_weight_to_fp16, x = input_591_cast_fp16)[name = string("input_593_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2039284]> module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145243072))), indices_nonzero_data = tensor<uint6, [2039284]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145767424))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147296960))))[name = string("module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_100_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_593_cast_fp16)[name = string("linear_100_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_597_cast_fp16 = silu(x = linear_100_cast_fp16)[name = string("input_597_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2009766]> module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147362560))), indices_nonzero_data = tensor<uint6, [2009766]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147886912))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149394304))))[name = string("module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_101_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_597_cast_fp16)[name = string("linear_101_cast_fp16")];
+            fp16 var_2025_to_fp16 = const()[name = string("op_2025_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_2026_cast_fp16 = mul(x = linear_101_cast_fp16, y = var_2025_to_fp16)[name = string("op_2026_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_603_cast_fp16 = add(x = input_591_cast_fp16, y = var_2026_cast_fp16)[name = string("input_603_cast_fp16")];
+            tensor<int32, [1]> query_23_axes_0 = const()[name = string("query_23_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_11_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_11_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149410752)))];
+            tensor<fp16, [1024]> module_layers_11_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_11_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149412864)))];
+            tensor<fp16, [1, 188, 1024]> query_23_cast_fp16 = layer_norm(axes = query_23_axes_0, beta = module_layers_11_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_self_att_weight_to_fp16, x = input_603_cast_fp16)[name = string("query_23_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525654]> module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149414976))), indices_nonzero_data = tensor<uint6, [525654]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149546112))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149940480))))[name = string("module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_102_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_23_cast_fp16)[name = string("linear_102_cast_fp16")];
+            tensor<int32, [4]> var_2042 = const()[name = string("op_2042"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_67_cast_fp16 = reshape(shape = var_2042, x = linear_102_cast_fp16)[name = string("q_67_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522850]> module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149956928))), indices_nonzero_data = tensor<uint6, [522850]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150088064))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150480320))))[name = string("module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_103_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_23_cast_fp16)[name = string("linear_103_cast_fp16")];
+            tensor<int32, [4]> var_2046 = const()[name = string("op_2046"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_45_cast_fp16 = reshape(shape = var_2046, x = linear_103_cast_fp16)[name = string("k_45_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522138]> module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150496768))), indices_nonzero_data = tensor<uint6, [522138]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150627904))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151019584))))[name = string("module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_104_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_23_cast_fp16)[name = string("linear_104_cast_fp16")];
+            tensor<int32, [4]> var_2050 = const()[name = string("op_2050"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_23_cast_fp16 = reshape(shape = var_2050, x = linear_104_cast_fp16)[name = string("v_23_cast_fp16")];
+            tensor<int32, [4]> value_25_perm_0 = const()[name = string("value_25_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_11_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_11_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151036032)))];
+            tensor<fp16, [1, 188, 8, 128]> var_2062_cast_fp16 = add(x = q_67_cast_fp16, y = module_layers_11_self_attn_pos_bias_u_to_fp16)[name = string("op_2062_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_11_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_11_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151038144)))];
+            tensor<fp16, [1, 188, 8, 128]> var_2064_cast_fp16 = add(x = q_67_cast_fp16, y = module_layers_11_self_attn_pos_bias_v_to_fp16)[name = string("op_2064_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_23_perm_0 = const()[name = string("q_with_bias_v_23_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_249_transpose_x_0 = const()[name = string("x_249_transpose_x_0"), val = bool(false)];
+            bool x_249_transpose_y_0 = const()[name = string("x_249_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_2066_to_fp16 = const()[name = string("op_2066_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(328406656)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_23_cast_fp16 = transpose(perm = q_with_bias_v_23_perm_0, x = var_2064_cast_fp16)[name = string("transpose_235")];
+            tensor<fp16, [1, 8, 188, 375]> x_249_cast_fp16 = matmul(transpose_x = x_249_transpose_x_0, transpose_y = x_249_transpose_y_0, x = q_with_bias_v_23_cast_fp16, y = var_2066_to_fp16)[name = string("x_249_cast_fp16")];
+            tensor<int32, [8]> x_251_pad_0 = const()[name = string("x_251_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_251_mode_0 = const()[name = string("x_251_mode_0"), val = string("constant")];
+            fp16 const_124_to_fp16 = const()[name = string("const_124_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_251_cast_fp16 = pad(constant_val = const_124_to_fp16, mode = x_251_mode_0, pad = x_251_pad_0, x = x_249_cast_fp16)[name = string("x_251_cast_fp16")];
+            tensor<int32, [4]> var_2074 = const()[name = string("op_2074"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_253_cast_fp16 = reshape(shape = var_2074, x = x_251_cast_fp16)[name = string("x_253_cast_fp16")];
+            tensor<int32, [4]> var_2078_begin_0 = const()[name = string("op_2078_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2078_end_0 = const()[name = string("op_2078_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_2078_end_mask_0 = const()[name = string("op_2078_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_2078_cast_fp16 = slice_by_index(begin = var_2078_begin_0, end = var_2078_end_0, end_mask = var_2078_end_mask_0, x = x_253_cast_fp16)[name = string("op_2078_cast_fp16")];
+            tensor<int32, [4]> var_2079 = const()[name = string("op_2079"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_45_cast_fp16 = reshape(shape = var_2079, x = var_2078_cast_fp16)[name = string("matrix_bd_45_cast_fp16")];
+            bool matrix_ac_23_transpose_x_0 = const()[name = string("matrix_ac_23_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_23_transpose_y_0 = const()[name = string("matrix_ac_23_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_118_perm_0 = const()[name = string("transpose_118_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_119_perm_0 = const()[name = string("transpose_119_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_119 = transpose(perm = transpose_119_perm_0, x = k_45_cast_fp16)[name = string("transpose_233")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_118 = transpose(perm = transpose_118_perm_0, x = var_2062_cast_fp16)[name = string("transpose_234")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_23_cast_fp16 = matmul(transpose_x = matrix_ac_23_transpose_x_0, transpose_y = matrix_ac_23_transpose_y_0, x = transpose_118, y = transpose_119)[name = string("matrix_ac_23_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_47_begin_0 = const()[name = string("matrix_bd_47_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_47_end_0 = const()[name = string("matrix_bd_47_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_47_end_mask_0 = const()[name = string("matrix_bd_47_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_47_cast_fp16 = slice_by_index(begin = matrix_bd_47_begin_0, end = matrix_bd_47_end_0, end_mask = matrix_bd_47_end_mask_0, x = matrix_bd_45_cast_fp16)[name = string("matrix_bd_47_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_2088_cast_fp16 = add(x = matrix_ac_23_cast_fp16, y = matrix_bd_47_cast_fp16)[name = string("op_2088_cast_fp16")];
+            fp16 _inversed_scores_45_y_0_to_fp16 = const()[name = string("_inversed_scores_45_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_45_cast_fp16 = mul(x = var_2088_cast_fp16, y = _inversed_scores_45_y_0_to_fp16)[name = string("_inversed_scores_45_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_47_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_45_cast_fp16, cond = mask_3)[name = string("scores_47_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_2094_cast_fp16 = softmax(axis = var_30, x = scores_47_cast_fp16)[name = string("op_2094_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_605_cast_fp16 = select(a = var_11_to_fp16, b = var_2094_cast_fp16, cond = mask_3)[name = string("input_605_cast_fp16")];
+            bool x_255_transpose_x_0 = const()[name = string("x_255_transpose_x_0"), val = bool(false)];
+            bool x_255_transpose_y_0 = const()[name = string("x_255_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_25_cast_fp16 = transpose(perm = value_25_perm_0, x = v_23_cast_fp16)[name = string("transpose_232")];
+            tensor<fp16, [1, 8, 188, 128]> x_255_cast_fp16 = matmul(transpose_x = x_255_transpose_x_0, transpose_y = x_255_transpose_y_0, x = input_605_cast_fp16, y = value_25_cast_fp16)[name = string("x_255_cast_fp16")];
+            tensor<int32, [4]> var_2098_perm_0 = const()[name = string("op_2098_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2099 = const()[name = string("op_2099"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_2098_cast_fp16 = transpose(perm = var_2098_perm_0, x = x_255_cast_fp16)[name = string("transpose_231")];
+            tensor<fp16, [1, 188, 1024]> input_607_cast_fp16 = reshape(shape = var_2099, x = var_2098_cast_fp16)[name = string("input_607_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524206]> module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151300416))), indices_nonzero_data = tensor<uint6, [524206]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151431552))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151824832))))[name = string("module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_106_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_607_cast_fp16)[name = string("linear_106_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_611_cast_fp16 = add(x = input_603_cast_fp16, y = linear_106_cast_fp16)[name = string("input_611_cast_fp16")];
+            tensor<int32, [1]> x_259_axes_0 = const()[name = string("x_259_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_11_norm_conv_weight_to_fp16 = const()[name = string("module_layers_11_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151841280)))];
+            tensor<fp16, [1024]> module_layers_11_norm_conv_bias_to_fp16 = const()[name = string("module_layers_11_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151843392)))];
+            tensor<fp16, [1, 188, 1024]> x_259_cast_fp16 = layer_norm(axes = x_259_axes_0, beta = module_layers_11_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_conv_weight_to_fp16, x = input_611_cast_fp16)[name = string("x_259_cast_fp16")];
+            tensor<int32, [3]> input_613_perm_0 = const()[name = string("input_613_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_615_pad_type_0 = const()[name = string("input_615_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_615_strides_0 = const()[name = string("input_615_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_615_pad_0 = const()[name = string("input_615_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_615_dilations_0 = const()[name = string("input_615_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_615_groups_0 = const()[name = string("input_615_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1049241]> module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151845504))), indices_nonzero_data = tensor<uint6, [1049241]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152107712))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152894720))))[name = string("module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_613_cast_fp16 = transpose(perm = input_613_perm_0, x = x_259_cast_fp16)[name = string("transpose_230")];
+            tensor<fp16, [1, 2048, 188]> input_615_cast_fp16 = conv(dilations = input_615_dilations_0, groups = input_615_groups_0, pad = input_615_pad_0, pad_type = input_615_pad_type_0, strides = input_615_strides_0, weight = module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_613_cast_fp16)[name = string("input_615_cast_fp16")];
+            int32 x_261_split_num_splits_0 = const()[name = string("x_261_split_num_splits_0"), val = int32(2)];
+            int32 x_261_split_axis_0 = const()[name = string("x_261_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_261_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_261_split_cast_fp16_1 = split(axis = x_261_split_axis_0, num_splits = x_261_split_num_splits_0, x = input_615_cast_fp16)[name = string("x_261_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_261_split_1_sigmoid_cast_fp16 = sigmoid(x = x_261_split_cast_fp16_1)[name = string("x_261_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_261_cast_fp16 = mul(x = x_261_split_cast_fp16_0, y = x_261_split_1_sigmoid_cast_fp16)[name = string("x_261_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_617_cast_fp16 = select(a = var_11_to_fp16, b = x_261_cast_fp16, cond = var_328)[name = string("input_617_cast_fp16")];
+            tensor<int32, [6]> input_619_pad_0 = const()[name = string("input_619_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_619_mode_0 = const()[name = string("input_619_mode_0"), val = string("constant")];
+            fp16 const_127_to_fp16 = const()[name = string("const_127_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_619_cast_fp16 = pad(constant_val = const_127_to_fp16, mode = input_619_mode_0, pad = input_619_pad_0, x = input_617_cast_fp16)[name = string("input_619_cast_fp16")];
+            string input_621_pad_type_0 = const()[name = string("input_621_pad_type_0"), val = string("valid")];
+            int32 input_621_groups_0 = const()[name = string("input_621_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_621_strides_0 = const()[name = string("input_621_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_621_pad_0 = const()[name = string("input_621_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_621_dilations_0 = const()[name = string("input_621_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_270_to_fp16 = const()[name = string("const_270_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152927552)))];
+            tensor<fp16, [1024]> const_271_to_fp16 = const()[name = string("const_271_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152946048)))];
+            tensor<fp16, [1, 1024, 188]> input_623_cast_fp16 = conv(bias = const_271_to_fp16, dilations = input_621_dilations_0, groups = input_621_groups_0, pad = input_621_pad_0, pad_type = input_621_pad_type_0, strides = input_621_strides_0, weight = const_270_to_fp16, x = input_619_cast_fp16)[name = string("input_623_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_625_cast_fp16 = silu(x = input_623_cast_fp16)[name = string("input_625_cast_fp16")];
+            string x_263_pad_type_0 = const()[name = string("x_263_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_263_strides_0 = const()[name = string("x_263_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_263_pad_0 = const()[name = string("x_263_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_263_dilations_0 = const()[name = string("x_263_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_263_groups_0 = const()[name = string("x_263_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522671]> module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152948160))), indices_nonzero_data = tensor<uint6, [522671]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153079296))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153471424))))[name = string("module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_263_cast_fp16 = conv(dilations = x_263_dilations_0, groups = x_263_groups_0, pad = x_263_pad_0, pad_type = x_263_pad_type_0, strides = x_263_strides_0, weight = module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_625_cast_fp16)[name = string("x_263_cast_fp16")];
+            tensor<int32, [3]> input_627_perm_0 = const()[name = string("input_627_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_627_cast_fp16 = transpose(perm = input_627_perm_0, x = x_263_cast_fp16)[name = string("transpose_229")];
+            tensor<fp16, [1, 188, 1024]> input_629_cast_fp16 = add(x = input_611_cast_fp16, y = input_627_cast_fp16)[name = string("input_629_cast_fp16")];
+            tensor<int32, [1]> input_631_axes_0 = const()[name = string("input_631_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_11_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_11_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153487872)))];
+            tensor<fp16, [1024]> module_layers_11_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_11_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153489984)))];
+            tensor<fp16, [1, 188, 1024]> input_631_cast_fp16 = layer_norm(axes = input_631_axes_0, beta = module_layers_11_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_feed_forward2_weight_to_fp16, x = input_629_cast_fp16)[name = string("input_631_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2065473]> module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153492096))), indices_nonzero_data = tensor<uint6, [2065473]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154016448))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155565632))))[name = string("module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_107_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_631_cast_fp16)[name = string("linear_107_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_635_cast_fp16 = silu(x = linear_107_cast_fp16)[name = string("input_635_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1993236]> module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155631232))), indices_nonzero_data = tensor<uint6, [1993236]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156155584))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157650624))))[name = string("module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_108_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_635_cast_fp16)[name = string("linear_108_cast_fp16")];
+            fp16 var_2159_to_fp16 = const()[name = string("op_2159_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_2160_cast_fp16 = mul(x = linear_108_cast_fp16, y = var_2159_to_fp16)[name = string("op_2160_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_641_cast_fp16 = add(x = input_629_cast_fp16, y = var_2160_cast_fp16)[name = string("input_641_cast_fp16")];
+            tensor<int32, [1]> input_643_axes_0 = const()[name = string("input_643_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_11_norm_out_weight_to_fp16 = const()[name = string("module_layers_11_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157667072)))];
+            tensor<fp16, [1024]> module_layers_11_norm_out_bias_to_fp16 = const()[name = string("module_layers_11_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157669184)))];
+            tensor<fp16, [1, 188, 1024]> input_643_cast_fp16 = layer_norm(axes = input_643_axes_0, beta = module_layers_11_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_out_weight_to_fp16, x = input_641_cast_fp16)[name = string("input_643_cast_fp16")];
+            tensor<int32, [1]> input_645_axes_0 = const()[name = string("input_645_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_12_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_12_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157671296)))];
+            tensor<fp16, [1024]> module_layers_12_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_12_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157673408)))];
+            tensor<fp16, [1, 188, 1024]> input_645_cast_fp16 = layer_norm(axes = input_645_axes_0, beta = module_layers_12_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_feed_forward1_weight_to_fp16, x = input_643_cast_fp16)[name = string("input_645_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2047312]> module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157675520))), indices_nonzero_data = tensor<uint6, [2047312]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158199872))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159735424))))[name = string("module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_109_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_645_cast_fp16)[name = string("linear_109_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_649_cast_fp16 = silu(x = linear_109_cast_fp16)[name = string("input_649_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1984634]> module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159801024))), indices_nonzero_data = tensor<uint6, [1984634]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160325376))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161813952))))[name = string("module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_110_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_649_cast_fp16)[name = string("linear_110_cast_fp16")];
+            fp16 var_2188_to_fp16 = const()[name = string("op_2188_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_2189_cast_fp16 = mul(x = linear_110_cast_fp16, y = var_2188_to_fp16)[name = string("op_2189_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_655_cast_fp16 = add(x = input_643_cast_fp16, y = var_2189_cast_fp16)[name = string("input_655_cast_fp16")];
+            tensor<int32, [1]> query_25_axes_0 = const()[name = string("query_25_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_12_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_12_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161830400)))];
+            tensor<fp16, [1024]> module_layers_12_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_12_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161832512)))];
+            tensor<fp16, [1, 188, 1024]> query_25_cast_fp16 = layer_norm(axes = query_25_axes_0, beta = module_layers_12_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_self_att_weight_to_fp16, x = input_655_cast_fp16)[name = string("query_25_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525969]> module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161834624))), indices_nonzero_data = tensor<uint6, [525969]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161965760))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162360320))))[name = string("module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_111_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_25_cast_fp16)[name = string("linear_111_cast_fp16")];
+            tensor<int32, [4]> var_2205 = const()[name = string("op_2205"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_73_cast_fp16 = reshape(shape = var_2205, x = linear_111_cast_fp16)[name = string("q_73_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522730]> module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162376768))), indices_nonzero_data = tensor<uint6, [522730]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162507904))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162900032))))[name = string("module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_112_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_25_cast_fp16)[name = string("linear_112_cast_fp16")];
+            tensor<int32, [4]> var_2209 = const()[name = string("op_2209"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_49_cast_fp16 = reshape(shape = var_2209, x = linear_112_cast_fp16)[name = string("k_49_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521113]> module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162916480))), indices_nonzero_data = tensor<uint6, [521113]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163047616))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163438528))))[name = string("module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_113_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_25_cast_fp16)[name = string("linear_113_cast_fp16")];
+            tensor<int32, [4]> var_2213 = const()[name = string("op_2213"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_25_cast_fp16 = reshape(shape = var_2213, x = linear_113_cast_fp16)[name = string("v_25_cast_fp16")];
+            tensor<int32, [4]> value_27_perm_0 = const()[name = string("value_27_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_12_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_12_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163454976)))];
+            tensor<fp16, [1, 188, 8, 128]> var_2225_cast_fp16 = add(x = q_73_cast_fp16, y = module_layers_12_self_attn_pos_bias_u_to_fp16)[name = string("op_2225_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_12_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_12_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163457088)))];
+            tensor<fp16, [1, 188, 8, 128]> var_2227_cast_fp16 = add(x = q_73_cast_fp16, y = module_layers_12_self_attn_pos_bias_v_to_fp16)[name = string("op_2227_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_25_perm_0 = const()[name = string("q_with_bias_v_25_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_271_transpose_x_0 = const()[name = string("x_271_transpose_x_0"), val = bool(false)];
+            bool x_271_transpose_y_0 = const()[name = string("x_271_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_2229_to_fp16 = const()[name = string("op_2229_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329174720)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_25_cast_fp16 = transpose(perm = q_with_bias_v_25_perm_0, x = var_2227_cast_fp16)[name = string("transpose_228")];
+            tensor<fp16, [1, 8, 188, 375]> x_271_cast_fp16 = matmul(transpose_x = x_271_transpose_x_0, transpose_y = x_271_transpose_y_0, x = q_with_bias_v_25_cast_fp16, y = var_2229_to_fp16)[name = string("x_271_cast_fp16")];
+            tensor<int32, [8]> x_273_pad_0 = const()[name = string("x_273_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_273_mode_0 = const()[name = string("x_273_mode_0"), val = string("constant")];
+            fp16 const_134_to_fp16 = const()[name = string("const_134_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_273_cast_fp16 = pad(constant_val = const_134_to_fp16, mode = x_273_mode_0, pad = x_273_pad_0, x = x_271_cast_fp16)[name = string("x_273_cast_fp16")];
+            tensor<int32, [4]> var_2237 = const()[name = string("op_2237"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_275_cast_fp16 = reshape(shape = var_2237, x = x_273_cast_fp16)[name = string("x_275_cast_fp16")];
+            tensor<int32, [4]> var_2241_begin_0 = const()[name = string("op_2241_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2241_end_0 = const()[name = string("op_2241_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_2241_end_mask_0 = const()[name = string("op_2241_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_2241_cast_fp16 = slice_by_index(begin = var_2241_begin_0, end = var_2241_end_0, end_mask = var_2241_end_mask_0, x = x_275_cast_fp16)[name = string("op_2241_cast_fp16")];
+            tensor<int32, [4]> var_2242 = const()[name = string("op_2242"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_49_cast_fp16 = reshape(shape = var_2242, x = var_2241_cast_fp16)[name = string("matrix_bd_49_cast_fp16")];
+            bool matrix_ac_25_transpose_x_0 = const()[name = string("matrix_ac_25_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_25_transpose_y_0 = const()[name = string("matrix_ac_25_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_120_perm_0 = const()[name = string("transpose_120_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_121_perm_0 = const()[name = string("transpose_121_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_121 = transpose(perm = transpose_121_perm_0, x = k_49_cast_fp16)[name = string("transpose_226")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_120 = transpose(perm = transpose_120_perm_0, x = var_2225_cast_fp16)[name = string("transpose_227")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_25_cast_fp16 = matmul(transpose_x = matrix_ac_25_transpose_x_0, transpose_y = matrix_ac_25_transpose_y_0, x = transpose_120, y = transpose_121)[name = string("matrix_ac_25_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_51_begin_0 = const()[name = string("matrix_bd_51_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_51_end_0 = const()[name = string("matrix_bd_51_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_51_end_mask_0 = const()[name = string("matrix_bd_51_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_51_cast_fp16 = slice_by_index(begin = matrix_bd_51_begin_0, end = matrix_bd_51_end_0, end_mask = matrix_bd_51_end_mask_0, x = matrix_bd_49_cast_fp16)[name = string("matrix_bd_51_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_2251_cast_fp16 = add(x = matrix_ac_25_cast_fp16, y = matrix_bd_51_cast_fp16)[name = string("op_2251_cast_fp16")];
+            fp16 _inversed_scores_49_y_0_to_fp16 = const()[name = string("_inversed_scores_49_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_49_cast_fp16 = mul(x = var_2251_cast_fp16, y = _inversed_scores_49_y_0_to_fp16)[name = string("_inversed_scores_49_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_51_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_49_cast_fp16, cond = mask_3)[name = string("scores_51_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_2257_cast_fp16 = softmax(axis = var_30, x = scores_51_cast_fp16)[name = string("op_2257_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_657_cast_fp16 = select(a = var_11_to_fp16, b = var_2257_cast_fp16, cond = mask_3)[name = string("input_657_cast_fp16")];
+            bool x_277_transpose_x_0 = const()[name = string("x_277_transpose_x_0"), val = bool(false)];
+            bool x_277_transpose_y_0 = const()[name = string("x_277_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_27_cast_fp16 = transpose(perm = value_27_perm_0, x = v_25_cast_fp16)[name = string("transpose_225")];
+            tensor<fp16, [1, 8, 188, 128]> x_277_cast_fp16 = matmul(transpose_x = x_277_transpose_x_0, transpose_y = x_277_transpose_y_0, x = input_657_cast_fp16, y = value_27_cast_fp16)[name = string("x_277_cast_fp16")];
+            tensor<int32, [4]> var_2261_perm_0 = const()[name = string("op_2261_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2262 = const()[name = string("op_2262"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_2261_cast_fp16 = transpose(perm = var_2261_perm_0, x = x_277_cast_fp16)[name = string("transpose_224")];
+            tensor<fp16, [1, 188, 1024]> input_659_cast_fp16 = reshape(shape = var_2262, x = var_2261_cast_fp16)[name = string("input_659_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [527263]> module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163719360))), indices_nonzero_data = tensor<uint6, [527263]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163850496))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164246016))))[name = string("module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_115_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_659_cast_fp16)[name = string("linear_115_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_663_cast_fp16 = add(x = input_655_cast_fp16, y = linear_115_cast_fp16)[name = string("input_663_cast_fp16")];
+            tensor<int32, [1]> x_281_axes_0 = const()[name = string("x_281_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_12_norm_conv_weight_to_fp16 = const()[name = string("module_layers_12_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164262464)))];
+            tensor<fp16, [1024]> module_layers_12_norm_conv_bias_to_fp16 = const()[name = string("module_layers_12_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164264576)))];
+            tensor<fp16, [1, 188, 1024]> x_281_cast_fp16 = layer_norm(axes = x_281_axes_0, beta = module_layers_12_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_conv_weight_to_fp16, x = input_663_cast_fp16)[name = string("x_281_cast_fp16")];
+            tensor<int32, [3]> input_665_perm_0 = const()[name = string("input_665_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_667_pad_type_0 = const()[name = string("input_667_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_667_strides_0 = const()[name = string("input_667_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_667_pad_0 = const()[name = string("input_667_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_667_dilations_0 = const()[name = string("input_667_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_667_groups_0 = const()[name = string("input_667_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1044574]> module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164266688))), indices_nonzero_data = tensor<uint6, [1044574]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164528896))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165312448))))[name = string("module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_665_cast_fp16 = transpose(perm = input_665_perm_0, x = x_281_cast_fp16)[name = string("transpose_223")];
+            tensor<fp16, [1, 2048, 188]> input_667_cast_fp16 = conv(dilations = input_667_dilations_0, groups = input_667_groups_0, pad = input_667_pad_0, pad_type = input_667_pad_type_0, strides = input_667_strides_0, weight = module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_665_cast_fp16)[name = string("input_667_cast_fp16")];
+            int32 x_283_split_num_splits_0 = const()[name = string("x_283_split_num_splits_0"), val = int32(2)];
+            int32 x_283_split_axis_0 = const()[name = string("x_283_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_283_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_283_split_cast_fp16_1 = split(axis = x_283_split_axis_0, num_splits = x_283_split_num_splits_0, x = input_667_cast_fp16)[name = string("x_283_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_283_split_1_sigmoid_cast_fp16 = sigmoid(x = x_283_split_cast_fp16_1)[name = string("x_283_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_283_cast_fp16 = mul(x = x_283_split_cast_fp16_0, y = x_283_split_1_sigmoid_cast_fp16)[name = string("x_283_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_669_cast_fp16 = select(a = var_11_to_fp16, b = x_283_cast_fp16, cond = var_328)[name = string("input_669_cast_fp16")];
+            tensor<int32, [6]> input_671_pad_0 = const()[name = string("input_671_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_671_mode_0 = const()[name = string("input_671_mode_0"), val = string("constant")];
+            fp16 const_137_to_fp16 = const()[name = string("const_137_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_671_cast_fp16 = pad(constant_val = const_137_to_fp16, mode = input_671_mode_0, pad = input_671_pad_0, x = input_669_cast_fp16)[name = string("input_671_cast_fp16")];
+            string input_673_pad_type_0 = const()[name = string("input_673_pad_type_0"), val = string("valid")];
+            int32 input_673_groups_0 = const()[name = string("input_673_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_673_strides_0 = const()[name = string("input_673_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_673_pad_0 = const()[name = string("input_673_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_673_dilations_0 = const()[name = string("input_673_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_272_to_fp16 = const()[name = string("const_272_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165345280)))];
+            tensor<fp16, [1024]> const_273_to_fp16 = const()[name = string("const_273_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165363776)))];
+            tensor<fp16, [1, 1024, 188]> input_675_cast_fp16 = conv(bias = const_273_to_fp16, dilations = input_673_dilations_0, groups = input_673_groups_0, pad = input_673_pad_0, pad_type = input_673_pad_type_0, strides = input_673_strides_0, weight = const_272_to_fp16, x = input_671_cast_fp16)[name = string("input_675_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_677_cast_fp16 = silu(x = input_675_cast_fp16)[name = string("input_677_cast_fp16")];
+            string x_285_pad_type_0 = const()[name = string("x_285_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_285_strides_0 = const()[name = string("x_285_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_285_pad_0 = const()[name = string("x_285_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_285_dilations_0 = const()[name = string("x_285_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_285_groups_0 = const()[name = string("x_285_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522908]> module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165365888))), indices_nonzero_data = tensor<uint6, [522908]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165497024))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165889280))))[name = string("module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_285_cast_fp16 = conv(dilations = x_285_dilations_0, groups = x_285_groups_0, pad = x_285_pad_0, pad_type = x_285_pad_type_0, strides = x_285_strides_0, weight = module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_677_cast_fp16)[name = string("x_285_cast_fp16")];
+            tensor<int32, [3]> input_679_perm_0 = const()[name = string("input_679_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_679_cast_fp16 = transpose(perm = input_679_perm_0, x = x_285_cast_fp16)[name = string("transpose_222")];
+            tensor<fp16, [1, 188, 1024]> input_681_cast_fp16 = add(x = input_663_cast_fp16, y = input_679_cast_fp16)[name = string("input_681_cast_fp16")];
+            tensor<int32, [1]> input_683_axes_0 = const()[name = string("input_683_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_12_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_12_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165905728)))];
+            tensor<fp16, [1024]> module_layers_12_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_12_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165907840)))];
+            tensor<fp16, [1, 188, 1024]> input_683_cast_fp16 = layer_norm(axes = input_683_axes_0, beta = module_layers_12_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_feed_forward2_weight_to_fp16, x = input_681_cast_fp16)[name = string("input_683_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2072965]> module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165909952))), indices_nonzero_data = tensor<uint6, [2072965]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166434304))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167989120))))[name = string("module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_116_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_683_cast_fp16)[name = string("linear_116_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_687_cast_fp16 = silu(x = linear_116_cast_fp16)[name = string("input_687_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1985707]> module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168054720))), indices_nonzero_data = tensor<uint6, [1985707]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168579072))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170068480))))[name = string("module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_117_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_687_cast_fp16)[name = string("linear_117_cast_fp16")];
+            fp16 var_2322_to_fp16 = const()[name = string("op_2322_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_2323_cast_fp16 = mul(x = linear_117_cast_fp16, y = var_2322_to_fp16)[name = string("op_2323_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_693_cast_fp16 = add(x = input_681_cast_fp16, y = var_2323_cast_fp16)[name = string("input_693_cast_fp16")];
+            tensor<int32, [1]> input_695_axes_0 = const()[name = string("input_695_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_12_norm_out_weight_to_fp16 = const()[name = string("module_layers_12_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170084928)))];
+            tensor<fp16, [1024]> module_layers_12_norm_out_bias_to_fp16 = const()[name = string("module_layers_12_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170087040)))];
+            tensor<fp16, [1, 188, 1024]> input_695_cast_fp16 = layer_norm(axes = input_695_axes_0, beta = module_layers_12_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_out_weight_to_fp16, x = input_693_cast_fp16)[name = string("input_695_cast_fp16")];
+            tensor<int32, [1]> input_697_axes_0 = const()[name = string("input_697_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_13_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_13_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170089152)))];
+            tensor<fp16, [1024]> module_layers_13_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_13_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170091264)))];
+            tensor<fp16, [1, 188, 1024]> input_697_cast_fp16 = layer_norm(axes = input_697_axes_0, beta = module_layers_13_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_feed_forward1_weight_to_fp16, x = input_695_cast_fp16)[name = string("input_697_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2052653]> module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170093376))), indices_nonzero_data = tensor<uint6, [2052653]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170617728))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172157312))))[name = string("module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_118_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_697_cast_fp16)[name = string("linear_118_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_701_cast_fp16 = silu(x = linear_118_cast_fp16)[name = string("input_701_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1946459]> module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172222912))), indices_nonzero_data = tensor<uint6, [1946459]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172747264))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174207232))))[name = string("module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_119_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_701_cast_fp16)[name = string("linear_119_cast_fp16")];
+            fp16 var_2351_to_fp16 = const()[name = string("op_2351_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_2352_cast_fp16 = mul(x = linear_119_cast_fp16, y = var_2351_to_fp16)[name = string("op_2352_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_707_cast_fp16 = add(x = input_695_cast_fp16, y = var_2352_cast_fp16)[name = string("input_707_cast_fp16")];
+            tensor<int32, [1]> query_27_axes_0 = const()[name = string("query_27_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_13_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_13_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174223680)))];
+            tensor<fp16, [1024]> module_layers_13_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_13_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174225792)))];
+            tensor<fp16, [1, 188, 1024]> query_27_cast_fp16 = layer_norm(axes = query_27_axes_0, beta = module_layers_13_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_self_att_weight_to_fp16, x = input_707_cast_fp16)[name = string("query_27_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524122]> module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174227904))), indices_nonzero_data = tensor<uint6, [524122]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174359040))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174752256))))[name = string("module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_120_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_27_cast_fp16)[name = string("linear_120_cast_fp16")];
+            tensor<int32, [4]> var_2368 = const()[name = string("op_2368"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_79_cast_fp16 = reshape(shape = var_2368, x = linear_120_cast_fp16)[name = string("q_79_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521808]> module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174768704))), indices_nonzero_data = tensor<uint6, [521808]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174899840))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175291264))))[name = string("module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_121_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_27_cast_fp16)[name = string("linear_121_cast_fp16")];
+            tensor<int32, [4]> var_2372 = const()[name = string("op_2372"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_53_cast_fp16 = reshape(shape = var_2372, x = linear_121_cast_fp16)[name = string("k_53_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521835]> module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175307712))), indices_nonzero_data = tensor<uint6, [521835]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175438848))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175830336))))[name = string("module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_122_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_27_cast_fp16)[name = string("linear_122_cast_fp16")];
+            tensor<int32, [4]> var_2376 = const()[name = string("op_2376"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_27_cast_fp16 = reshape(shape = var_2376, x = linear_122_cast_fp16)[name = string("v_27_cast_fp16")];
+            tensor<int32, [4]> value_29_perm_0 = const()[name = string("value_29_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_13_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_13_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175846784)))];
+            tensor<fp16, [1, 188, 8, 128]> var_2388_cast_fp16 = add(x = q_79_cast_fp16, y = module_layers_13_self_attn_pos_bias_u_to_fp16)[name = string("op_2388_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_13_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_13_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175848896)))];
+            tensor<fp16, [1, 188, 8, 128]> var_2390_cast_fp16 = add(x = q_79_cast_fp16, y = module_layers_13_self_attn_pos_bias_v_to_fp16)[name = string("op_2390_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_27_perm_0 = const()[name = string("q_with_bias_v_27_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_293_transpose_x_0 = const()[name = string("x_293_transpose_x_0"), val = bool(false)];
+            bool x_293_transpose_y_0 = const()[name = string("x_293_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_2392_to_fp16 = const()[name = string("op_2392_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329942784)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_27_cast_fp16 = transpose(perm = q_with_bias_v_27_perm_0, x = var_2390_cast_fp16)[name = string("transpose_221")];
+            tensor<fp16, [1, 8, 188, 375]> x_293_cast_fp16 = matmul(transpose_x = x_293_transpose_x_0, transpose_y = x_293_transpose_y_0, x = q_with_bias_v_27_cast_fp16, y = var_2392_to_fp16)[name = string("x_293_cast_fp16")];
+            tensor<int32, [8]> x_295_pad_0 = const()[name = string("x_295_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_295_mode_0 = const()[name = string("x_295_mode_0"), val = string("constant")];
+            fp16 const_144_to_fp16 = const()[name = string("const_144_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_295_cast_fp16 = pad(constant_val = const_144_to_fp16, mode = x_295_mode_0, pad = x_295_pad_0, x = x_293_cast_fp16)[name = string("x_295_cast_fp16")];
+            tensor<int32, [4]> var_2400 = const()[name = string("op_2400"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_297_cast_fp16 = reshape(shape = var_2400, x = x_295_cast_fp16)[name = string("x_297_cast_fp16")];
+            tensor<int32, [4]> var_2404_begin_0 = const()[name = string("op_2404_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2404_end_0 = const()[name = string("op_2404_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_2404_end_mask_0 = const()[name = string("op_2404_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_2404_cast_fp16 = slice_by_index(begin = var_2404_begin_0, end = var_2404_end_0, end_mask = var_2404_end_mask_0, x = x_297_cast_fp16)[name = string("op_2404_cast_fp16")];
+            tensor<int32, [4]> var_2405 = const()[name = string("op_2405"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_53_cast_fp16 = reshape(shape = var_2405, x = var_2404_cast_fp16)[name = string("matrix_bd_53_cast_fp16")];
+            bool matrix_ac_27_transpose_x_0 = const()[name = string("matrix_ac_27_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_27_transpose_y_0 = const()[name = string("matrix_ac_27_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_122_perm_0 = const()[name = string("transpose_122_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_123_perm_0 = const()[name = string("transpose_123_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_123 = transpose(perm = transpose_123_perm_0, x = k_53_cast_fp16)[name = string("transpose_219")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_122 = transpose(perm = transpose_122_perm_0, x = var_2388_cast_fp16)[name = string("transpose_220")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_27_cast_fp16 = matmul(transpose_x = matrix_ac_27_transpose_x_0, transpose_y = matrix_ac_27_transpose_y_0, x = transpose_122, y = transpose_123)[name = string("matrix_ac_27_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_55_begin_0 = const()[name = string("matrix_bd_55_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_55_end_0 = const()[name = string("matrix_bd_55_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_55_end_mask_0 = const()[name = string("matrix_bd_55_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_55_cast_fp16 = slice_by_index(begin = matrix_bd_55_begin_0, end = matrix_bd_55_end_0, end_mask = matrix_bd_55_end_mask_0, x = matrix_bd_53_cast_fp16)[name = string("matrix_bd_55_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_2414_cast_fp16 = add(x = matrix_ac_27_cast_fp16, y = matrix_bd_55_cast_fp16)[name = string("op_2414_cast_fp16")];
+            fp16 _inversed_scores_53_y_0_to_fp16 = const()[name = string("_inversed_scores_53_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_53_cast_fp16 = mul(x = var_2414_cast_fp16, y = _inversed_scores_53_y_0_to_fp16)[name = string("_inversed_scores_53_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_55_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_53_cast_fp16, cond = mask_3)[name = string("scores_55_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_2420_cast_fp16 = softmax(axis = var_30, x = scores_55_cast_fp16)[name = string("op_2420_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_709_cast_fp16 = select(a = var_11_to_fp16, b = var_2420_cast_fp16, cond = mask_3)[name = string("input_709_cast_fp16")];
+            bool x_299_transpose_x_0 = const()[name = string("x_299_transpose_x_0"), val = bool(false)];
+            bool x_299_transpose_y_0 = const()[name = string("x_299_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_29_cast_fp16 = transpose(perm = value_29_perm_0, x = v_27_cast_fp16)[name = string("transpose_218")];
+            tensor<fp16, [1, 8, 188, 128]> x_299_cast_fp16 = matmul(transpose_x = x_299_transpose_x_0, transpose_y = x_299_transpose_y_0, x = input_709_cast_fp16, y = value_29_cast_fp16)[name = string("x_299_cast_fp16")];
+            tensor<int32, [4]> var_2424_perm_0 = const()[name = string("op_2424_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2425 = const()[name = string("op_2425"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_2424_cast_fp16 = transpose(perm = var_2424_perm_0, x = x_299_cast_fp16)[name = string("transpose_217")];
+            tensor<fp16, [1, 188, 1024]> input_711_cast_fp16 = reshape(shape = var_2425, x = var_2424_cast_fp16)[name = string("input_711_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [528285]> module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176111168))), indices_nonzero_data = tensor<uint6, [528285]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176242304))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176638592))))[name = string("module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_124_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_711_cast_fp16)[name = string("linear_124_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_715_cast_fp16 = add(x = input_707_cast_fp16, y = linear_124_cast_fp16)[name = string("input_715_cast_fp16")];
+            tensor<int32, [1]> x_303_axes_0 = const()[name = string("x_303_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_13_norm_conv_weight_to_fp16 = const()[name = string("module_layers_13_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176655040)))];
+            tensor<fp16, [1024]> module_layers_13_norm_conv_bias_to_fp16 = const()[name = string("module_layers_13_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176657152)))];
+            tensor<fp16, [1, 188, 1024]> x_303_cast_fp16 = layer_norm(axes = x_303_axes_0, beta = module_layers_13_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_conv_weight_to_fp16, x = input_715_cast_fp16)[name = string("x_303_cast_fp16")];
+            tensor<int32, [3]> input_717_perm_0 = const()[name = string("input_717_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_719_pad_type_0 = const()[name = string("input_719_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_719_strides_0 = const()[name = string("input_719_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_719_pad_0 = const()[name = string("input_719_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_719_dilations_0 = const()[name = string("input_719_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_719_groups_0 = const()[name = string("input_719_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1045290]> module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176659264))), indices_nonzero_data = tensor<uint6, [1045290]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176921472))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177705536))))[name = string("module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_717_cast_fp16 = transpose(perm = input_717_perm_0, x = x_303_cast_fp16)[name = string("transpose_216")];
+            tensor<fp16, [1, 2048, 188]> input_719_cast_fp16 = conv(dilations = input_719_dilations_0, groups = input_719_groups_0, pad = input_719_pad_0, pad_type = input_719_pad_type_0, strides = input_719_strides_0, weight = module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_717_cast_fp16)[name = string("input_719_cast_fp16")];
+            int32 x_305_split_num_splits_0 = const()[name = string("x_305_split_num_splits_0"), val = int32(2)];
+            int32 x_305_split_axis_0 = const()[name = string("x_305_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_305_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_305_split_cast_fp16_1 = split(axis = x_305_split_axis_0, num_splits = x_305_split_num_splits_0, x = input_719_cast_fp16)[name = string("x_305_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_305_split_1_sigmoid_cast_fp16 = sigmoid(x = x_305_split_cast_fp16_1)[name = string("x_305_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_305_cast_fp16 = mul(x = x_305_split_cast_fp16_0, y = x_305_split_1_sigmoid_cast_fp16)[name = string("x_305_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_721_cast_fp16 = select(a = var_11_to_fp16, b = x_305_cast_fp16, cond = var_328)[name = string("input_721_cast_fp16")];
+            tensor<int32, [6]> input_723_pad_0 = const()[name = string("input_723_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_723_mode_0 = const()[name = string("input_723_mode_0"), val = string("constant")];
+            fp16 const_147_to_fp16 = const()[name = string("const_147_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_723_cast_fp16 = pad(constant_val = const_147_to_fp16, mode = input_723_mode_0, pad = input_723_pad_0, x = input_721_cast_fp16)[name = string("input_723_cast_fp16")];
+            string input_725_pad_type_0 = const()[name = string("input_725_pad_type_0"), val = string("valid")];
+            int32 input_725_groups_0 = const()[name = string("input_725_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_725_strides_0 = const()[name = string("input_725_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_725_pad_0 = const()[name = string("input_725_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_725_dilations_0 = const()[name = string("input_725_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_274_to_fp16 = const()[name = string("const_274_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177738368)))];
+            tensor<fp16, [1024]> const_275_to_fp16 = const()[name = string("const_275_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177756864)))];
+            tensor<fp16, [1, 1024, 188]> input_727_cast_fp16 = conv(bias = const_275_to_fp16, dilations = input_725_dilations_0, groups = input_725_groups_0, pad = input_725_pad_0, pad_type = input_725_pad_type_0, strides = input_725_strides_0, weight = const_274_to_fp16, x = input_723_cast_fp16)[name = string("input_727_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_729_cast_fp16 = silu(x = input_727_cast_fp16)[name = string("input_729_cast_fp16")];
+            string x_307_pad_type_0 = const()[name = string("x_307_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_307_strides_0 = const()[name = string("x_307_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_307_pad_0 = const()[name = string("x_307_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_307_dilations_0 = const()[name = string("x_307_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_307_groups_0 = const()[name = string("x_307_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524845]> module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177758976))), indices_nonzero_data = tensor<uint6, [524845]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177890112))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178283840))))[name = string("module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_307_cast_fp16 = conv(dilations = x_307_dilations_0, groups = x_307_groups_0, pad = x_307_pad_0, pad_type = x_307_pad_type_0, strides = x_307_strides_0, weight = module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_729_cast_fp16)[name = string("x_307_cast_fp16")];
+            tensor<int32, [3]> input_731_perm_0 = const()[name = string("input_731_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_731_cast_fp16 = transpose(perm = input_731_perm_0, x = x_307_cast_fp16)[name = string("transpose_215")];
+            tensor<fp16, [1, 188, 1024]> input_733_cast_fp16 = add(x = input_715_cast_fp16, y = input_731_cast_fp16)[name = string("input_733_cast_fp16")];
+            tensor<int32, [1]> input_735_axes_0 = const()[name = string("input_735_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_13_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_13_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178300288)))];
+            tensor<fp16, [1024]> module_layers_13_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_13_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178302400)))];
+            tensor<fp16, [1, 188, 1024]> input_735_cast_fp16 = layer_norm(axes = input_735_axes_0, beta = module_layers_13_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_feed_forward2_weight_to_fp16, x = input_733_cast_fp16)[name = string("input_735_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2074364]> module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178304512))), indices_nonzero_data = tensor<uint6, [2074364]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178828864))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180384704))))[name = string("module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_125_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_735_cast_fp16)[name = string("linear_125_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_739_cast_fp16 = silu(x = linear_125_cast_fp16)[name = string("input_739_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1986819]> module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180450304))), indices_nonzero_data = tensor<uint6, [1986819]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180974656))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182464896))))[name = string("module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_126_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_739_cast_fp16)[name = string("linear_126_cast_fp16")];
+            fp16 var_2485_to_fp16 = const()[name = string("op_2485_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_2486_cast_fp16 = mul(x = linear_126_cast_fp16, y = var_2485_to_fp16)[name = string("op_2486_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_745_cast_fp16 = add(x = input_733_cast_fp16, y = var_2486_cast_fp16)[name = string("input_745_cast_fp16")];
+            tensor<int32, [1]> input_747_axes_0 = const()[name = string("input_747_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_13_norm_out_weight_to_fp16 = const()[name = string("module_layers_13_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182481344)))];
+            tensor<fp16, [1024]> module_layers_13_norm_out_bias_to_fp16 = const()[name = string("module_layers_13_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182483456)))];
+            tensor<fp16, [1, 188, 1024]> input_747_cast_fp16 = layer_norm(axes = input_747_axes_0, beta = module_layers_13_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_out_weight_to_fp16, x = input_745_cast_fp16)[name = string("input_747_cast_fp16")];
+            tensor<int32, [1]> input_749_axes_0 = const()[name = string("input_749_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_14_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_14_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182485568)))];
+            tensor<fp16, [1024]> module_layers_14_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_14_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182487680)))];
+            tensor<fp16, [1, 188, 1024]> input_749_cast_fp16 = layer_norm(axes = input_749_axes_0, beta = module_layers_14_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_feed_forward1_weight_to_fp16, x = input_747_cast_fp16)[name = string("input_749_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2059717]> module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182489792))), indices_nonzero_data = tensor<uint6, [2059717]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183014144))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184559040))))[name = string("module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_127_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_749_cast_fp16)[name = string("linear_127_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_753_cast_fp16 = silu(x = linear_127_cast_fp16)[name = string("input_753_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1924564]> module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184624640))), indices_nonzero_data = tensor<uint6, [1924564]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185148992))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186592512))))[name = string("module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_128_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_753_cast_fp16)[name = string("linear_128_cast_fp16")];
+            fp16 var_2514_to_fp16 = const()[name = string("op_2514_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_2515_cast_fp16 = mul(x = linear_128_cast_fp16, y = var_2514_to_fp16)[name = string("op_2515_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_759_cast_fp16 = add(x = input_747_cast_fp16, y = var_2515_cast_fp16)[name = string("input_759_cast_fp16")];
+            tensor<int32, [1]> query_29_axes_0 = const()[name = string("query_29_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_14_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_14_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186608960)))];
+            tensor<fp16, [1024]> module_layers_14_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_14_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186611072)))];
+            tensor<fp16, [1, 188, 1024]> query_29_cast_fp16 = layer_norm(axes = query_29_axes_0, beta = module_layers_14_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_self_att_weight_to_fp16, x = input_759_cast_fp16)[name = string("query_29_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525274]> module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186613184))), indices_nonzero_data = tensor<uint6, [525274]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186744320))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187138368))))[name = string("module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_129_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_29_cast_fp16)[name = string("linear_129_cast_fp16")];
+            tensor<int32, [4]> var_2531 = const()[name = string("op_2531"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_85_cast_fp16 = reshape(shape = var_2531, x = linear_129_cast_fp16)[name = string("q_85_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522998]> module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187154816))), indices_nonzero_data = tensor<uint6, [522998]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187285952))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187678272))))[name = string("module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_130_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_29_cast_fp16)[name = string("linear_130_cast_fp16")];
+            tensor<int32, [4]> var_2535 = const()[name = string("op_2535"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_57_cast_fp16 = reshape(shape = var_2535, x = linear_130_cast_fp16)[name = string("k_57_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522224]> module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187694720))), indices_nonzero_data = tensor<uint6, [522224]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187825856))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188217600))))[name = string("module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_131_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_29_cast_fp16)[name = string("linear_131_cast_fp16")];
+            tensor<int32, [4]> var_2539 = const()[name = string("op_2539"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_29_cast_fp16 = reshape(shape = var_2539, x = linear_131_cast_fp16)[name = string("v_29_cast_fp16")];
+            tensor<int32, [4]> value_31_perm_0 = const()[name = string("value_31_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_14_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_14_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188234048)))];
+            tensor<fp16, [1, 188, 8, 128]> var_2551_cast_fp16 = add(x = q_85_cast_fp16, y = module_layers_14_self_attn_pos_bias_u_to_fp16)[name = string("op_2551_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_14_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_14_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188236160)))];
+            tensor<fp16, [1, 188, 8, 128]> var_2553_cast_fp16 = add(x = q_85_cast_fp16, y = module_layers_14_self_attn_pos_bias_v_to_fp16)[name = string("op_2553_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_29_perm_0 = const()[name = string("q_with_bias_v_29_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_315_transpose_x_0 = const()[name = string("x_315_transpose_x_0"), val = bool(false)];
+            bool x_315_transpose_y_0 = const()[name = string("x_315_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_2555_to_fp16 = const()[name = string("op_2555_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(330710848)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_29_cast_fp16 = transpose(perm = q_with_bias_v_29_perm_0, x = var_2553_cast_fp16)[name = string("transpose_214")];
+            tensor<fp16, [1, 8, 188, 375]> x_315_cast_fp16 = matmul(transpose_x = x_315_transpose_x_0, transpose_y = x_315_transpose_y_0, x = q_with_bias_v_29_cast_fp16, y = var_2555_to_fp16)[name = string("x_315_cast_fp16")];
+            tensor<int32, [8]> x_317_pad_0 = const()[name = string("x_317_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_317_mode_0 = const()[name = string("x_317_mode_0"), val = string("constant")];
+            fp16 const_154_to_fp16 = const()[name = string("const_154_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_317_cast_fp16 = pad(constant_val = const_154_to_fp16, mode = x_317_mode_0, pad = x_317_pad_0, x = x_315_cast_fp16)[name = string("x_317_cast_fp16")];
+            tensor<int32, [4]> var_2563 = const()[name = string("op_2563"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_319_cast_fp16 = reshape(shape = var_2563, x = x_317_cast_fp16)[name = string("x_319_cast_fp16")];
+            tensor<int32, [4]> var_2567_begin_0 = const()[name = string("op_2567_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2567_end_0 = const()[name = string("op_2567_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_2567_end_mask_0 = const()[name = string("op_2567_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_2567_cast_fp16 = slice_by_index(begin = var_2567_begin_0, end = var_2567_end_0, end_mask = var_2567_end_mask_0, x = x_319_cast_fp16)[name = string("op_2567_cast_fp16")];
+            tensor<int32, [4]> var_2568 = const()[name = string("op_2568"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_57_cast_fp16 = reshape(shape = var_2568, x = var_2567_cast_fp16)[name = string("matrix_bd_57_cast_fp16")];
+            bool matrix_ac_29_transpose_x_0 = const()[name = string("matrix_ac_29_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_29_transpose_y_0 = const()[name = string("matrix_ac_29_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_124_perm_0 = const()[name = string("transpose_124_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_125_perm_0 = const()[name = string("transpose_125_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_125 = transpose(perm = transpose_125_perm_0, x = k_57_cast_fp16)[name = string("transpose_212")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_124 = transpose(perm = transpose_124_perm_0, x = var_2551_cast_fp16)[name = string("transpose_213")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_29_cast_fp16 = matmul(transpose_x = matrix_ac_29_transpose_x_0, transpose_y = matrix_ac_29_transpose_y_0, x = transpose_124, y = transpose_125)[name = string("matrix_ac_29_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_59_begin_0 = const()[name = string("matrix_bd_59_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_59_end_0 = const()[name = string("matrix_bd_59_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_59_end_mask_0 = const()[name = string("matrix_bd_59_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_59_cast_fp16 = slice_by_index(begin = matrix_bd_59_begin_0, end = matrix_bd_59_end_0, end_mask = matrix_bd_59_end_mask_0, x = matrix_bd_57_cast_fp16)[name = string("matrix_bd_59_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_2577_cast_fp16 = add(x = matrix_ac_29_cast_fp16, y = matrix_bd_59_cast_fp16)[name = string("op_2577_cast_fp16")];
+            fp16 _inversed_scores_57_y_0_to_fp16 = const()[name = string("_inversed_scores_57_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_57_cast_fp16 = mul(x = var_2577_cast_fp16, y = _inversed_scores_57_y_0_to_fp16)[name = string("_inversed_scores_57_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_59_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_57_cast_fp16, cond = mask_3)[name = string("scores_59_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_2583_cast_fp16 = softmax(axis = var_30, x = scores_59_cast_fp16)[name = string("op_2583_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_761_cast_fp16 = select(a = var_11_to_fp16, b = var_2583_cast_fp16, cond = mask_3)[name = string("input_761_cast_fp16")];
+            bool x_321_transpose_x_0 = const()[name = string("x_321_transpose_x_0"), val = bool(false)];
+            bool x_321_transpose_y_0 = const()[name = string("x_321_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_31_cast_fp16 = transpose(perm = value_31_perm_0, x = v_29_cast_fp16)[name = string("transpose_211")];
+            tensor<fp16, [1, 8, 188, 128]> x_321_cast_fp16 = matmul(transpose_x = x_321_transpose_x_0, transpose_y = x_321_transpose_y_0, x = input_761_cast_fp16, y = value_31_cast_fp16)[name = string("x_321_cast_fp16")];
+            tensor<int32, [4]> var_2587_perm_0 = const()[name = string("op_2587_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2588 = const()[name = string("op_2588"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_2587_cast_fp16 = transpose(perm = var_2587_perm_0, x = x_321_cast_fp16)[name = string("transpose_210")];
+            tensor<fp16, [1, 188, 1024]> input_763_cast_fp16 = reshape(shape = var_2588, x = var_2587_cast_fp16)[name = string("input_763_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524764]> module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188498432))), indices_nonzero_data = tensor<uint6, [524764]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188629568))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189023232))))[name = string("module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_133_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_763_cast_fp16)[name = string("linear_133_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_767_cast_fp16 = add(x = input_759_cast_fp16, y = linear_133_cast_fp16)[name = string("input_767_cast_fp16")];
+            tensor<int32, [1]> x_325_axes_0 = const()[name = string("x_325_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_14_norm_conv_weight_to_fp16 = const()[name = string("module_layers_14_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189039680)))];
+            tensor<fp16, [1024]> module_layers_14_norm_conv_bias_to_fp16 = const()[name = string("module_layers_14_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189041792)))];
+            tensor<fp16, [1, 188, 1024]> x_325_cast_fp16 = layer_norm(axes = x_325_axes_0, beta = module_layers_14_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_conv_weight_to_fp16, x = input_767_cast_fp16)[name = string("x_325_cast_fp16")];
+            tensor<int32, [3]> input_769_perm_0 = const()[name = string("input_769_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_771_pad_type_0 = const()[name = string("input_771_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_771_strides_0 = const()[name = string("input_771_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_771_pad_0 = const()[name = string("input_771_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_771_dilations_0 = const()[name = string("input_771_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_771_groups_0 = const()[name = string("input_771_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1043300]> module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189043904))), indices_nonzero_data = tensor<uint6, [1043300]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189306112))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190088704))))[name = string("module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_769_cast_fp16 = transpose(perm = input_769_perm_0, x = x_325_cast_fp16)[name = string("transpose_209")];
+            tensor<fp16, [1, 2048, 188]> input_771_cast_fp16 = conv(dilations = input_771_dilations_0, groups = input_771_groups_0, pad = input_771_pad_0, pad_type = input_771_pad_type_0, strides = input_771_strides_0, weight = module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_769_cast_fp16)[name = string("input_771_cast_fp16")];
+            int32 x_327_split_num_splits_0 = const()[name = string("x_327_split_num_splits_0"), val = int32(2)];
+            int32 x_327_split_axis_0 = const()[name = string("x_327_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_327_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_327_split_cast_fp16_1 = split(axis = x_327_split_axis_0, num_splits = x_327_split_num_splits_0, x = input_771_cast_fp16)[name = string("x_327_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_327_split_1_sigmoid_cast_fp16 = sigmoid(x = x_327_split_cast_fp16_1)[name = string("x_327_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_327_cast_fp16 = mul(x = x_327_split_cast_fp16_0, y = x_327_split_1_sigmoid_cast_fp16)[name = string("x_327_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_773_cast_fp16 = select(a = var_11_to_fp16, b = x_327_cast_fp16, cond = var_328)[name = string("input_773_cast_fp16")];
+            tensor<int32, [6]> input_775_pad_0 = const()[name = string("input_775_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_775_mode_0 = const()[name = string("input_775_mode_0"), val = string("constant")];
+            fp16 const_157_to_fp16 = const()[name = string("const_157_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_775_cast_fp16 = pad(constant_val = const_157_to_fp16, mode = input_775_mode_0, pad = input_775_pad_0, x = input_773_cast_fp16)[name = string("input_775_cast_fp16")];
+            string input_777_pad_type_0 = const()[name = string("input_777_pad_type_0"), val = string("valid")];
+            int32 input_777_groups_0 = const()[name = string("input_777_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_777_strides_0 = const()[name = string("input_777_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_777_pad_0 = const()[name = string("input_777_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_777_dilations_0 = const()[name = string("input_777_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_276_to_fp16 = const()[name = string("const_276_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190121536)))];
+            tensor<fp16, [1024]> const_277_to_fp16 = const()[name = string("const_277_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190140032)))];
+            tensor<fp16, [1, 1024, 188]> input_779_cast_fp16 = conv(bias = const_277_to_fp16, dilations = input_777_dilations_0, groups = input_777_groups_0, pad = input_777_pad_0, pad_type = input_777_pad_type_0, strides = input_777_strides_0, weight = const_276_to_fp16, x = input_775_cast_fp16)[name = string("input_779_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_781_cast_fp16 = silu(x = input_779_cast_fp16)[name = string("input_781_cast_fp16")];
+            string x_329_pad_type_0 = const()[name = string("x_329_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_329_strides_0 = const()[name = string("x_329_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_329_pad_0 = const()[name = string("x_329_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_329_dilations_0 = const()[name = string("x_329_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_329_groups_0 = const()[name = string("x_329_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525590]> module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190142144))), indices_nonzero_data = tensor<uint6, [525590]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190273280))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190667584))))[name = string("module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_329_cast_fp16 = conv(dilations = x_329_dilations_0, groups = x_329_groups_0, pad = x_329_pad_0, pad_type = x_329_pad_type_0, strides = x_329_strides_0, weight = module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_781_cast_fp16)[name = string("x_329_cast_fp16")];
+            tensor<int32, [3]> input_783_perm_0 = const()[name = string("input_783_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_783_cast_fp16 = transpose(perm = input_783_perm_0, x = x_329_cast_fp16)[name = string("transpose_208")];
+            tensor<fp16, [1, 188, 1024]> input_785_cast_fp16 = add(x = input_767_cast_fp16, y = input_783_cast_fp16)[name = string("input_785_cast_fp16")];
+            tensor<int32, [1]> input_787_axes_0 = const()[name = string("input_787_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_14_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_14_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190684032)))];
+            tensor<fp16, [1024]> module_layers_14_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_14_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190686144)))];
+            tensor<fp16, [1, 188, 1024]> input_787_cast_fp16 = layer_norm(axes = input_787_axes_0, beta = module_layers_14_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_feed_forward2_weight_to_fp16, x = input_785_cast_fp16)[name = string("input_787_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2073313]> module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190688256))), indices_nonzero_data = tensor<uint6, [2073313]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191212608))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192767680))))[name = string("module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_134_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_787_cast_fp16)[name = string("linear_134_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_791_cast_fp16 = silu(x = linear_134_cast_fp16)[name = string("input_791_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1989781]> module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192833280))), indices_nonzero_data = tensor<uint6, [1989781]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193357632))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194850048))))[name = string("module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_135_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_791_cast_fp16)[name = string("linear_135_cast_fp16")];
+            fp16 var_2648_to_fp16 = const()[name = string("op_2648_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_2649_cast_fp16 = mul(x = linear_135_cast_fp16, y = var_2648_to_fp16)[name = string("op_2649_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_797_cast_fp16 = add(x = input_785_cast_fp16, y = var_2649_cast_fp16)[name = string("input_797_cast_fp16")];
+            tensor<int32, [1]> input_799_axes_0 = const()[name = string("input_799_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_14_norm_out_weight_to_fp16 = const()[name = string("module_layers_14_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194866496)))];
+            tensor<fp16, [1024]> module_layers_14_norm_out_bias_to_fp16 = const()[name = string("module_layers_14_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194868608)))];
+            tensor<fp16, [1, 188, 1024]> input_799_cast_fp16 = layer_norm(axes = input_799_axes_0, beta = module_layers_14_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_out_weight_to_fp16, x = input_797_cast_fp16)[name = string("input_799_cast_fp16")];
+            tensor<int32, [1]> input_801_axes_0 = const()[name = string("input_801_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_15_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_15_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194870720)))];
+            tensor<fp16, [1024]> module_layers_15_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_15_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194872832)))];
+            tensor<fp16, [1, 188, 1024]> input_801_cast_fp16 = layer_norm(axes = input_801_axes_0, beta = module_layers_15_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_feed_forward1_weight_to_fp16, x = input_799_cast_fp16)[name = string("input_801_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2064877]> module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194874944))), indices_nonzero_data = tensor<uint6, [2064877]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195399296))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196948032))))[name = string("module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_136_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_801_cast_fp16)[name = string("linear_136_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_805_cast_fp16 = silu(x = linear_136_cast_fp16)[name = string("input_805_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1919271]> module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197013632))), indices_nonzero_data = tensor<uint6, [1919271]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197537984))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198977536))))[name = string("module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_137_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_805_cast_fp16)[name = string("linear_137_cast_fp16")];
+            fp16 var_2677_to_fp16 = const()[name = string("op_2677_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_2678_cast_fp16 = mul(x = linear_137_cast_fp16, y = var_2677_to_fp16)[name = string("op_2678_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_811_cast_fp16 = add(x = input_799_cast_fp16, y = var_2678_cast_fp16)[name = string("input_811_cast_fp16")];
+            tensor<int32, [1]> query_31_axes_0 = const()[name = string("query_31_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_15_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_15_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198993984)))];
+            tensor<fp16, [1024]> module_layers_15_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_15_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198996096)))];
+            tensor<fp16, [1, 188, 1024]> query_31_cast_fp16 = layer_norm(axes = query_31_axes_0, beta = module_layers_15_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_self_att_weight_to_fp16, x = input_811_cast_fp16)[name = string("query_31_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525795]> module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198998208))), indices_nonzero_data = tensor<uint6, [525795]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199129344))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199523776))))[name = string("module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_138_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_31_cast_fp16)[name = string("linear_138_cast_fp16")];
+            tensor<int32, [4]> var_2694 = const()[name = string("op_2694"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_91_cast_fp16 = reshape(shape = var_2694, x = linear_138_cast_fp16)[name = string("q_91_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523094]> module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199540224))), indices_nonzero_data = tensor<uint6, [523094]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199671360))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200063808))))[name = string("module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_139_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_31_cast_fp16)[name = string("linear_139_cast_fp16")];
+            tensor<int32, [4]> var_2698 = const()[name = string("op_2698"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_61_cast_fp16 = reshape(shape = var_2698, x = linear_139_cast_fp16)[name = string("k_61_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523685]> module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200080256))), indices_nonzero_data = tensor<uint6, [523685]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200211392))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200604224))))[name = string("module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_140_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_31_cast_fp16)[name = string("linear_140_cast_fp16")];
+            tensor<int32, [4]> var_2702 = const()[name = string("op_2702"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_31_cast_fp16 = reshape(shape = var_2702, x = linear_140_cast_fp16)[name = string("v_31_cast_fp16")];
+            tensor<int32, [4]> value_33_perm_0 = const()[name = string("value_33_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_15_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_15_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200620672)))];
+            tensor<fp16, [1, 188, 8, 128]> var_2714_cast_fp16 = add(x = q_91_cast_fp16, y = module_layers_15_self_attn_pos_bias_u_to_fp16)[name = string("op_2714_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_15_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_15_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200622784)))];
+            tensor<fp16, [1, 188, 8, 128]> var_2716_cast_fp16 = add(x = q_91_cast_fp16, y = module_layers_15_self_attn_pos_bias_v_to_fp16)[name = string("op_2716_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_31_perm_0 = const()[name = string("q_with_bias_v_31_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_337_transpose_x_0 = const()[name = string("x_337_transpose_x_0"), val = bool(false)];
+            bool x_337_transpose_y_0 = const()[name = string("x_337_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_2718_to_fp16 = const()[name = string("op_2718_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(331478912)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_31_cast_fp16 = transpose(perm = q_with_bias_v_31_perm_0, x = var_2716_cast_fp16)[name = string("transpose_207")];
+            tensor<fp16, [1, 8, 188, 375]> x_337_cast_fp16 = matmul(transpose_x = x_337_transpose_x_0, transpose_y = x_337_transpose_y_0, x = q_with_bias_v_31_cast_fp16, y = var_2718_to_fp16)[name = string("x_337_cast_fp16")];
+            tensor<int32, [8]> x_339_pad_0 = const()[name = string("x_339_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_339_mode_0 = const()[name = string("x_339_mode_0"), val = string("constant")];
+            fp16 const_164_to_fp16 = const()[name = string("const_164_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_339_cast_fp16 = pad(constant_val = const_164_to_fp16, mode = x_339_mode_0, pad = x_339_pad_0, x = x_337_cast_fp16)[name = string("x_339_cast_fp16")];
+            tensor<int32, [4]> var_2726 = const()[name = string("op_2726"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_341_cast_fp16 = reshape(shape = var_2726, x = x_339_cast_fp16)[name = string("x_341_cast_fp16")];
+            tensor<int32, [4]> var_2730_begin_0 = const()[name = string("op_2730_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2730_end_0 = const()[name = string("op_2730_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_2730_end_mask_0 = const()[name = string("op_2730_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_2730_cast_fp16 = slice_by_index(begin = var_2730_begin_0, end = var_2730_end_0, end_mask = var_2730_end_mask_0, x = x_341_cast_fp16)[name = string("op_2730_cast_fp16")];
+            tensor<int32, [4]> var_2731 = const()[name = string("op_2731"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_61_cast_fp16 = reshape(shape = var_2731, x = var_2730_cast_fp16)[name = string("matrix_bd_61_cast_fp16")];
+            bool matrix_ac_31_transpose_x_0 = const()[name = string("matrix_ac_31_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_31_transpose_y_0 = const()[name = string("matrix_ac_31_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_126_perm_0 = const()[name = string("transpose_126_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_127_perm_0 = const()[name = string("transpose_127_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_127 = transpose(perm = transpose_127_perm_0, x = k_61_cast_fp16)[name = string("transpose_205")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_126 = transpose(perm = transpose_126_perm_0, x = var_2714_cast_fp16)[name = string("transpose_206")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_31_cast_fp16 = matmul(transpose_x = matrix_ac_31_transpose_x_0, transpose_y = matrix_ac_31_transpose_y_0, x = transpose_126, y = transpose_127)[name = string("matrix_ac_31_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_63_begin_0 = const()[name = string("matrix_bd_63_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_63_end_0 = const()[name = string("matrix_bd_63_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_63_end_mask_0 = const()[name = string("matrix_bd_63_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_63_cast_fp16 = slice_by_index(begin = matrix_bd_63_begin_0, end = matrix_bd_63_end_0, end_mask = matrix_bd_63_end_mask_0, x = matrix_bd_61_cast_fp16)[name = string("matrix_bd_63_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_2740_cast_fp16 = add(x = matrix_ac_31_cast_fp16, y = matrix_bd_63_cast_fp16)[name = string("op_2740_cast_fp16")];
+            fp16 _inversed_scores_61_y_0_to_fp16 = const()[name = string("_inversed_scores_61_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_61_cast_fp16 = mul(x = var_2740_cast_fp16, y = _inversed_scores_61_y_0_to_fp16)[name = string("_inversed_scores_61_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_63_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_61_cast_fp16, cond = mask_3)[name = string("scores_63_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_2746_cast_fp16 = softmax(axis = var_30, x = scores_63_cast_fp16)[name = string("op_2746_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_813_cast_fp16 = select(a = var_11_to_fp16, b = var_2746_cast_fp16, cond = mask_3)[name = string("input_813_cast_fp16")];
+            bool x_343_transpose_x_0 = const()[name = string("x_343_transpose_x_0"), val = bool(false)];
+            bool x_343_transpose_y_0 = const()[name = string("x_343_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_33_cast_fp16 = transpose(perm = value_33_perm_0, x = v_31_cast_fp16)[name = string("transpose_204")];
+            tensor<fp16, [1, 8, 188, 128]> x_343_cast_fp16 = matmul(transpose_x = x_343_transpose_x_0, transpose_y = x_343_transpose_y_0, x = input_813_cast_fp16, y = value_33_cast_fp16)[name = string("x_343_cast_fp16")];
+            tensor<int32, [4]> var_2750_perm_0 = const()[name = string("op_2750_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2751 = const()[name = string("op_2751"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_2750_cast_fp16 = transpose(perm = var_2750_perm_0, x = x_343_cast_fp16)[name = string("transpose_203")];
+            tensor<fp16, [1, 188, 1024]> input_815_cast_fp16 = reshape(shape = var_2751, x = var_2750_cast_fp16)[name = string("input_815_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [526464]> module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200885056))), indices_nonzero_data = tensor<uint6, [526464]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201016192))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201411136))))[name = string("module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_142_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_815_cast_fp16)[name = string("linear_142_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_819_cast_fp16 = add(x = input_811_cast_fp16, y = linear_142_cast_fp16)[name = string("input_819_cast_fp16")];
+            tensor<int32, [1]> x_347_axes_0 = const()[name = string("x_347_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_15_norm_conv_weight_to_fp16 = const()[name = string("module_layers_15_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201427584)))];
+            tensor<fp16, [1024]> module_layers_15_norm_conv_bias_to_fp16 = const()[name = string("module_layers_15_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201429696)))];
+            tensor<fp16, [1, 188, 1024]> x_347_cast_fp16 = layer_norm(axes = x_347_axes_0, beta = module_layers_15_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_conv_weight_to_fp16, x = input_819_cast_fp16)[name = string("x_347_cast_fp16")];
+            tensor<int32, [3]> input_821_perm_0 = const()[name = string("input_821_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_823_pad_type_0 = const()[name = string("input_823_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_823_strides_0 = const()[name = string("input_823_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_823_pad_0 = const()[name = string("input_823_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_823_dilations_0 = const()[name = string("input_823_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_823_groups_0 = const()[name = string("input_823_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1046264]> module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201431808))), indices_nonzero_data = tensor<uint6, [1046264]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201694016))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202478784))))[name = string("module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_821_cast_fp16 = transpose(perm = input_821_perm_0, x = x_347_cast_fp16)[name = string("transpose_202")];
+            tensor<fp16, [1, 2048, 188]> input_823_cast_fp16 = conv(dilations = input_823_dilations_0, groups = input_823_groups_0, pad = input_823_pad_0, pad_type = input_823_pad_type_0, strides = input_823_strides_0, weight = module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_821_cast_fp16)[name = string("input_823_cast_fp16")];
+            int32 x_349_split_num_splits_0 = const()[name = string("x_349_split_num_splits_0"), val = int32(2)];
+            int32 x_349_split_axis_0 = const()[name = string("x_349_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_349_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_349_split_cast_fp16_1 = split(axis = x_349_split_axis_0, num_splits = x_349_split_num_splits_0, x = input_823_cast_fp16)[name = string("x_349_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_349_split_1_sigmoid_cast_fp16 = sigmoid(x = x_349_split_cast_fp16_1)[name = string("x_349_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_349_cast_fp16 = mul(x = x_349_split_cast_fp16_0, y = x_349_split_1_sigmoid_cast_fp16)[name = string("x_349_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_825_cast_fp16 = select(a = var_11_to_fp16, b = x_349_cast_fp16, cond = var_328)[name = string("input_825_cast_fp16")];
+            tensor<int32, [6]> input_827_pad_0 = const()[name = string("input_827_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_827_mode_0 = const()[name = string("input_827_mode_0"), val = string("constant")];
+            fp16 const_167_to_fp16 = const()[name = string("const_167_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_827_cast_fp16 = pad(constant_val = const_167_to_fp16, mode = input_827_mode_0, pad = input_827_pad_0, x = input_825_cast_fp16)[name = string("input_827_cast_fp16")];
+            string input_829_pad_type_0 = const()[name = string("input_829_pad_type_0"), val = string("valid")];
+            int32 input_829_groups_0 = const()[name = string("input_829_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_829_strides_0 = const()[name = string("input_829_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_829_pad_0 = const()[name = string("input_829_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_829_dilations_0 = const()[name = string("input_829_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_278_to_fp16 = const()[name = string("const_278_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202511616)))];
+            tensor<fp16, [1024]> const_279_to_fp16 = const()[name = string("const_279_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202530112)))];
+            tensor<fp16, [1, 1024, 188]> input_831_cast_fp16 = conv(bias = const_279_to_fp16, dilations = input_829_dilations_0, groups = input_829_groups_0, pad = input_829_pad_0, pad_type = input_829_pad_type_0, strides = input_829_strides_0, weight = const_278_to_fp16, x = input_827_cast_fp16)[name = string("input_831_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_833_cast_fp16 = silu(x = input_831_cast_fp16)[name = string("input_833_cast_fp16")];
+            string x_351_pad_type_0 = const()[name = string("x_351_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_351_strides_0 = const()[name = string("x_351_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_351_pad_0 = const()[name = string("x_351_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_351_dilations_0 = const()[name = string("x_351_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_351_groups_0 = const()[name = string("x_351_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524462]> module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202532224))), indices_nonzero_data = tensor<uint6, [524462]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202663360))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203056832))))[name = string("module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_351_cast_fp16 = conv(dilations = x_351_dilations_0, groups = x_351_groups_0, pad = x_351_pad_0, pad_type = x_351_pad_type_0, strides = x_351_strides_0, weight = module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_833_cast_fp16)[name = string("x_351_cast_fp16")];
+            tensor<int32, [3]> input_835_perm_0 = const()[name = string("input_835_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_835_cast_fp16 = transpose(perm = input_835_perm_0, x = x_351_cast_fp16)[name = string("transpose_201")];
+            tensor<fp16, [1, 188, 1024]> input_837_cast_fp16 = add(x = input_819_cast_fp16, y = input_835_cast_fp16)[name = string("input_837_cast_fp16")];
+            tensor<int32, [1]> input_839_axes_0 = const()[name = string("input_839_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_15_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_15_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203073280)))];
+            tensor<fp16, [1024]> module_layers_15_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_15_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203075392)))];
+            tensor<fp16, [1, 188, 1024]> input_839_cast_fp16 = layer_norm(axes = input_839_axes_0, beta = module_layers_15_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_feed_forward2_weight_to_fp16, x = input_837_cast_fp16)[name = string("input_839_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2078588]> module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203077504))), indices_nonzero_data = tensor<uint6, [2078588]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203601856))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205160896))))[name = string("module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_143_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_839_cast_fp16)[name = string("linear_143_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_843_cast_fp16 = silu(x = linear_143_cast_fp16)[name = string("input_843_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1967312]> module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205226496))), indices_nonzero_data = tensor<uint6, [1967312]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205750848))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207226432))))[name = string("module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_144_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_843_cast_fp16)[name = string("linear_144_cast_fp16")];
+            fp16 var_2811_to_fp16 = const()[name = string("op_2811_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_2812_cast_fp16 = mul(x = linear_144_cast_fp16, y = var_2811_to_fp16)[name = string("op_2812_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_849_cast_fp16 = add(x = input_837_cast_fp16, y = var_2812_cast_fp16)[name = string("input_849_cast_fp16")];
+            tensor<int32, [1]> input_851_axes_0 = const()[name = string("input_851_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_15_norm_out_weight_to_fp16 = const()[name = string("module_layers_15_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207242880)))];
+            tensor<fp16, [1024]> module_layers_15_norm_out_bias_to_fp16 = const()[name = string("module_layers_15_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207244992)))];
+            tensor<fp16, [1, 188, 1024]> input_851_cast_fp16 = layer_norm(axes = input_851_axes_0, beta = module_layers_15_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_out_weight_to_fp16, x = input_849_cast_fp16)[name = string("input_851_cast_fp16")];
+            tensor<int32, [1]> input_853_axes_0 = const()[name = string("input_853_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_16_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_16_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207247104)))];
+            tensor<fp16, [1024]> module_layers_16_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_16_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207249216)))];
+            tensor<fp16, [1, 188, 1024]> input_853_cast_fp16 = layer_norm(axes = input_853_axes_0, beta = module_layers_16_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_feed_forward1_weight_to_fp16, x = input_851_cast_fp16)[name = string("input_853_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2069902]> module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207251328))), indices_nonzero_data = tensor<uint6, [2069902]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207775680))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209328192))))[name = string("module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_145_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_853_cast_fp16)[name = string("linear_145_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_857_cast_fp16 = silu(x = linear_145_cast_fp16)[name = string("input_857_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1927421]> module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209393792))), indices_nonzero_data = tensor<uint6, [1927421]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209918144))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211363776))))[name = string("module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_146_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_857_cast_fp16)[name = string("linear_146_cast_fp16")];
+            fp16 var_2840_to_fp16 = const()[name = string("op_2840_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_2841_cast_fp16 = mul(x = linear_146_cast_fp16, y = var_2840_to_fp16)[name = string("op_2841_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_863_cast_fp16 = add(x = input_851_cast_fp16, y = var_2841_cast_fp16)[name = string("input_863_cast_fp16")];
+            tensor<int32, [1]> query_33_axes_0 = const()[name = string("query_33_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_16_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_16_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211380224)))];
+            tensor<fp16, [1024]> module_layers_16_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_16_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211382336)))];
+            tensor<fp16, [1, 188, 1024]> query_33_cast_fp16 = layer_norm(axes = query_33_axes_0, beta = module_layers_16_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_self_att_weight_to_fp16, x = input_863_cast_fp16)[name = string("query_33_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524478]> module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211384448))), indices_nonzero_data = tensor<uint6, [524478]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211515584))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211909056))))[name = string("module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_147_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_33_cast_fp16)[name = string("linear_147_cast_fp16")];
+            tensor<int32, [4]> var_2857 = const()[name = string("op_2857"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_97_cast_fp16 = reshape(shape = var_2857, x = linear_147_cast_fp16)[name = string("q_97_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524779]> module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211925504))), indices_nonzero_data = tensor<uint6, [524779]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212056640))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212450304))))[name = string("module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_148_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_33_cast_fp16)[name = string("linear_148_cast_fp16")];
+            tensor<int32, [4]> var_2861 = const()[name = string("op_2861"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_65_cast_fp16 = reshape(shape = var_2861, x = linear_148_cast_fp16)[name = string("k_65_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525290]> module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212466752))), indices_nonzero_data = tensor<uint6, [525290]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212597888))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212991936))))[name = string("module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_149_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_33_cast_fp16)[name = string("linear_149_cast_fp16")];
+            tensor<int32, [4]> var_2865 = const()[name = string("op_2865"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_33_cast_fp16 = reshape(shape = var_2865, x = linear_149_cast_fp16)[name = string("v_33_cast_fp16")];
+            tensor<int32, [4]> value_35_perm_0 = const()[name = string("value_35_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_16_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_16_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213008384)))];
+            tensor<fp16, [1, 188, 8, 128]> var_2877_cast_fp16 = add(x = q_97_cast_fp16, y = module_layers_16_self_attn_pos_bias_u_to_fp16)[name = string("op_2877_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_16_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_16_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213010496)))];
+            tensor<fp16, [1, 188, 8, 128]> var_2879_cast_fp16 = add(x = q_97_cast_fp16, y = module_layers_16_self_attn_pos_bias_v_to_fp16)[name = string("op_2879_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_33_perm_0 = const()[name = string("q_with_bias_v_33_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_359_transpose_x_0 = const()[name = string("x_359_transpose_x_0"), val = bool(false)];
+            bool x_359_transpose_y_0 = const()[name = string("x_359_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_2881_to_fp16 = const()[name = string("op_2881_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(332246976)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_33_cast_fp16 = transpose(perm = q_with_bias_v_33_perm_0, x = var_2879_cast_fp16)[name = string("transpose_200")];
+            tensor<fp16, [1, 8, 188, 375]> x_359_cast_fp16 = matmul(transpose_x = x_359_transpose_x_0, transpose_y = x_359_transpose_y_0, x = q_with_bias_v_33_cast_fp16, y = var_2881_to_fp16)[name = string("x_359_cast_fp16")];
+            tensor<int32, [8]> x_361_pad_0 = const()[name = string("x_361_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_361_mode_0 = const()[name = string("x_361_mode_0"), val = string("constant")];
+            fp16 const_174_to_fp16 = const()[name = string("const_174_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_361_cast_fp16 = pad(constant_val = const_174_to_fp16, mode = x_361_mode_0, pad = x_361_pad_0, x = x_359_cast_fp16)[name = string("x_361_cast_fp16")];
+            tensor<int32, [4]> var_2889 = const()[name = string("op_2889"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_363_cast_fp16 = reshape(shape = var_2889, x = x_361_cast_fp16)[name = string("x_363_cast_fp16")];
+            tensor<int32, [4]> var_2893_begin_0 = const()[name = string("op_2893_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2893_end_0 = const()[name = string("op_2893_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_2893_end_mask_0 = const()[name = string("op_2893_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_2893_cast_fp16 = slice_by_index(begin = var_2893_begin_0, end = var_2893_end_0, end_mask = var_2893_end_mask_0, x = x_363_cast_fp16)[name = string("op_2893_cast_fp16")];
+            tensor<int32, [4]> var_2894 = const()[name = string("op_2894"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_65_cast_fp16 = reshape(shape = var_2894, x = var_2893_cast_fp16)[name = string("matrix_bd_65_cast_fp16")];
+            bool matrix_ac_33_transpose_x_0 = const()[name = string("matrix_ac_33_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_33_transpose_y_0 = const()[name = string("matrix_ac_33_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_128_perm_0 = const()[name = string("transpose_128_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_129_perm_0 = const()[name = string("transpose_129_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_129 = transpose(perm = transpose_129_perm_0, x = k_65_cast_fp16)[name = string("transpose_198")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_128 = transpose(perm = transpose_128_perm_0, x = var_2877_cast_fp16)[name = string("transpose_199")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_33_cast_fp16 = matmul(transpose_x = matrix_ac_33_transpose_x_0, transpose_y = matrix_ac_33_transpose_y_0, x = transpose_128, y = transpose_129)[name = string("matrix_ac_33_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_67_begin_0 = const()[name = string("matrix_bd_67_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_67_end_0 = const()[name = string("matrix_bd_67_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_67_end_mask_0 = const()[name = string("matrix_bd_67_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_67_cast_fp16 = slice_by_index(begin = matrix_bd_67_begin_0, end = matrix_bd_67_end_0, end_mask = matrix_bd_67_end_mask_0, x = matrix_bd_65_cast_fp16)[name = string("matrix_bd_67_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_2903_cast_fp16 = add(x = matrix_ac_33_cast_fp16, y = matrix_bd_67_cast_fp16)[name = string("op_2903_cast_fp16")];
+            fp16 _inversed_scores_65_y_0_to_fp16 = const()[name = string("_inversed_scores_65_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_65_cast_fp16 = mul(x = var_2903_cast_fp16, y = _inversed_scores_65_y_0_to_fp16)[name = string("_inversed_scores_65_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_67_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_65_cast_fp16, cond = mask_3)[name = string("scores_67_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_2909_cast_fp16 = softmax(axis = var_30, x = scores_67_cast_fp16)[name = string("op_2909_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_865_cast_fp16 = select(a = var_11_to_fp16, b = var_2909_cast_fp16, cond = mask_3)[name = string("input_865_cast_fp16")];
+            bool x_365_transpose_x_0 = const()[name = string("x_365_transpose_x_0"), val = bool(false)];
+            bool x_365_transpose_y_0 = const()[name = string("x_365_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_35_cast_fp16 = transpose(perm = value_35_perm_0, x = v_33_cast_fp16)[name = string("transpose_197")];
+            tensor<fp16, [1, 8, 188, 128]> x_365_cast_fp16 = matmul(transpose_x = x_365_transpose_x_0, transpose_y = x_365_transpose_y_0, x = input_865_cast_fp16, y = value_35_cast_fp16)[name = string("x_365_cast_fp16")];
+            tensor<int32, [4]> var_2913_perm_0 = const()[name = string("op_2913_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2914 = const()[name = string("op_2914"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_2913_cast_fp16 = transpose(perm = var_2913_perm_0, x = x_365_cast_fp16)[name = string("transpose_196")];
+            tensor<fp16, [1, 188, 1024]> input_867_cast_fp16 = reshape(shape = var_2914, x = var_2913_cast_fp16)[name = string("input_867_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523537]> module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213272768))), indices_nonzero_data = tensor<uint6, [523537]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213403904))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213796672))))[name = string("module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_151_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_867_cast_fp16)[name = string("linear_151_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_871_cast_fp16 = add(x = input_863_cast_fp16, y = linear_151_cast_fp16)[name = string("input_871_cast_fp16")];
+            tensor<int32, [1]> x_369_axes_0 = const()[name = string("x_369_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_16_norm_conv_weight_to_fp16 = const()[name = string("module_layers_16_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213813120)))];
+            tensor<fp16, [1024]> module_layers_16_norm_conv_bias_to_fp16 = const()[name = string("module_layers_16_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213815232)))];
+            tensor<fp16, [1, 188, 1024]> x_369_cast_fp16 = layer_norm(axes = x_369_axes_0, beta = module_layers_16_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_conv_weight_to_fp16, x = input_871_cast_fp16)[name = string("x_369_cast_fp16")];
+            tensor<int32, [3]> input_873_perm_0 = const()[name = string("input_873_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_875_pad_type_0 = const()[name = string("input_875_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_875_strides_0 = const()[name = string("input_875_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_875_pad_0 = const()[name = string("input_875_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_875_dilations_0 = const()[name = string("input_875_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_875_groups_0 = const()[name = string("input_875_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047665]> module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213817344))), indices_nonzero_data = tensor<uint6, [1047665]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214079552))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214865408))))[name = string("module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_873_cast_fp16 = transpose(perm = input_873_perm_0, x = x_369_cast_fp16)[name = string("transpose_195")];
+            tensor<fp16, [1, 2048, 188]> input_875_cast_fp16 = conv(dilations = input_875_dilations_0, groups = input_875_groups_0, pad = input_875_pad_0, pad_type = input_875_pad_type_0, strides = input_875_strides_0, weight = module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_873_cast_fp16)[name = string("input_875_cast_fp16")];
+            int32 x_371_split_num_splits_0 = const()[name = string("x_371_split_num_splits_0"), val = int32(2)];
+            int32 x_371_split_axis_0 = const()[name = string("x_371_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_371_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_371_split_cast_fp16_1 = split(axis = x_371_split_axis_0, num_splits = x_371_split_num_splits_0, x = input_875_cast_fp16)[name = string("x_371_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_371_split_1_sigmoid_cast_fp16 = sigmoid(x = x_371_split_cast_fp16_1)[name = string("x_371_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_371_cast_fp16 = mul(x = x_371_split_cast_fp16_0, y = x_371_split_1_sigmoid_cast_fp16)[name = string("x_371_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_877_cast_fp16 = select(a = var_11_to_fp16, b = x_371_cast_fp16, cond = var_328)[name = string("input_877_cast_fp16")];
+            tensor<int32, [6]> input_879_pad_0 = const()[name = string("input_879_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_879_mode_0 = const()[name = string("input_879_mode_0"), val = string("constant")];
+            fp16 const_177_to_fp16 = const()[name = string("const_177_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_879_cast_fp16 = pad(constant_val = const_177_to_fp16, mode = input_879_mode_0, pad = input_879_pad_0, x = input_877_cast_fp16)[name = string("input_879_cast_fp16")];
+            string input_881_pad_type_0 = const()[name = string("input_881_pad_type_0"), val = string("valid")];
+            int32 input_881_groups_0 = const()[name = string("input_881_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_881_strides_0 = const()[name = string("input_881_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_881_pad_0 = const()[name = string("input_881_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_881_dilations_0 = const()[name = string("input_881_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_280_to_fp16 = const()[name = string("const_280_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214898240)))];
+            tensor<fp16, [1024]> const_281_to_fp16 = const()[name = string("const_281_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214916736)))];
+            tensor<fp16, [1, 1024, 188]> input_883_cast_fp16 = conv(bias = const_281_to_fp16, dilations = input_881_dilations_0, groups = input_881_groups_0, pad = input_881_pad_0, pad_type = input_881_pad_type_0, strides = input_881_strides_0, weight = const_280_to_fp16, x = input_879_cast_fp16)[name = string("input_883_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_885_cast_fp16 = silu(x = input_883_cast_fp16)[name = string("input_885_cast_fp16")];
+            string x_373_pad_type_0 = const()[name = string("x_373_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_373_strides_0 = const()[name = string("x_373_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_373_pad_0 = const()[name = string("x_373_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_373_dilations_0 = const()[name = string("x_373_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_373_groups_0 = const()[name = string("x_373_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523727]> module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214918848))), indices_nonzero_data = tensor<uint6, [523727]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215049984))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215442880))))[name = string("module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_373_cast_fp16 = conv(dilations = x_373_dilations_0, groups = x_373_groups_0, pad = x_373_pad_0, pad_type = x_373_pad_type_0, strides = x_373_strides_0, weight = module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_885_cast_fp16)[name = string("x_373_cast_fp16")];
+            tensor<int32, [3]> input_887_perm_0 = const()[name = string("input_887_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_887_cast_fp16 = transpose(perm = input_887_perm_0, x = x_373_cast_fp16)[name = string("transpose_194")];
+            tensor<fp16, [1, 188, 1024]> input_889_cast_fp16 = add(x = input_871_cast_fp16, y = input_887_cast_fp16)[name = string("input_889_cast_fp16")];
+            tensor<int32, [1]> input_891_axes_0 = const()[name = string("input_891_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_16_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_16_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215459328)))];
+            tensor<fp16, [1024]> module_layers_16_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_16_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215461440)))];
+            tensor<fp16, [1, 188, 1024]> input_891_cast_fp16 = layer_norm(axes = input_891_axes_0, beta = module_layers_16_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_feed_forward2_weight_to_fp16, x = input_889_cast_fp16)[name = string("input_891_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2085434]> module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215463552))), indices_nonzero_data = tensor<uint6, [2085434]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215987904))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217552064))))[name = string("module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_152_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_891_cast_fp16)[name = string("linear_152_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_895_cast_fp16 = silu(x = linear_152_cast_fp16)[name = string("input_895_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2028445]> module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217617664))), indices_nonzero_data = tensor<uint6, [2028445]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218142016))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219663424))))[name = string("module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_153_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_895_cast_fp16)[name = string("linear_153_cast_fp16")];
+            fp16 var_2974_to_fp16 = const()[name = string("op_2974_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_2975_cast_fp16 = mul(x = linear_153_cast_fp16, y = var_2974_to_fp16)[name = string("op_2975_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_901_cast_fp16 = add(x = input_889_cast_fp16, y = var_2975_cast_fp16)[name = string("input_901_cast_fp16")];
+            tensor<int32, [1]> input_903_axes_0 = const()[name = string("input_903_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_16_norm_out_weight_to_fp16 = const()[name = string("module_layers_16_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219679872)))];
+            tensor<fp16, [1024]> module_layers_16_norm_out_bias_to_fp16 = const()[name = string("module_layers_16_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219681984)))];
+            tensor<fp16, [1, 188, 1024]> input_903_cast_fp16 = layer_norm(axes = input_903_axes_0, beta = module_layers_16_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_out_weight_to_fp16, x = input_901_cast_fp16)[name = string("input_903_cast_fp16")];
+            tensor<int32, [1]> input_905_axes_0 = const()[name = string("input_905_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_17_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_17_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219684096)))];
+            tensor<fp16, [1024]> module_layers_17_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_17_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219686208)))];
+            tensor<fp16, [1, 188, 1024]> input_905_cast_fp16 = layer_norm(axes = input_905_axes_0, beta = module_layers_17_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_feed_forward1_weight_to_fp16, x = input_903_cast_fp16)[name = string("input_905_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2078697]> module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219688320))), indices_nonzero_data = tensor<uint6, [2078697]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220212672))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221771776))))[name = string("module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_154_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_905_cast_fp16)[name = string("linear_154_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_909_cast_fp16 = silu(x = linear_154_cast_fp16)[name = string("input_909_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2015245]> module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221837376))), indices_nonzero_data = tensor<uint6, [2015245]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222361728))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223873280))))[name = string("module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_155_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_909_cast_fp16)[name = string("linear_155_cast_fp16")];
+            fp16 var_3003_to_fp16 = const()[name = string("op_3003_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_3004_cast_fp16 = mul(x = linear_155_cast_fp16, y = var_3003_to_fp16)[name = string("op_3004_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_915_cast_fp16 = add(x = input_903_cast_fp16, y = var_3004_cast_fp16)[name = string("input_915_cast_fp16")];
+            tensor<int32, [1]> query_35_axes_0 = const()[name = string("query_35_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_17_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_17_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223889728)))];
+            tensor<fp16, [1024]> module_layers_17_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_17_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223891840)))];
+            tensor<fp16, [1, 188, 1024]> query_35_cast_fp16 = layer_norm(axes = query_35_axes_0, beta = module_layers_17_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_self_att_weight_to_fp16, x = input_915_cast_fp16)[name = string("query_35_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525433]> module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223893952))), indices_nonzero_data = tensor<uint6, [525433]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224025088))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224419264))))[name = string("module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_156_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_35_cast_fp16)[name = string("linear_156_cast_fp16")];
+            tensor<int32, [4]> var_3020 = const()[name = string("op_3020"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_103_cast_fp16 = reshape(shape = var_3020, x = linear_156_cast_fp16)[name = string("q_103_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521661]> module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224435712))), indices_nonzero_data = tensor<uint6, [521661]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224566848))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224958208))))[name = string("module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_157_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_35_cast_fp16)[name = string("linear_157_cast_fp16")];
+            tensor<int32, [4]> var_3024 = const()[name = string("op_3024"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_69_cast_fp16 = reshape(shape = var_3024, x = linear_157_cast_fp16)[name = string("k_69_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521931]> module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224974656))), indices_nonzero_data = tensor<uint6, [521931]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225105792))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225497344))))[name = string("module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_158_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_35_cast_fp16)[name = string("linear_158_cast_fp16")];
+            tensor<int32, [4]> var_3028 = const()[name = string("op_3028"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_35_cast_fp16 = reshape(shape = var_3028, x = linear_158_cast_fp16)[name = string("v_35_cast_fp16")];
+            tensor<int32, [4]> value_37_perm_0 = const()[name = string("value_37_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_17_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_17_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225513792)))];
+            tensor<fp16, [1, 188, 8, 128]> var_3040_cast_fp16 = add(x = q_103_cast_fp16, y = module_layers_17_self_attn_pos_bias_u_to_fp16)[name = string("op_3040_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_17_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_17_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225515904)))];
+            tensor<fp16, [1, 188, 8, 128]> var_3042_cast_fp16 = add(x = q_103_cast_fp16, y = module_layers_17_self_attn_pos_bias_v_to_fp16)[name = string("op_3042_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_35_perm_0 = const()[name = string("q_with_bias_v_35_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_381_transpose_x_0 = const()[name = string("x_381_transpose_x_0"), val = bool(false)];
+            bool x_381_transpose_y_0 = const()[name = string("x_381_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_3044_to_fp16 = const()[name = string("op_3044_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(333015040)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_35_cast_fp16 = transpose(perm = q_with_bias_v_35_perm_0, x = var_3042_cast_fp16)[name = string("transpose_193")];
+            tensor<fp16, [1, 8, 188, 375]> x_381_cast_fp16 = matmul(transpose_x = x_381_transpose_x_0, transpose_y = x_381_transpose_y_0, x = q_with_bias_v_35_cast_fp16, y = var_3044_to_fp16)[name = string("x_381_cast_fp16")];
+            tensor<int32, [8]> x_383_pad_0 = const()[name = string("x_383_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_383_mode_0 = const()[name = string("x_383_mode_0"), val = string("constant")];
+            fp16 const_184_to_fp16 = const()[name = string("const_184_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_383_cast_fp16 = pad(constant_val = const_184_to_fp16, mode = x_383_mode_0, pad = x_383_pad_0, x = x_381_cast_fp16)[name = string("x_383_cast_fp16")];
+            tensor<int32, [4]> var_3052 = const()[name = string("op_3052"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_385_cast_fp16 = reshape(shape = var_3052, x = x_383_cast_fp16)[name = string("x_385_cast_fp16")];
+            tensor<int32, [4]> var_3056_begin_0 = const()[name = string("op_3056_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3056_end_0 = const()[name = string("op_3056_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_3056_end_mask_0 = const()[name = string("op_3056_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_3056_cast_fp16 = slice_by_index(begin = var_3056_begin_0, end = var_3056_end_0, end_mask = var_3056_end_mask_0, x = x_385_cast_fp16)[name = string("op_3056_cast_fp16")];
+            tensor<int32, [4]> var_3057 = const()[name = string("op_3057"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_69_cast_fp16 = reshape(shape = var_3057, x = var_3056_cast_fp16)[name = string("matrix_bd_69_cast_fp16")];
+            bool matrix_ac_35_transpose_x_0 = const()[name = string("matrix_ac_35_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_35_transpose_y_0 = const()[name = string("matrix_ac_35_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_130_perm_0 = const()[name = string("transpose_130_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_131_perm_0 = const()[name = string("transpose_131_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_131 = transpose(perm = transpose_131_perm_0, x = k_69_cast_fp16)[name = string("transpose_191")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_130 = transpose(perm = transpose_130_perm_0, x = var_3040_cast_fp16)[name = string("transpose_192")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_35_cast_fp16 = matmul(transpose_x = matrix_ac_35_transpose_x_0, transpose_y = matrix_ac_35_transpose_y_0, x = transpose_130, y = transpose_131)[name = string("matrix_ac_35_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_71_begin_0 = const()[name = string("matrix_bd_71_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_71_end_0 = const()[name = string("matrix_bd_71_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_71_end_mask_0 = const()[name = string("matrix_bd_71_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_71_cast_fp16 = slice_by_index(begin = matrix_bd_71_begin_0, end = matrix_bd_71_end_0, end_mask = matrix_bd_71_end_mask_0, x = matrix_bd_69_cast_fp16)[name = string("matrix_bd_71_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_3066_cast_fp16 = add(x = matrix_ac_35_cast_fp16, y = matrix_bd_71_cast_fp16)[name = string("op_3066_cast_fp16")];
+            fp16 _inversed_scores_69_y_0_to_fp16 = const()[name = string("_inversed_scores_69_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_69_cast_fp16 = mul(x = var_3066_cast_fp16, y = _inversed_scores_69_y_0_to_fp16)[name = string("_inversed_scores_69_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_71_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_69_cast_fp16, cond = mask_3)[name = string("scores_71_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_3072_cast_fp16 = softmax(axis = var_30, x = scores_71_cast_fp16)[name = string("op_3072_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_917_cast_fp16 = select(a = var_11_to_fp16, b = var_3072_cast_fp16, cond = mask_3)[name = string("input_917_cast_fp16")];
+            bool x_387_transpose_x_0 = const()[name = string("x_387_transpose_x_0"), val = bool(false)];
+            bool x_387_transpose_y_0 = const()[name = string("x_387_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_37_cast_fp16 = transpose(perm = value_37_perm_0, x = v_35_cast_fp16)[name = string("transpose_190")];
+            tensor<fp16, [1, 8, 188, 128]> x_387_cast_fp16 = matmul(transpose_x = x_387_transpose_x_0, transpose_y = x_387_transpose_y_0, x = input_917_cast_fp16, y = value_37_cast_fp16)[name = string("x_387_cast_fp16")];
+            tensor<int32, [4]> var_3076_perm_0 = const()[name = string("op_3076_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3077 = const()[name = string("op_3077"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_3076_cast_fp16 = transpose(perm = var_3076_perm_0, x = x_387_cast_fp16)[name = string("transpose_189")];
+            tensor<fp16, [1, 188, 1024]> input_919_cast_fp16 = reshape(shape = var_3077, x = var_3076_cast_fp16)[name = string("input_919_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [518816]> module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225778176))), indices_nonzero_data = tensor<uint6, [518816]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225909312))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226298496))))[name = string("module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_160_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_919_cast_fp16)[name = string("linear_160_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_923_cast_fp16 = add(x = input_915_cast_fp16, y = linear_160_cast_fp16)[name = string("input_923_cast_fp16")];
+            tensor<int32, [1]> x_391_axes_0 = const()[name = string("x_391_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_17_norm_conv_weight_to_fp16 = const()[name = string("module_layers_17_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226314944)))];
+            tensor<fp16, [1024]> module_layers_17_norm_conv_bias_to_fp16 = const()[name = string("module_layers_17_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226317056)))];
+            tensor<fp16, [1, 188, 1024]> x_391_cast_fp16 = layer_norm(axes = x_391_axes_0, beta = module_layers_17_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_conv_weight_to_fp16, x = input_923_cast_fp16)[name = string("x_391_cast_fp16")];
+            tensor<int32, [3]> input_925_perm_0 = const()[name = string("input_925_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_927_pad_type_0 = const()[name = string("input_927_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_927_strides_0 = const()[name = string("input_927_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_927_pad_0 = const()[name = string("input_927_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_927_dilations_0 = const()[name = string("input_927_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_927_groups_0 = const()[name = string("input_927_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047749]> module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226319168))), indices_nonzero_data = tensor<uint6, [1047749]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226581376))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227367296))))[name = string("module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_925_cast_fp16 = transpose(perm = input_925_perm_0, x = x_391_cast_fp16)[name = string("transpose_188")];
+            tensor<fp16, [1, 2048, 188]> input_927_cast_fp16 = conv(dilations = input_927_dilations_0, groups = input_927_groups_0, pad = input_927_pad_0, pad_type = input_927_pad_type_0, strides = input_927_strides_0, weight = module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_925_cast_fp16)[name = string("input_927_cast_fp16")];
+            int32 x_393_split_num_splits_0 = const()[name = string("x_393_split_num_splits_0"), val = int32(2)];
+            int32 x_393_split_axis_0 = const()[name = string("x_393_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_393_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_393_split_cast_fp16_1 = split(axis = x_393_split_axis_0, num_splits = x_393_split_num_splits_0, x = input_927_cast_fp16)[name = string("x_393_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_393_split_1_sigmoid_cast_fp16 = sigmoid(x = x_393_split_cast_fp16_1)[name = string("x_393_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_393_cast_fp16 = mul(x = x_393_split_cast_fp16_0, y = x_393_split_1_sigmoid_cast_fp16)[name = string("x_393_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_929_cast_fp16 = select(a = var_11_to_fp16, b = x_393_cast_fp16, cond = var_328)[name = string("input_929_cast_fp16")];
+            tensor<int32, [6]> input_931_pad_0 = const()[name = string("input_931_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_931_mode_0 = const()[name = string("input_931_mode_0"), val = string("constant")];
+            fp16 const_187_to_fp16 = const()[name = string("const_187_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_931_cast_fp16 = pad(constant_val = const_187_to_fp16, mode = input_931_mode_0, pad = input_931_pad_0, x = input_929_cast_fp16)[name = string("input_931_cast_fp16")];
+            string input_933_pad_type_0 = const()[name = string("input_933_pad_type_0"), val = string("valid")];
+            int32 input_933_groups_0 = const()[name = string("input_933_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_933_strides_0 = const()[name = string("input_933_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_933_pad_0 = const()[name = string("input_933_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_933_dilations_0 = const()[name = string("input_933_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_282_to_fp16 = const()[name = string("const_282_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227400128)))];
+            tensor<fp16, [1024]> const_283_to_fp16 = const()[name = string("const_283_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227418624)))];
+            tensor<fp16, [1, 1024, 188]> input_935_cast_fp16 = conv(bias = const_283_to_fp16, dilations = input_933_dilations_0, groups = input_933_groups_0, pad = input_933_pad_0, pad_type = input_933_pad_type_0, strides = input_933_strides_0, weight = const_282_to_fp16, x = input_931_cast_fp16)[name = string("input_935_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_937_cast_fp16 = silu(x = input_935_cast_fp16)[name = string("input_937_cast_fp16")];
+            string x_395_pad_type_0 = const()[name = string("x_395_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_395_strides_0 = const()[name = string("x_395_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_395_pad_0 = const()[name = string("x_395_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_395_dilations_0 = const()[name = string("x_395_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_395_groups_0 = const()[name = string("x_395_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521069]> module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227420736))), indices_nonzero_data = tensor<uint6, [521069]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227551872))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227942784))))[name = string("module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_395_cast_fp16 = conv(dilations = x_395_dilations_0, groups = x_395_groups_0, pad = x_395_pad_0, pad_type = x_395_pad_type_0, strides = x_395_strides_0, weight = module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_937_cast_fp16)[name = string("x_395_cast_fp16")];
+            tensor<int32, [3]> input_939_perm_0 = const()[name = string("input_939_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_939_cast_fp16 = transpose(perm = input_939_perm_0, x = x_395_cast_fp16)[name = string("transpose_187")];
+            tensor<fp16, [1, 188, 1024]> input_941_cast_fp16 = add(x = input_923_cast_fp16, y = input_939_cast_fp16)[name = string("input_941_cast_fp16")];
+            tensor<int32, [1]> input_943_axes_0 = const()[name = string("input_943_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_17_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_17_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227959232)))];
+            tensor<fp16, [1024]> module_layers_17_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_17_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227961344)))];
+            tensor<fp16, [1, 188, 1024]> input_943_cast_fp16 = layer_norm(axes = input_943_axes_0, beta = module_layers_17_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_feed_forward2_weight_to_fp16, x = input_941_cast_fp16)[name = string("input_943_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2085405]> module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227963456))), indices_nonzero_data = tensor<uint6, [2085405]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228487808))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230051968))))[name = string("module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_161_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_943_cast_fp16)[name = string("linear_161_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_947_cast_fp16 = silu(x = linear_161_cast_fp16)[name = string("input_947_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2032104]> module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230117568))), indices_nonzero_data = tensor<uint6, [2032104]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230641920))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232166080))))[name = string("module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_162_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_947_cast_fp16)[name = string("linear_162_cast_fp16")];
+            fp16 var_3137_to_fp16 = const()[name = string("op_3137_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_3138_cast_fp16 = mul(x = linear_162_cast_fp16, y = var_3137_to_fp16)[name = string("op_3138_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_953_cast_fp16 = add(x = input_941_cast_fp16, y = var_3138_cast_fp16)[name = string("input_953_cast_fp16")];
+            tensor<int32, [1]> input_955_axes_0 = const()[name = string("input_955_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_17_norm_out_weight_to_fp16 = const()[name = string("module_layers_17_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232182528)))];
+            tensor<fp16, [1024]> module_layers_17_norm_out_bias_to_fp16 = const()[name = string("module_layers_17_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232184640)))];
+            tensor<fp16, [1, 188, 1024]> input_955_cast_fp16 = layer_norm(axes = input_955_axes_0, beta = module_layers_17_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_out_weight_to_fp16, x = input_953_cast_fp16)[name = string("input_955_cast_fp16")];
+            tensor<int32, [1]> input_957_axes_0 = const()[name = string("input_957_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_18_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_18_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232186752)))];
+            tensor<fp16, [1024]> module_layers_18_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_18_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232188864)))];
+            tensor<fp16, [1, 188, 1024]> input_957_cast_fp16 = layer_norm(axes = input_957_axes_0, beta = module_layers_18_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_feed_forward1_weight_to_fp16, x = input_955_cast_fp16)[name = string("input_957_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2077029]> module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232190976))), indices_nonzero_data = tensor<uint6, [2077029]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232715328))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234273216))))[name = string("module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_163_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_957_cast_fp16)[name = string("linear_163_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_961_cast_fp16 = silu(x = linear_163_cast_fp16)[name = string("input_961_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2028343]> module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234338816))), indices_nonzero_data = tensor<uint6, [2028343]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234863168))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236384512))))[name = string("module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_164_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_961_cast_fp16)[name = string("linear_164_cast_fp16")];
+            fp16 var_3166_to_fp16 = const()[name = string("op_3166_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_3167_cast_fp16 = mul(x = linear_164_cast_fp16, y = var_3166_to_fp16)[name = string("op_3167_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_967_cast_fp16 = add(x = input_955_cast_fp16, y = var_3167_cast_fp16)[name = string("input_967_cast_fp16")];
+            tensor<int32, [1]> query_37_axes_0 = const()[name = string("query_37_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_18_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_18_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236400960)))];
+            tensor<fp16, [1024]> module_layers_18_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_18_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236403072)))];
+            tensor<fp16, [1, 188, 1024]> query_37_cast_fp16 = layer_norm(axes = query_37_axes_0, beta = module_layers_18_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_self_att_weight_to_fp16, x = input_967_cast_fp16)[name = string("query_37_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525392]> module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236405184))), indices_nonzero_data = tensor<uint6, [525392]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236536320))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236930432))))[name = string("module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_165_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_37_cast_fp16)[name = string("linear_165_cast_fp16")];
+            tensor<int32, [4]> var_3183 = const()[name = string("op_3183"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_109_cast_fp16 = reshape(shape = var_3183, x = linear_165_cast_fp16)[name = string("q_109_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524473]> module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236946880))), indices_nonzero_data = tensor<uint6, [524473]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237078016))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237471488))))[name = string("module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_166_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_37_cast_fp16)[name = string("linear_166_cast_fp16")];
+            tensor<int32, [4]> var_3187 = const()[name = string("op_3187"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_73_cast_fp16 = reshape(shape = var_3187, x = linear_166_cast_fp16)[name = string("k_73_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523564]> module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237487936))), indices_nonzero_data = tensor<uint6, [523564]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237619072))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238011840))))[name = string("module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_167_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_37_cast_fp16)[name = string("linear_167_cast_fp16")];
+            tensor<int32, [4]> var_3191 = const()[name = string("op_3191"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_37_cast_fp16 = reshape(shape = var_3191, x = linear_167_cast_fp16)[name = string("v_37_cast_fp16")];
+            tensor<int32, [4]> value_39_perm_0 = const()[name = string("value_39_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_18_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_18_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238028288)))];
+            tensor<fp16, [1, 188, 8, 128]> var_3203_cast_fp16 = add(x = q_109_cast_fp16, y = module_layers_18_self_attn_pos_bias_u_to_fp16)[name = string("op_3203_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_18_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_18_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238030400)))];
+            tensor<fp16, [1, 188, 8, 128]> var_3205_cast_fp16 = add(x = q_109_cast_fp16, y = module_layers_18_self_attn_pos_bias_v_to_fp16)[name = string("op_3205_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_37_perm_0 = const()[name = string("q_with_bias_v_37_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_403_transpose_x_0 = const()[name = string("x_403_transpose_x_0"), val = bool(false)];
+            bool x_403_transpose_y_0 = const()[name = string("x_403_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_3207_to_fp16 = const()[name = string("op_3207_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(333783104)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_37_cast_fp16 = transpose(perm = q_with_bias_v_37_perm_0, x = var_3205_cast_fp16)[name = string("transpose_186")];
+            tensor<fp16, [1, 8, 188, 375]> x_403_cast_fp16 = matmul(transpose_x = x_403_transpose_x_0, transpose_y = x_403_transpose_y_0, x = q_with_bias_v_37_cast_fp16, y = var_3207_to_fp16)[name = string("x_403_cast_fp16")];
+            tensor<int32, [8]> x_405_pad_0 = const()[name = string("x_405_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_405_mode_0 = const()[name = string("x_405_mode_0"), val = string("constant")];
+            fp16 const_194_to_fp16 = const()[name = string("const_194_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_405_cast_fp16 = pad(constant_val = const_194_to_fp16, mode = x_405_mode_0, pad = x_405_pad_0, x = x_403_cast_fp16)[name = string("x_405_cast_fp16")];
+            tensor<int32, [4]> var_3215 = const()[name = string("op_3215"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_407_cast_fp16 = reshape(shape = var_3215, x = x_405_cast_fp16)[name = string("x_407_cast_fp16")];
+            tensor<int32, [4]> var_3219_begin_0 = const()[name = string("op_3219_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3219_end_0 = const()[name = string("op_3219_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_3219_end_mask_0 = const()[name = string("op_3219_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_3219_cast_fp16 = slice_by_index(begin = var_3219_begin_0, end = var_3219_end_0, end_mask = var_3219_end_mask_0, x = x_407_cast_fp16)[name = string("op_3219_cast_fp16")];
+            tensor<int32, [4]> var_3220 = const()[name = string("op_3220"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_73_cast_fp16 = reshape(shape = var_3220, x = var_3219_cast_fp16)[name = string("matrix_bd_73_cast_fp16")];
+            bool matrix_ac_37_transpose_x_0 = const()[name = string("matrix_ac_37_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_37_transpose_y_0 = const()[name = string("matrix_ac_37_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_132_perm_0 = const()[name = string("transpose_132_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_133_perm_0 = const()[name = string("transpose_133_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_133 = transpose(perm = transpose_133_perm_0, x = k_73_cast_fp16)[name = string("transpose_184")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_132 = transpose(perm = transpose_132_perm_0, x = var_3203_cast_fp16)[name = string("transpose_185")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_37_cast_fp16 = matmul(transpose_x = matrix_ac_37_transpose_x_0, transpose_y = matrix_ac_37_transpose_y_0, x = transpose_132, y = transpose_133)[name = string("matrix_ac_37_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_75_begin_0 = const()[name = string("matrix_bd_75_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_75_end_0 = const()[name = string("matrix_bd_75_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_75_end_mask_0 = const()[name = string("matrix_bd_75_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_75_cast_fp16 = slice_by_index(begin = matrix_bd_75_begin_0, end = matrix_bd_75_end_0, end_mask = matrix_bd_75_end_mask_0, x = matrix_bd_73_cast_fp16)[name = string("matrix_bd_75_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_3229_cast_fp16 = add(x = matrix_ac_37_cast_fp16, y = matrix_bd_75_cast_fp16)[name = string("op_3229_cast_fp16")];
+            fp16 _inversed_scores_73_y_0_to_fp16 = const()[name = string("_inversed_scores_73_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_73_cast_fp16 = mul(x = var_3229_cast_fp16, y = _inversed_scores_73_y_0_to_fp16)[name = string("_inversed_scores_73_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_75_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_73_cast_fp16, cond = mask_3)[name = string("scores_75_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_3235_cast_fp16 = softmax(axis = var_30, x = scores_75_cast_fp16)[name = string("op_3235_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_969_cast_fp16 = select(a = var_11_to_fp16, b = var_3235_cast_fp16, cond = mask_3)[name = string("input_969_cast_fp16")];
+            bool x_409_transpose_x_0 = const()[name = string("x_409_transpose_x_0"), val = bool(false)];
+            bool x_409_transpose_y_0 = const()[name = string("x_409_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_39_cast_fp16 = transpose(perm = value_39_perm_0, x = v_37_cast_fp16)[name = string("transpose_183")];
+            tensor<fp16, [1, 8, 188, 128]> x_409_cast_fp16 = matmul(transpose_x = x_409_transpose_x_0, transpose_y = x_409_transpose_y_0, x = input_969_cast_fp16, y = value_39_cast_fp16)[name = string("x_409_cast_fp16")];
+            tensor<int32, [4]> var_3239_perm_0 = const()[name = string("op_3239_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3240 = const()[name = string("op_3240"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_3239_cast_fp16 = transpose(perm = var_3239_perm_0, x = x_409_cast_fp16)[name = string("transpose_182")];
+            tensor<fp16, [1, 188, 1024]> input_971_cast_fp16 = reshape(shape = var_3240, x = var_3239_cast_fp16)[name = string("input_971_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520692]> module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238292672))), indices_nonzero_data = tensor<uint6, [520692]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238423808))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238814400))))[name = string("module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_169_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_971_cast_fp16)[name = string("linear_169_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_975_cast_fp16 = add(x = input_967_cast_fp16, y = linear_169_cast_fp16)[name = string("input_975_cast_fp16")];
+            tensor<int32, [1]> x_413_axes_0 = const()[name = string("x_413_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_18_norm_conv_weight_to_fp16 = const()[name = string("module_layers_18_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238830848)))];
+            tensor<fp16, [1024]> module_layers_18_norm_conv_bias_to_fp16 = const()[name = string("module_layers_18_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238832960)))];
+            tensor<fp16, [1, 188, 1024]> x_413_cast_fp16 = layer_norm(axes = x_413_axes_0, beta = module_layers_18_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_conv_weight_to_fp16, x = input_975_cast_fp16)[name = string("x_413_cast_fp16")];
+            tensor<int32, [3]> input_977_perm_0 = const()[name = string("input_977_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_979_pad_type_0 = const()[name = string("input_979_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_979_strides_0 = const()[name = string("input_979_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_979_pad_0 = const()[name = string("input_979_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_979_dilations_0 = const()[name = string("input_979_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_979_groups_0 = const()[name = string("input_979_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047234]> module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238835072))), indices_nonzero_data = tensor<uint6, [1047234]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239097280))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239882816))))[name = string("module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_977_cast_fp16 = transpose(perm = input_977_perm_0, x = x_413_cast_fp16)[name = string("transpose_181")];
+            tensor<fp16, [1, 2048, 188]> input_979_cast_fp16 = conv(dilations = input_979_dilations_0, groups = input_979_groups_0, pad = input_979_pad_0, pad_type = input_979_pad_type_0, strides = input_979_strides_0, weight = module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_977_cast_fp16)[name = string("input_979_cast_fp16")];
+            int32 x_415_split_num_splits_0 = const()[name = string("x_415_split_num_splits_0"), val = int32(2)];
+            int32 x_415_split_axis_0 = const()[name = string("x_415_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_415_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_415_split_cast_fp16_1 = split(axis = x_415_split_axis_0, num_splits = x_415_split_num_splits_0, x = input_979_cast_fp16)[name = string("x_415_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_415_split_1_sigmoid_cast_fp16 = sigmoid(x = x_415_split_cast_fp16_1)[name = string("x_415_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_415_cast_fp16 = mul(x = x_415_split_cast_fp16_0, y = x_415_split_1_sigmoid_cast_fp16)[name = string("x_415_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_981_cast_fp16 = select(a = var_11_to_fp16, b = x_415_cast_fp16, cond = var_328)[name = string("input_981_cast_fp16")];
+            tensor<int32, [6]> input_983_pad_0 = const()[name = string("input_983_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_983_mode_0 = const()[name = string("input_983_mode_0"), val = string("constant")];
+            fp16 const_197_to_fp16 = const()[name = string("const_197_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_983_cast_fp16 = pad(constant_val = const_197_to_fp16, mode = input_983_mode_0, pad = input_983_pad_0, x = input_981_cast_fp16)[name = string("input_983_cast_fp16")];
+            string input_985_pad_type_0 = const()[name = string("input_985_pad_type_0"), val = string("valid")];
+            int32 input_985_groups_0 = const()[name = string("input_985_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_985_strides_0 = const()[name = string("input_985_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_985_pad_0 = const()[name = string("input_985_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_985_dilations_0 = const()[name = string("input_985_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_284_to_fp16 = const()[name = string("const_284_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239915648)))];
+            tensor<fp16, [1024]> const_285_to_fp16 = const()[name = string("const_285_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239934144)))];
+            tensor<fp16, [1, 1024, 188]> input_987_cast_fp16 = conv(bias = const_285_to_fp16, dilations = input_985_dilations_0, groups = input_985_groups_0, pad = input_985_pad_0, pad_type = input_985_pad_type_0, strides = input_985_strides_0, weight = const_284_to_fp16, x = input_983_cast_fp16)[name = string("input_987_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_989_cast_fp16 = silu(x = input_987_cast_fp16)[name = string("input_989_cast_fp16")];
+            string x_417_pad_type_0 = const()[name = string("x_417_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_417_strides_0 = const()[name = string("x_417_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_417_pad_0 = const()[name = string("x_417_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_417_dilations_0 = const()[name = string("x_417_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_417_groups_0 = const()[name = string("x_417_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521466]> module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239936256))), indices_nonzero_data = tensor<uint6, [521466]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240067392))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240458560))))[name = string("module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_417_cast_fp16 = conv(dilations = x_417_dilations_0, groups = x_417_groups_0, pad = x_417_pad_0, pad_type = x_417_pad_type_0, strides = x_417_strides_0, weight = module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_989_cast_fp16)[name = string("x_417_cast_fp16")];
+            tensor<int32, [3]> input_991_perm_0 = const()[name = string("input_991_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_991_cast_fp16 = transpose(perm = input_991_perm_0, x = x_417_cast_fp16)[name = string("transpose_180")];
+            tensor<fp16, [1, 188, 1024]> input_993_cast_fp16 = add(x = input_975_cast_fp16, y = input_991_cast_fp16)[name = string("input_993_cast_fp16")];
+            tensor<int32, [1]> input_995_axes_0 = const()[name = string("input_995_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_18_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_18_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240475008)))];
+            tensor<fp16, [1024]> module_layers_18_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_18_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240477120)))];
+            tensor<fp16, [1, 188, 1024]> input_995_cast_fp16 = layer_norm(axes = input_995_axes_0, beta = module_layers_18_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_feed_forward2_weight_to_fp16, x = input_993_cast_fp16)[name = string("input_995_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2081130]> module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240479232))), indices_nonzero_data = tensor<uint6, [2081130]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241003584))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242564544))))[name = string("module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_170_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_995_cast_fp16)[name = string("linear_170_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_999_cast_fp16 = silu(x = linear_170_cast_fp16)[name = string("input_999_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2068517]> module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242630144))), indices_nonzero_data = tensor<uint6, [2068517]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243154496))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244705984))))[name = string("module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_171_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_999_cast_fp16)[name = string("linear_171_cast_fp16")];
+            fp16 var_3300_to_fp16 = const()[name = string("op_3300_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_3301_cast_fp16 = mul(x = linear_171_cast_fp16, y = var_3300_to_fp16)[name = string("op_3301_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_1005_cast_fp16 = add(x = input_993_cast_fp16, y = var_3301_cast_fp16)[name = string("input_1005_cast_fp16")];
+            tensor<int32, [1]> input_1007_axes_0 = const()[name = string("input_1007_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_18_norm_out_weight_to_fp16 = const()[name = string("module_layers_18_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244722432)))];
+            tensor<fp16, [1024]> module_layers_18_norm_out_bias_to_fp16 = const()[name = string("module_layers_18_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244724544)))];
+            tensor<fp16, [1, 188, 1024]> input_1007_cast_fp16 = layer_norm(axes = input_1007_axes_0, beta = module_layers_18_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_out_weight_to_fp16, x = input_1005_cast_fp16)[name = string("input_1007_cast_fp16")];
+            tensor<int32, [1]> input_1009_axes_0 = const()[name = string("input_1009_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_19_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_19_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244726656)))];
+            tensor<fp16, [1024]> module_layers_19_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_19_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244728768)))];
+            tensor<fp16, [1, 188, 1024]> input_1009_cast_fp16 = layer_norm(axes = input_1009_axes_0, beta = module_layers_19_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_feed_forward1_weight_to_fp16, x = input_1007_cast_fp16)[name = string("input_1009_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2077149]> module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244730880))), indices_nonzero_data = tensor<uint6, [2077149]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245255232))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246813184))))[name = string("module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_172_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_1009_cast_fp16)[name = string("linear_172_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_1013_cast_fp16 = silu(x = linear_172_cast_fp16)[name = string("input_1013_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2050240]> module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246878784))), indices_nonzero_data = tensor<uint6, [2050240]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247403136))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248940928))))[name = string("module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_173_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_1013_cast_fp16)[name = string("linear_173_cast_fp16")];
+            fp16 var_3329_to_fp16 = const()[name = string("op_3329_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_3330_cast_fp16 = mul(x = linear_173_cast_fp16, y = var_3329_to_fp16)[name = string("op_3330_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_1019_cast_fp16 = add(x = input_1007_cast_fp16, y = var_3330_cast_fp16)[name = string("input_1019_cast_fp16")];
+            tensor<int32, [1]> query_39_axes_0 = const()[name = string("query_39_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_19_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_19_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248957376)))];
+            tensor<fp16, [1024]> module_layers_19_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_19_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248959488)))];
+            tensor<fp16, [1, 188, 1024]> query_39_cast_fp16 = layer_norm(axes = query_39_axes_0, beta = module_layers_19_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_self_att_weight_to_fp16, x = input_1019_cast_fp16)[name = string("query_39_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524663]> module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248961600))), indices_nonzero_data = tensor<uint6, [524663]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249092736))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249486336))))[name = string("module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_174_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_39_cast_fp16)[name = string("linear_174_cast_fp16")];
+            tensor<int32, [4]> var_3346 = const()[name = string("op_3346"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_115_cast_fp16 = reshape(shape = var_3346, x = linear_174_cast_fp16)[name = string("q_115_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525503]> module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249502784))), indices_nonzero_data = tensor<uint6, [525503]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249633920))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250028160))))[name = string("module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_175_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_39_cast_fp16)[name = string("linear_175_cast_fp16")];
+            tensor<int32, [4]> var_3350 = const()[name = string("op_3350"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_77_cast_fp16 = reshape(shape = var_3350, x = linear_175_cast_fp16)[name = string("k_77_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523873]> module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250044608))), indices_nonzero_data = tensor<uint6, [523873]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250175744))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250568768))))[name = string("module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_176_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_39_cast_fp16)[name = string("linear_176_cast_fp16")];
+            tensor<int32, [4]> var_3354 = const()[name = string("op_3354"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_39_cast_fp16 = reshape(shape = var_3354, x = linear_176_cast_fp16)[name = string("v_39_cast_fp16")];
+            tensor<int32, [4]> value_41_perm_0 = const()[name = string("value_41_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_19_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_19_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250585216)))];
+            tensor<fp16, [1, 188, 8, 128]> var_3366_cast_fp16 = add(x = q_115_cast_fp16, y = module_layers_19_self_attn_pos_bias_u_to_fp16)[name = string("op_3366_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_19_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_19_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250587328)))];
+            tensor<fp16, [1, 188, 8, 128]> var_3368_cast_fp16 = add(x = q_115_cast_fp16, y = module_layers_19_self_attn_pos_bias_v_to_fp16)[name = string("op_3368_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_39_perm_0 = const()[name = string("q_with_bias_v_39_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_425_transpose_x_0 = const()[name = string("x_425_transpose_x_0"), val = bool(false)];
+            bool x_425_transpose_y_0 = const()[name = string("x_425_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_3370_to_fp16 = const()[name = string("op_3370_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(334551168)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_39_cast_fp16 = transpose(perm = q_with_bias_v_39_perm_0, x = var_3368_cast_fp16)[name = string("transpose_179")];
+            tensor<fp16, [1, 8, 188, 375]> x_425_cast_fp16 = matmul(transpose_x = x_425_transpose_x_0, transpose_y = x_425_transpose_y_0, x = q_with_bias_v_39_cast_fp16, y = var_3370_to_fp16)[name = string("x_425_cast_fp16")];
+            tensor<int32, [8]> x_427_pad_0 = const()[name = string("x_427_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_427_mode_0 = const()[name = string("x_427_mode_0"), val = string("constant")];
+            fp16 const_204_to_fp16 = const()[name = string("const_204_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_427_cast_fp16 = pad(constant_val = const_204_to_fp16, mode = x_427_mode_0, pad = x_427_pad_0, x = x_425_cast_fp16)[name = string("x_427_cast_fp16")];
+            tensor<int32, [4]> var_3378 = const()[name = string("op_3378"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_429_cast_fp16 = reshape(shape = var_3378, x = x_427_cast_fp16)[name = string("x_429_cast_fp16")];
+            tensor<int32, [4]> var_3382_begin_0 = const()[name = string("op_3382_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3382_end_0 = const()[name = string("op_3382_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_3382_end_mask_0 = const()[name = string("op_3382_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_3382_cast_fp16 = slice_by_index(begin = var_3382_begin_0, end = var_3382_end_0, end_mask = var_3382_end_mask_0, x = x_429_cast_fp16)[name = string("op_3382_cast_fp16")];
+            tensor<int32, [4]> var_3383 = const()[name = string("op_3383"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_77_cast_fp16 = reshape(shape = var_3383, x = var_3382_cast_fp16)[name = string("matrix_bd_77_cast_fp16")];
+            bool matrix_ac_39_transpose_x_0 = const()[name = string("matrix_ac_39_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_39_transpose_y_0 = const()[name = string("matrix_ac_39_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_134_perm_0 = const()[name = string("transpose_134_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_135_perm_0 = const()[name = string("transpose_135_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_135 = transpose(perm = transpose_135_perm_0, x = k_77_cast_fp16)[name = string("transpose_177")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_134 = transpose(perm = transpose_134_perm_0, x = var_3366_cast_fp16)[name = string("transpose_178")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_39_cast_fp16 = matmul(transpose_x = matrix_ac_39_transpose_x_0, transpose_y = matrix_ac_39_transpose_y_0, x = transpose_134, y = transpose_135)[name = string("matrix_ac_39_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_79_begin_0 = const()[name = string("matrix_bd_79_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_79_end_0 = const()[name = string("matrix_bd_79_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_79_end_mask_0 = const()[name = string("matrix_bd_79_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_79_cast_fp16 = slice_by_index(begin = matrix_bd_79_begin_0, end = matrix_bd_79_end_0, end_mask = matrix_bd_79_end_mask_0, x = matrix_bd_77_cast_fp16)[name = string("matrix_bd_79_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_3392_cast_fp16 = add(x = matrix_ac_39_cast_fp16, y = matrix_bd_79_cast_fp16)[name = string("op_3392_cast_fp16")];
+            fp16 _inversed_scores_77_y_0_to_fp16 = const()[name = string("_inversed_scores_77_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_77_cast_fp16 = mul(x = var_3392_cast_fp16, y = _inversed_scores_77_y_0_to_fp16)[name = string("_inversed_scores_77_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_79_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_77_cast_fp16, cond = mask_3)[name = string("scores_79_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_3398_cast_fp16 = softmax(axis = var_30, x = scores_79_cast_fp16)[name = string("op_3398_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_1021_cast_fp16 = select(a = var_11_to_fp16, b = var_3398_cast_fp16, cond = mask_3)[name = string("input_1021_cast_fp16")];
+            bool x_431_transpose_x_0 = const()[name = string("x_431_transpose_x_0"), val = bool(false)];
+            bool x_431_transpose_y_0 = const()[name = string("x_431_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_41_cast_fp16 = transpose(perm = value_41_perm_0, x = v_39_cast_fp16)[name = string("transpose_176")];
+            tensor<fp16, [1, 8, 188, 128]> x_431_cast_fp16 = matmul(transpose_x = x_431_transpose_x_0, transpose_y = x_431_transpose_y_0, x = input_1021_cast_fp16, y = value_41_cast_fp16)[name = string("x_431_cast_fp16")];
+            tensor<int32, [4]> var_3402_perm_0 = const()[name = string("op_3402_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3403 = const()[name = string("op_3403"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_3402_cast_fp16 = transpose(perm = var_3402_perm_0, x = x_431_cast_fp16)[name = string("transpose_175")];
+            tensor<fp16, [1, 188, 1024]> input_1023_cast_fp16 = reshape(shape = var_3403, x = var_3402_cast_fp16)[name = string("input_1023_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523881]> module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250849600))), indices_nonzero_data = tensor<uint6, [523881]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250980736))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251373760))))[name = string("module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_178_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_1023_cast_fp16)[name = string("linear_178_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_1027_cast_fp16 = add(x = input_1019_cast_fp16, y = linear_178_cast_fp16)[name = string("input_1027_cast_fp16")];
+            tensor<int32, [1]> x_435_axes_0 = const()[name = string("x_435_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_19_norm_conv_weight_to_fp16 = const()[name = string("module_layers_19_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251390208)))];
+            tensor<fp16, [1024]> module_layers_19_norm_conv_bias_to_fp16 = const()[name = string("module_layers_19_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251392320)))];
+            tensor<fp16, [1, 188, 1024]> x_435_cast_fp16 = layer_norm(axes = x_435_axes_0, beta = module_layers_19_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_conv_weight_to_fp16, x = input_1027_cast_fp16)[name = string("x_435_cast_fp16")];
+            tensor<int32, [3]> input_1029_perm_0 = const()[name = string("input_1029_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_1031_pad_type_0 = const()[name = string("input_1031_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_1031_strides_0 = const()[name = string("input_1031_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1031_pad_0 = const()[name = string("input_1031_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1031_dilations_0 = const()[name = string("input_1031_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_1031_groups_0 = const()[name = string("input_1031_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047423]> module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251394432))), indices_nonzero_data = tensor<uint6, [1047423]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251656640))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252442304))))[name = string("module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_1029_cast_fp16 = transpose(perm = input_1029_perm_0, x = x_435_cast_fp16)[name = string("transpose_174")];
+            tensor<fp16, [1, 2048, 188]> input_1031_cast_fp16 = conv(dilations = input_1031_dilations_0, groups = input_1031_groups_0, pad = input_1031_pad_0, pad_type = input_1031_pad_type_0, strides = input_1031_strides_0, weight = module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_1029_cast_fp16)[name = string("input_1031_cast_fp16")];
+            int32 x_437_split_num_splits_0 = const()[name = string("x_437_split_num_splits_0"), val = int32(2)];
+            int32 x_437_split_axis_0 = const()[name = string("x_437_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_437_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_437_split_cast_fp16_1 = split(axis = x_437_split_axis_0, num_splits = x_437_split_num_splits_0, x = input_1031_cast_fp16)[name = string("x_437_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_437_split_1_sigmoid_cast_fp16 = sigmoid(x = x_437_split_cast_fp16_1)[name = string("x_437_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_437_cast_fp16 = mul(x = x_437_split_cast_fp16_0, y = x_437_split_1_sigmoid_cast_fp16)[name = string("x_437_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_1033_cast_fp16 = select(a = var_11_to_fp16, b = x_437_cast_fp16, cond = var_328)[name = string("input_1033_cast_fp16")];
+            tensor<int32, [6]> input_1035_pad_0 = const()[name = string("input_1035_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_1035_mode_0 = const()[name = string("input_1035_mode_0"), val = string("constant")];
+            fp16 const_207_to_fp16 = const()[name = string("const_207_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_1035_cast_fp16 = pad(constant_val = const_207_to_fp16, mode = input_1035_mode_0, pad = input_1035_pad_0, x = input_1033_cast_fp16)[name = string("input_1035_cast_fp16")];
+            string input_1037_pad_type_0 = const()[name = string("input_1037_pad_type_0"), val = string("valid")];
+            int32 input_1037_groups_0 = const()[name = string("input_1037_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_1037_strides_0 = const()[name = string("input_1037_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1037_pad_0 = const()[name = string("input_1037_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1037_dilations_0 = const()[name = string("input_1037_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_286_to_fp16 = const()[name = string("const_286_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252475136)))];
+            tensor<fp16, [1024]> const_287_to_fp16 = const()[name = string("const_287_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252493632)))];
+            tensor<fp16, [1, 1024, 188]> input_1039_cast_fp16 = conv(bias = const_287_to_fp16, dilations = input_1037_dilations_0, groups = input_1037_groups_0, pad = input_1037_pad_0, pad_type = input_1037_pad_type_0, strides = input_1037_strides_0, weight = const_286_to_fp16, x = input_1035_cast_fp16)[name = string("input_1039_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_1041_cast_fp16 = silu(x = input_1039_cast_fp16)[name = string("input_1041_cast_fp16")];
+            string x_439_pad_type_0 = const()[name = string("x_439_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_439_strides_0 = const()[name = string("x_439_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_439_pad_0 = const()[name = string("x_439_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_439_dilations_0 = const()[name = string("x_439_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_439_groups_0 = const()[name = string("x_439_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519548]> module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252495744))), indices_nonzero_data = tensor<uint6, [519548]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252626880))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253016640))))[name = string("module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_439_cast_fp16 = conv(dilations = x_439_dilations_0, groups = x_439_groups_0, pad = x_439_pad_0, pad_type = x_439_pad_type_0, strides = x_439_strides_0, weight = module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_1041_cast_fp16)[name = string("x_439_cast_fp16")];
+            tensor<int32, [3]> input_1043_perm_0 = const()[name = string("input_1043_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_1043_cast_fp16 = transpose(perm = input_1043_perm_0, x = x_439_cast_fp16)[name = string("transpose_173")];
+            tensor<fp16, [1, 188, 1024]> input_1045_cast_fp16 = add(x = input_1027_cast_fp16, y = input_1043_cast_fp16)[name = string("input_1045_cast_fp16")];
+            tensor<int32, [1]> input_1047_axes_0 = const()[name = string("input_1047_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_19_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_19_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253033088)))];
+            tensor<fp16, [1024]> module_layers_19_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_19_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253035200)))];
+            tensor<fp16, [1, 188, 1024]> input_1047_cast_fp16 = layer_norm(axes = input_1047_axes_0, beta = module_layers_19_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_feed_forward2_weight_to_fp16, x = input_1045_cast_fp16)[name = string("input_1047_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2091309]> module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253037312))), indices_nonzero_data = tensor<uint6, [2091309]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253561664))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(255130240))))[name = string("module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_179_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_1047_cast_fp16)[name = string("linear_179_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_1051_cast_fp16 = silu(x = linear_179_cast_fp16)[name = string("input_1051_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2085070]> module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(255195840))), indices_nonzero_data = tensor<uint6, [2085070]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(255720192))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257284096))))[name = string("module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_180_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_1051_cast_fp16)[name = string("linear_180_cast_fp16")];
+            fp16 var_3463_to_fp16 = const()[name = string("op_3463_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_3464_cast_fp16 = mul(x = linear_180_cast_fp16, y = var_3463_to_fp16)[name = string("op_3464_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_1057_cast_fp16 = add(x = input_1045_cast_fp16, y = var_3464_cast_fp16)[name = string("input_1057_cast_fp16")];
+            tensor<int32, [1]> input_1059_axes_0 = const()[name = string("input_1059_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_19_norm_out_weight_to_fp16 = const()[name = string("module_layers_19_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257300544)))];
+            tensor<fp16, [1024]> module_layers_19_norm_out_bias_to_fp16 = const()[name = string("module_layers_19_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257302656)))];
+            tensor<fp16, [1, 188, 1024]> input_1059_cast_fp16 = layer_norm(axes = input_1059_axes_0, beta = module_layers_19_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_out_weight_to_fp16, x = input_1057_cast_fp16)[name = string("input_1059_cast_fp16")];
+            tensor<int32, [1]> input_1061_axes_0 = const()[name = string("input_1061_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_20_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_20_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257304768)))];
+            tensor<fp16, [1024]> module_layers_20_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_20_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257306880)))];
+            tensor<fp16, [1, 188, 1024]> input_1061_cast_fp16 = layer_norm(axes = input_1061_axes_0, beta = module_layers_20_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_feed_forward1_weight_to_fp16, x = input_1059_cast_fp16)[name = string("input_1061_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2088569]> module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257308992))), indices_nonzero_data = tensor<uint6, [2088569]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257833344))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259399872))))[name = string("module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_181_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_1061_cast_fp16)[name = string("linear_181_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_1065_cast_fp16 = silu(x = linear_181_cast_fp16)[name = string("input_1065_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2053925]> module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259465472))), indices_nonzero_data = tensor<uint6, [2053925]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259989824))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261530368))))[name = string("module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_182_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_1065_cast_fp16)[name = string("linear_182_cast_fp16")];
+            fp16 var_3492_to_fp16 = const()[name = string("op_3492_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_3493_cast_fp16 = mul(x = linear_182_cast_fp16, y = var_3492_to_fp16)[name = string("op_3493_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_1071_cast_fp16 = add(x = input_1059_cast_fp16, y = var_3493_cast_fp16)[name = string("input_1071_cast_fp16")];
+            tensor<int32, [1]> query_41_axes_0 = const()[name = string("query_41_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_20_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_20_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261546816)))];
+            tensor<fp16, [1024]> module_layers_20_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_20_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261548928)))];
+            tensor<fp16, [1, 188, 1024]> query_41_cast_fp16 = layer_norm(axes = query_41_axes_0, beta = module_layers_20_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_self_att_weight_to_fp16, x = input_1071_cast_fp16)[name = string("query_41_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525329]> module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261551040))), indices_nonzero_data = tensor<uint6, [525329]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261682176))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262076288))))[name = string("module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_183_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_41_cast_fp16)[name = string("linear_183_cast_fp16")];
+            tensor<int32, [4]> var_3509 = const()[name = string("op_3509"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_121_cast_fp16 = reshape(shape = var_3509, x = linear_183_cast_fp16)[name = string("q_121_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525107]> module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262092736))), indices_nonzero_data = tensor<uint6, [525107]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262223872))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262617792))))[name = string("module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_184_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_41_cast_fp16)[name = string("linear_184_cast_fp16")];
+            tensor<int32, [4]> var_3513 = const()[name = string("op_3513"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_81_cast_fp16 = reshape(shape = var_3513, x = linear_184_cast_fp16)[name = string("k_81_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525707]> module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262634240))), indices_nonzero_data = tensor<uint6, [525707]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262765376))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263159744))))[name = string("module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_185_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_41_cast_fp16)[name = string("linear_185_cast_fp16")];
+            tensor<int32, [4]> var_3517 = const()[name = string("op_3517"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_41_cast_fp16 = reshape(shape = var_3517, x = linear_185_cast_fp16)[name = string("v_41_cast_fp16")];
+            tensor<int32, [4]> value_43_perm_0 = const()[name = string("value_43_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_20_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_20_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263176192)))];
+            tensor<fp16, [1, 188, 8, 128]> var_3529_cast_fp16 = add(x = q_121_cast_fp16, y = module_layers_20_self_attn_pos_bias_u_to_fp16)[name = string("op_3529_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_20_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_20_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263178304)))];
+            tensor<fp16, [1, 188, 8, 128]> var_3531_cast_fp16 = add(x = q_121_cast_fp16, y = module_layers_20_self_attn_pos_bias_v_to_fp16)[name = string("op_3531_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_41_perm_0 = const()[name = string("q_with_bias_v_41_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_447_transpose_x_0 = const()[name = string("x_447_transpose_x_0"), val = bool(false)];
+            bool x_447_transpose_y_0 = const()[name = string("x_447_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_3533_to_fp16 = const()[name = string("op_3533_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(335319232)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_41_cast_fp16 = transpose(perm = q_with_bias_v_41_perm_0, x = var_3531_cast_fp16)[name = string("transpose_172")];
+            tensor<fp16, [1, 8, 188, 375]> x_447_cast_fp16 = matmul(transpose_x = x_447_transpose_x_0, transpose_y = x_447_transpose_y_0, x = q_with_bias_v_41_cast_fp16, y = var_3533_to_fp16)[name = string("x_447_cast_fp16")];
+            tensor<int32, [8]> x_449_pad_0 = const()[name = string("x_449_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_449_mode_0 = const()[name = string("x_449_mode_0"), val = string("constant")];
+            fp16 const_214_to_fp16 = const()[name = string("const_214_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_449_cast_fp16 = pad(constant_val = const_214_to_fp16, mode = x_449_mode_0, pad = x_449_pad_0, x = x_447_cast_fp16)[name = string("x_449_cast_fp16")];
+            tensor<int32, [4]> var_3541 = const()[name = string("op_3541"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_451_cast_fp16 = reshape(shape = var_3541, x = x_449_cast_fp16)[name = string("x_451_cast_fp16")];
+            tensor<int32, [4]> var_3545_begin_0 = const()[name = string("op_3545_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3545_end_0 = const()[name = string("op_3545_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_3545_end_mask_0 = const()[name = string("op_3545_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_3545_cast_fp16 = slice_by_index(begin = var_3545_begin_0, end = var_3545_end_0, end_mask = var_3545_end_mask_0, x = x_451_cast_fp16)[name = string("op_3545_cast_fp16")];
+            tensor<int32, [4]> var_3546 = const()[name = string("op_3546"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_81_cast_fp16 = reshape(shape = var_3546, x = var_3545_cast_fp16)[name = string("matrix_bd_81_cast_fp16")];
+            bool matrix_ac_41_transpose_x_0 = const()[name = string("matrix_ac_41_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_41_transpose_y_0 = const()[name = string("matrix_ac_41_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_136_perm_0 = const()[name = string("transpose_136_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_137_perm_0 = const()[name = string("transpose_137_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_137 = transpose(perm = transpose_137_perm_0, x = k_81_cast_fp16)[name = string("transpose_170")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_136 = transpose(perm = transpose_136_perm_0, x = var_3529_cast_fp16)[name = string("transpose_171")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_41_cast_fp16 = matmul(transpose_x = matrix_ac_41_transpose_x_0, transpose_y = matrix_ac_41_transpose_y_0, x = transpose_136, y = transpose_137)[name = string("matrix_ac_41_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_83_begin_0 = const()[name = string("matrix_bd_83_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_83_end_0 = const()[name = string("matrix_bd_83_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_83_end_mask_0 = const()[name = string("matrix_bd_83_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_83_cast_fp16 = slice_by_index(begin = matrix_bd_83_begin_0, end = matrix_bd_83_end_0, end_mask = matrix_bd_83_end_mask_0, x = matrix_bd_81_cast_fp16)[name = string("matrix_bd_83_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_3555_cast_fp16 = add(x = matrix_ac_41_cast_fp16, y = matrix_bd_83_cast_fp16)[name = string("op_3555_cast_fp16")];
+            fp16 _inversed_scores_81_y_0_to_fp16 = const()[name = string("_inversed_scores_81_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_81_cast_fp16 = mul(x = var_3555_cast_fp16, y = _inversed_scores_81_y_0_to_fp16)[name = string("_inversed_scores_81_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_83_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_81_cast_fp16, cond = mask_3)[name = string("scores_83_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_3561_cast_fp16 = softmax(axis = var_30, x = scores_83_cast_fp16)[name = string("op_3561_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_1073_cast_fp16 = select(a = var_11_to_fp16, b = var_3561_cast_fp16, cond = mask_3)[name = string("input_1073_cast_fp16")];
+            bool x_453_transpose_x_0 = const()[name = string("x_453_transpose_x_0"), val = bool(false)];
+            bool x_453_transpose_y_0 = const()[name = string("x_453_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_43_cast_fp16 = transpose(perm = value_43_perm_0, x = v_41_cast_fp16)[name = string("transpose_169")];
+            tensor<fp16, [1, 8, 188, 128]> x_453_cast_fp16 = matmul(transpose_x = x_453_transpose_x_0, transpose_y = x_453_transpose_y_0, x = input_1073_cast_fp16, y = value_43_cast_fp16)[name = string("x_453_cast_fp16")];
+            tensor<int32, [4]> var_3565_perm_0 = const()[name = string("op_3565_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3566 = const()[name = string("op_3566"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_3565_cast_fp16 = transpose(perm = var_3565_perm_0, x = x_453_cast_fp16)[name = string("transpose_168")];
+            tensor<fp16, [1, 188, 1024]> input_1075_cast_fp16 = reshape(shape = var_3566, x = var_3565_cast_fp16)[name = string("input_1075_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [526127]> module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263440576))), indices_nonzero_data = tensor<uint6, [526127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263571712))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263966400))))[name = string("module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_187_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_1075_cast_fp16)[name = string("linear_187_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_1079_cast_fp16 = add(x = input_1071_cast_fp16, y = linear_187_cast_fp16)[name = string("input_1079_cast_fp16")];
+            tensor<int32, [1]> x_457_axes_0 = const()[name = string("x_457_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_20_norm_conv_weight_to_fp16 = const()[name = string("module_layers_20_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263982848)))];
+            tensor<fp16, [1024]> module_layers_20_norm_conv_bias_to_fp16 = const()[name = string("module_layers_20_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263984960)))];
+            tensor<fp16, [1, 188, 1024]> x_457_cast_fp16 = layer_norm(axes = x_457_axes_0, beta = module_layers_20_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_conv_weight_to_fp16, x = input_1079_cast_fp16)[name = string("x_457_cast_fp16")];
+            tensor<int32, [3]> input_1081_perm_0 = const()[name = string("input_1081_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_1083_pad_type_0 = const()[name = string("input_1083_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_1083_strides_0 = const()[name = string("input_1083_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1083_pad_0 = const()[name = string("input_1083_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1083_dilations_0 = const()[name = string("input_1083_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_1083_groups_0 = const()[name = string("input_1083_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1049203]> module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263987072))), indices_nonzero_data = tensor<uint6, [1049203]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264249280))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265036288))))[name = string("module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_1081_cast_fp16 = transpose(perm = input_1081_perm_0, x = x_457_cast_fp16)[name = string("transpose_167")];
+            tensor<fp16, [1, 2048, 188]> input_1083_cast_fp16 = conv(dilations = input_1083_dilations_0, groups = input_1083_groups_0, pad = input_1083_pad_0, pad_type = input_1083_pad_type_0, strides = input_1083_strides_0, weight = module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_1081_cast_fp16)[name = string("input_1083_cast_fp16")];
+            int32 x_459_split_num_splits_0 = const()[name = string("x_459_split_num_splits_0"), val = int32(2)];
+            int32 x_459_split_axis_0 = const()[name = string("x_459_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_459_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_459_split_cast_fp16_1 = split(axis = x_459_split_axis_0, num_splits = x_459_split_num_splits_0, x = input_1083_cast_fp16)[name = string("x_459_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_459_split_1_sigmoid_cast_fp16 = sigmoid(x = x_459_split_cast_fp16_1)[name = string("x_459_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_459_cast_fp16 = mul(x = x_459_split_cast_fp16_0, y = x_459_split_1_sigmoid_cast_fp16)[name = string("x_459_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_1085_cast_fp16 = select(a = var_11_to_fp16, b = x_459_cast_fp16, cond = var_328)[name = string("input_1085_cast_fp16")];
+            tensor<int32, [6]> input_1087_pad_0 = const()[name = string("input_1087_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_1087_mode_0 = const()[name = string("input_1087_mode_0"), val = string("constant")];
+            fp16 const_217_to_fp16 = const()[name = string("const_217_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_1087_cast_fp16 = pad(constant_val = const_217_to_fp16, mode = input_1087_mode_0, pad = input_1087_pad_0, x = input_1085_cast_fp16)[name = string("input_1087_cast_fp16")];
+            string input_1089_pad_type_0 = const()[name = string("input_1089_pad_type_0"), val = string("valid")];
+            int32 input_1089_groups_0 = const()[name = string("input_1089_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_1089_strides_0 = const()[name = string("input_1089_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1089_pad_0 = const()[name = string("input_1089_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1089_dilations_0 = const()[name = string("input_1089_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_288_to_fp16 = const()[name = string("const_288_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265069120)))];
+            tensor<fp16, [1024]> const_289_to_fp16 = const()[name = string("const_289_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265087616)))];
+            tensor<fp16, [1, 1024, 188]> input_1091_cast_fp16 = conv(bias = const_289_to_fp16, dilations = input_1089_dilations_0, groups = input_1089_groups_0, pad = input_1089_pad_0, pad_type = input_1089_pad_type_0, strides = input_1089_strides_0, weight = const_288_to_fp16, x = input_1087_cast_fp16)[name = string("input_1091_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_1093_cast_fp16 = silu(x = input_1091_cast_fp16)[name = string("input_1093_cast_fp16")];
+            string x_461_pad_type_0 = const()[name = string("x_461_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_461_strides_0 = const()[name = string("x_461_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_461_pad_0 = const()[name = string("x_461_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_461_dilations_0 = const()[name = string("x_461_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_461_groups_0 = const()[name = string("x_461_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [518556]> module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265089728))), indices_nonzero_data = tensor<uint6, [518556]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265220864))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265609856))))[name = string("module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_461_cast_fp16 = conv(dilations = x_461_dilations_0, groups = x_461_groups_0, pad = x_461_pad_0, pad_type = x_461_pad_type_0, strides = x_461_strides_0, weight = module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_1093_cast_fp16)[name = string("x_461_cast_fp16")];
+            tensor<int32, [3]> input_1095_perm_0 = const()[name = string("input_1095_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_1095_cast_fp16 = transpose(perm = input_1095_perm_0, x = x_461_cast_fp16)[name = string("transpose_166")];
+            tensor<fp16, [1, 188, 1024]> input_1097_cast_fp16 = add(x = input_1079_cast_fp16, y = input_1095_cast_fp16)[name = string("input_1097_cast_fp16")];
+            tensor<int32, [1]> input_1099_axes_0 = const()[name = string("input_1099_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_20_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_20_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265626304)))];
+            tensor<fp16, [1024]> module_layers_20_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_20_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265628416)))];
+            tensor<fp16, [1, 188, 1024]> input_1099_cast_fp16 = layer_norm(axes = input_1099_axes_0, beta = module_layers_20_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_feed_forward2_weight_to_fp16, x = input_1097_cast_fp16)[name = string("input_1099_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2095351]> module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265630528))), indices_nonzero_data = tensor<uint6, [2095351]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(266154880))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267726464))))[name = string("module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_188_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_1099_cast_fp16)[name = string("linear_188_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_1103_cast_fp16 = silu(x = linear_188_cast_fp16)[name = string("input_1103_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2074717]> module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267792064))), indices_nonzero_data = tensor<uint6, [2074717]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268316416))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269872576))))[name = string("module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_189_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_1103_cast_fp16)[name = string("linear_189_cast_fp16")];
+            fp16 var_3626_to_fp16 = const()[name = string("op_3626_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_3627_cast_fp16 = mul(x = linear_189_cast_fp16, y = var_3626_to_fp16)[name = string("op_3627_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_1109_cast_fp16 = add(x = input_1097_cast_fp16, y = var_3627_cast_fp16)[name = string("input_1109_cast_fp16")];
+            tensor<int32, [1]> input_1111_axes_0 = const()[name = string("input_1111_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_20_norm_out_weight_to_fp16 = const()[name = string("module_layers_20_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269889024)))];
+            tensor<fp16, [1024]> module_layers_20_norm_out_bias_to_fp16 = const()[name = string("module_layers_20_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269891136)))];
+            tensor<fp16, [1, 188, 1024]> input_1111_cast_fp16 = layer_norm(axes = input_1111_axes_0, beta = module_layers_20_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_out_weight_to_fp16, x = input_1109_cast_fp16)[name = string("input_1111_cast_fp16")];
+            tensor<int32, [1]> input_1113_axes_0 = const()[name = string("input_1113_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_21_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_21_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269893248)))];
+            tensor<fp16, [1024]> module_layers_21_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_21_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269895360)))];
+            tensor<fp16, [1, 188, 1024]> input_1113_cast_fp16 = layer_norm(axes = input_1113_axes_0, beta = module_layers_21_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_feed_forward1_weight_to_fp16, x = input_1111_cast_fp16)[name = string("input_1113_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2093447]> module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269897472))), indices_nonzero_data = tensor<uint6, [2093447]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270421824))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271992000))))[name = string("module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_190_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_1113_cast_fp16)[name = string("linear_190_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_1117_cast_fp16 = silu(x = linear_190_cast_fp16)[name = string("input_1117_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2029278]> module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272057600))), indices_nonzero_data = tensor<uint6, [2029278]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272581952))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274104000))))[name = string("module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_191_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_1117_cast_fp16)[name = string("linear_191_cast_fp16")];
+            fp16 var_3655_to_fp16 = const()[name = string("op_3655_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_3656_cast_fp16 = mul(x = linear_191_cast_fp16, y = var_3655_to_fp16)[name = string("op_3656_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_1123_cast_fp16 = add(x = input_1111_cast_fp16, y = var_3656_cast_fp16)[name = string("input_1123_cast_fp16")];
+            tensor<int32, [1]> query_43_axes_0 = const()[name = string("query_43_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_21_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_21_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274120448)))];
+            tensor<fp16, [1024]> module_layers_21_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_21_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274122560)))];
+            tensor<fp16, [1, 188, 1024]> query_43_cast_fp16 = layer_norm(axes = query_43_axes_0, beta = module_layers_21_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_self_att_weight_to_fp16, x = input_1123_cast_fp16)[name = string("query_43_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525507]> module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274124672))), indices_nonzero_data = tensor<uint6, [525507]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274255808))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274650048))))[name = string("module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_192_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_43_cast_fp16)[name = string("linear_192_cast_fp16")];
+            tensor<int32, [4]> var_3672 = const()[name = string("op_3672"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_127_cast_fp16 = reshape(shape = var_3672, x = linear_192_cast_fp16)[name = string("q_127_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525093]> module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274666496))), indices_nonzero_data = tensor<uint6, [525093]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274797632))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275191552))))[name = string("module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_193_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_43_cast_fp16)[name = string("linear_193_cast_fp16")];
+            tensor<int32, [4]> var_3676 = const()[name = string("op_3676"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_85_cast_fp16 = reshape(shape = var_3676, x = linear_193_cast_fp16)[name = string("k_85_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525954]> module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275208000))), indices_nonzero_data = tensor<uint6, [525954]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275339136))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275733696))))[name = string("module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_194_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_43_cast_fp16)[name = string("linear_194_cast_fp16")];
+            tensor<int32, [4]> var_3680 = const()[name = string("op_3680"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_43_cast_fp16 = reshape(shape = var_3680, x = linear_194_cast_fp16)[name = string("v_43_cast_fp16")];
+            tensor<int32, [4]> value_45_perm_0 = const()[name = string("value_45_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_21_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_21_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275750144)))];
+            tensor<fp16, [1, 188, 8, 128]> var_3692_cast_fp16 = add(x = q_127_cast_fp16, y = module_layers_21_self_attn_pos_bias_u_to_fp16)[name = string("op_3692_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_21_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_21_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275752256)))];
+            tensor<fp16, [1, 188, 8, 128]> var_3694_cast_fp16 = add(x = q_127_cast_fp16, y = module_layers_21_self_attn_pos_bias_v_to_fp16)[name = string("op_3694_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_43_perm_0 = const()[name = string("q_with_bias_v_43_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_469_transpose_x_0 = const()[name = string("x_469_transpose_x_0"), val = bool(false)];
+            bool x_469_transpose_y_0 = const()[name = string("x_469_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_3696_to_fp16 = const()[name = string("op_3696_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(336087296)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_43_cast_fp16 = transpose(perm = q_with_bias_v_43_perm_0, x = var_3694_cast_fp16)[name = string("transpose_165")];
+            tensor<fp16, [1, 8, 188, 375]> x_469_cast_fp16 = matmul(transpose_x = x_469_transpose_x_0, transpose_y = x_469_transpose_y_0, x = q_with_bias_v_43_cast_fp16, y = var_3696_to_fp16)[name = string("x_469_cast_fp16")];
+            tensor<int32, [8]> x_471_pad_0 = const()[name = string("x_471_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_471_mode_0 = const()[name = string("x_471_mode_0"), val = string("constant")];
+            fp16 const_224_to_fp16 = const()[name = string("const_224_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_471_cast_fp16 = pad(constant_val = const_224_to_fp16, mode = x_471_mode_0, pad = x_471_pad_0, x = x_469_cast_fp16)[name = string("x_471_cast_fp16")];
+            tensor<int32, [4]> var_3704 = const()[name = string("op_3704"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_473_cast_fp16 = reshape(shape = var_3704, x = x_471_cast_fp16)[name = string("x_473_cast_fp16")];
+            tensor<int32, [4]> var_3708_begin_0 = const()[name = string("op_3708_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3708_end_0 = const()[name = string("op_3708_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_3708_end_mask_0 = const()[name = string("op_3708_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_3708_cast_fp16 = slice_by_index(begin = var_3708_begin_0, end = var_3708_end_0, end_mask = var_3708_end_mask_0, x = x_473_cast_fp16)[name = string("op_3708_cast_fp16")];
+            tensor<int32, [4]> var_3709 = const()[name = string("op_3709"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_85_cast_fp16 = reshape(shape = var_3709, x = var_3708_cast_fp16)[name = string("matrix_bd_85_cast_fp16")];
+            bool matrix_ac_43_transpose_x_0 = const()[name = string("matrix_ac_43_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_43_transpose_y_0 = const()[name = string("matrix_ac_43_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_138_perm_0 = const()[name = string("transpose_138_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_139_perm_0 = const()[name = string("transpose_139_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_139 = transpose(perm = transpose_139_perm_0, x = k_85_cast_fp16)[name = string("transpose_163")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_138 = transpose(perm = transpose_138_perm_0, x = var_3692_cast_fp16)[name = string("transpose_164")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_43_cast_fp16 = matmul(transpose_x = matrix_ac_43_transpose_x_0, transpose_y = matrix_ac_43_transpose_y_0, x = transpose_138, y = transpose_139)[name = string("matrix_ac_43_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_87_begin_0 = const()[name = string("matrix_bd_87_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_87_end_0 = const()[name = string("matrix_bd_87_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_87_end_mask_0 = const()[name = string("matrix_bd_87_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_87_cast_fp16 = slice_by_index(begin = matrix_bd_87_begin_0, end = matrix_bd_87_end_0, end_mask = matrix_bd_87_end_mask_0, x = matrix_bd_85_cast_fp16)[name = string("matrix_bd_87_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_3718_cast_fp16 = add(x = matrix_ac_43_cast_fp16, y = matrix_bd_87_cast_fp16)[name = string("op_3718_cast_fp16")];
+            fp16 _inversed_scores_85_y_0_to_fp16 = const()[name = string("_inversed_scores_85_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_85_cast_fp16 = mul(x = var_3718_cast_fp16, y = _inversed_scores_85_y_0_to_fp16)[name = string("_inversed_scores_85_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_87_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_85_cast_fp16, cond = mask_3)[name = string("scores_87_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_3724_cast_fp16 = softmax(axis = var_30, x = scores_87_cast_fp16)[name = string("op_3724_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_1125_cast_fp16 = select(a = var_11_to_fp16, b = var_3724_cast_fp16, cond = mask_3)[name = string("input_1125_cast_fp16")];
+            bool x_475_transpose_x_0 = const()[name = string("x_475_transpose_x_0"), val = bool(false)];
+            bool x_475_transpose_y_0 = const()[name = string("x_475_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_45_cast_fp16 = transpose(perm = value_45_perm_0, x = v_43_cast_fp16)[name = string("transpose_162")];
+            tensor<fp16, [1, 8, 188, 128]> x_475_cast_fp16 = matmul(transpose_x = x_475_transpose_x_0, transpose_y = x_475_transpose_y_0, x = input_1125_cast_fp16, y = value_45_cast_fp16)[name = string("x_475_cast_fp16")];
+            tensor<int32, [4]> var_3728_perm_0 = const()[name = string("op_3728_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3729 = const()[name = string("op_3729"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_3728_cast_fp16 = transpose(perm = var_3728_perm_0, x = x_475_cast_fp16)[name = string("transpose_161")];
+            tensor<fp16, [1, 188, 1024]> input_1127_cast_fp16 = reshape(shape = var_3729, x = var_3728_cast_fp16)[name = string("input_1127_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [526102]> module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276014528))), indices_nonzero_data = tensor<uint6, [526102]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276145664))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276540352))))[name = string("module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_196_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_1127_cast_fp16)[name = string("linear_196_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_1131_cast_fp16 = add(x = input_1123_cast_fp16, y = linear_196_cast_fp16)[name = string("input_1131_cast_fp16")];
+            tensor<int32, [1]> x_479_axes_0 = const()[name = string("x_479_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_21_norm_conv_weight_to_fp16 = const()[name = string("module_layers_21_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276556800)))];
+            tensor<fp16, [1024]> module_layers_21_norm_conv_bias_to_fp16 = const()[name = string("module_layers_21_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276558912)))];
+            tensor<fp16, [1, 188, 1024]> x_479_cast_fp16 = layer_norm(axes = x_479_axes_0, beta = module_layers_21_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_conv_weight_to_fp16, x = input_1131_cast_fp16)[name = string("x_479_cast_fp16")];
+            tensor<int32, [3]> input_1133_perm_0 = const()[name = string("input_1133_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_1135_pad_type_0 = const()[name = string("input_1135_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_1135_strides_0 = const()[name = string("input_1135_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1135_pad_0 = const()[name = string("input_1135_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1135_dilations_0 = const()[name = string("input_1135_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_1135_groups_0 = const()[name = string("input_1135_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1049320]> module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276561024))), indices_nonzero_data = tensor<uint6, [1049320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276823232))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277610304))))[name = string("module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_1133_cast_fp16 = transpose(perm = input_1133_perm_0, x = x_479_cast_fp16)[name = string("transpose_160")];
+            tensor<fp16, [1, 2048, 188]> input_1135_cast_fp16 = conv(dilations = input_1135_dilations_0, groups = input_1135_groups_0, pad = input_1135_pad_0, pad_type = input_1135_pad_type_0, strides = input_1135_strides_0, weight = module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_1133_cast_fp16)[name = string("input_1135_cast_fp16")];
+            int32 x_481_split_num_splits_0 = const()[name = string("x_481_split_num_splits_0"), val = int32(2)];
+            int32 x_481_split_axis_0 = const()[name = string("x_481_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_481_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_481_split_cast_fp16_1 = split(axis = x_481_split_axis_0, num_splits = x_481_split_num_splits_0, x = input_1135_cast_fp16)[name = string("x_481_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_481_split_1_sigmoid_cast_fp16 = sigmoid(x = x_481_split_cast_fp16_1)[name = string("x_481_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_481_cast_fp16 = mul(x = x_481_split_cast_fp16_0, y = x_481_split_1_sigmoid_cast_fp16)[name = string("x_481_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_1137_cast_fp16 = select(a = var_11_to_fp16, b = x_481_cast_fp16, cond = var_328)[name = string("input_1137_cast_fp16")];
+            tensor<int32, [6]> input_1139_pad_0 = const()[name = string("input_1139_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_1139_mode_0 = const()[name = string("input_1139_mode_0"), val = string("constant")];
+            fp16 const_227_to_fp16 = const()[name = string("const_227_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_1139_cast_fp16 = pad(constant_val = const_227_to_fp16, mode = input_1139_mode_0, pad = input_1139_pad_0, x = input_1137_cast_fp16)[name = string("input_1139_cast_fp16")];
+            string input_1141_pad_type_0 = const()[name = string("input_1141_pad_type_0"), val = string("valid")];
+            int32 input_1141_groups_0 = const()[name = string("input_1141_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_1141_strides_0 = const()[name = string("input_1141_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1141_pad_0 = const()[name = string("input_1141_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1141_dilations_0 = const()[name = string("input_1141_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_290_to_fp16 = const()[name = string("const_290_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277643136)))];
+            tensor<fp16, [1024]> const_291_to_fp16 = const()[name = string("const_291_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277661632)))];
+            tensor<fp16, [1, 1024, 188]> input_1143_cast_fp16 = conv(bias = const_291_to_fp16, dilations = input_1141_dilations_0, groups = input_1141_groups_0, pad = input_1141_pad_0, pad_type = input_1141_pad_type_0, strides = input_1141_strides_0, weight = const_290_to_fp16, x = input_1139_cast_fp16)[name = string("input_1143_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_1145_cast_fp16 = silu(x = input_1143_cast_fp16)[name = string("input_1145_cast_fp16")];
+            string x_483_pad_type_0 = const()[name = string("x_483_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_483_strides_0 = const()[name = string("x_483_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_483_pad_0 = const()[name = string("x_483_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_483_dilations_0 = const()[name = string("x_483_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_483_groups_0 = const()[name = string("x_483_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519430]> module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277663744))), indices_nonzero_data = tensor<uint6, [519430]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277794880))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278184576))))[name = string("module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_483_cast_fp16 = conv(dilations = x_483_dilations_0, groups = x_483_groups_0, pad = x_483_pad_0, pad_type = x_483_pad_type_0, strides = x_483_strides_0, weight = module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_1145_cast_fp16)[name = string("x_483_cast_fp16")];
+            tensor<int32, [3]> input_1147_perm_0 = const()[name = string("input_1147_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_1147_cast_fp16 = transpose(perm = input_1147_perm_0, x = x_483_cast_fp16)[name = string("transpose_159")];
+            tensor<fp16, [1, 188, 1024]> input_1149_cast_fp16 = add(x = input_1131_cast_fp16, y = input_1147_cast_fp16)[name = string("input_1149_cast_fp16")];
+            tensor<int32, [1]> input_1151_axes_0 = const()[name = string("input_1151_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_21_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_21_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278201024)))];
+            tensor<fp16, [1024]> module_layers_21_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_21_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278203136)))];
+            tensor<fp16, [1, 188, 1024]> input_1151_cast_fp16 = layer_norm(axes = input_1151_axes_0, beta = module_layers_21_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_feed_forward2_weight_to_fp16, x = input_1149_cast_fp16)[name = string("input_1151_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2100215]> module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278205248))), indices_nonzero_data = tensor<uint6, [2100215]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278729600))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280304832))))[name = string("module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_197_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_1151_cast_fp16)[name = string("linear_197_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_1155_cast_fp16 = silu(x = linear_197_cast_fp16)[name = string("input_1155_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2058868]> module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280370432))), indices_nonzero_data = tensor<uint6, [2058868]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280894784))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282439040))))[name = string("module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_198_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_1155_cast_fp16)[name = string("linear_198_cast_fp16")];
+            fp16 var_3789_to_fp16 = const()[name = string("op_3789_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_3790_cast_fp16 = mul(x = linear_198_cast_fp16, y = var_3789_to_fp16)[name = string("op_3790_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_1161_cast_fp16 = add(x = input_1149_cast_fp16, y = var_3790_cast_fp16)[name = string("input_1161_cast_fp16")];
+            tensor<int32, [1]> input_1163_axes_0 = const()[name = string("input_1163_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_21_norm_out_weight_to_fp16 = const()[name = string("module_layers_21_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282455488)))];
+            tensor<fp16, [1024]> module_layers_21_norm_out_bias_to_fp16 = const()[name = string("module_layers_21_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282457600)))];
+            tensor<fp16, [1, 188, 1024]> input_1163_cast_fp16 = layer_norm(axes = input_1163_axes_0, beta = module_layers_21_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_out_weight_to_fp16, x = input_1161_cast_fp16)[name = string("input_1163_cast_fp16")];
+            tensor<int32, [1]> input_1165_axes_0 = const()[name = string("input_1165_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_22_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_22_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282459712)))];
+            tensor<fp16, [1024]> module_layers_22_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_22_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282461824)))];
+            tensor<fp16, [1, 188, 1024]> input_1165_cast_fp16 = layer_norm(axes = input_1165_axes_0, beta = module_layers_22_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_feed_forward1_weight_to_fp16, x = input_1163_cast_fp16)[name = string("input_1165_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2098607]> module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282463936))), indices_nonzero_data = tensor<uint6, [2098607]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282988288))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284562368))))[name = string("module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_199_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_1165_cast_fp16)[name = string("linear_199_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_1169_cast_fp16 = silu(x = linear_199_cast_fp16)[name = string("input_1169_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2035005]> module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284627968))), indices_nonzero_data = tensor<uint6, [2035005]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285152320))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286678656))))[name = string("module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_200_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_1169_cast_fp16)[name = string("linear_200_cast_fp16")];
+            fp16 var_3818_to_fp16 = const()[name = string("op_3818_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_3819_cast_fp16 = mul(x = linear_200_cast_fp16, y = var_3818_to_fp16)[name = string("op_3819_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_1175_cast_fp16 = add(x = input_1163_cast_fp16, y = var_3819_cast_fp16)[name = string("input_1175_cast_fp16")];
+            tensor<int32, [1]> query_45_axes_0 = const()[name = string("query_45_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_22_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_22_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286695104)))];
+            tensor<fp16, [1024]> module_layers_22_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_22_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286697216)))];
+            tensor<fp16, [1, 188, 1024]> query_45_cast_fp16 = layer_norm(axes = query_45_axes_0, beta = module_layers_22_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_self_att_weight_to_fp16, x = input_1175_cast_fp16)[name = string("query_45_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [527072]> module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286699328))), indices_nonzero_data = tensor<uint6, [527072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286830464))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287225856))))[name = string("module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_201_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_45_cast_fp16)[name = string("linear_201_cast_fp16")];
+            tensor<int32, [4]> var_3835 = const()[name = string("op_3835"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_133_cast_fp16 = reshape(shape = var_3835, x = linear_201_cast_fp16)[name = string("q_133_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525242]> module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287242304))), indices_nonzero_data = tensor<uint6, [525242]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287373440))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287767488))))[name = string("module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_202_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_45_cast_fp16)[name = string("linear_202_cast_fp16")];
+            tensor<int32, [4]> var_3839 = const()[name = string("op_3839"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_89_cast_fp16 = reshape(shape = var_3839, x = linear_202_cast_fp16)[name = string("k_89_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523544]> module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287783936))), indices_nonzero_data = tensor<uint6, [523544]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287915072))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288307840))))[name = string("module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_203_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_45_cast_fp16)[name = string("linear_203_cast_fp16")];
+            tensor<int32, [4]> var_3843 = const()[name = string("op_3843"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_45_cast_fp16 = reshape(shape = var_3843, x = linear_203_cast_fp16)[name = string("v_45_cast_fp16")];
+            tensor<int32, [4]> value_47_perm_0 = const()[name = string("value_47_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_22_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_22_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288324288)))];
+            tensor<fp16, [1, 188, 8, 128]> var_3855_cast_fp16 = add(x = q_133_cast_fp16, y = module_layers_22_self_attn_pos_bias_u_to_fp16)[name = string("op_3855_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_22_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_22_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288326400)))];
+            tensor<fp16, [1, 188, 8, 128]> var_3857_cast_fp16 = add(x = q_133_cast_fp16, y = module_layers_22_self_attn_pos_bias_v_to_fp16)[name = string("op_3857_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_45_perm_0 = const()[name = string("q_with_bias_v_45_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_491_transpose_x_0 = const()[name = string("x_491_transpose_x_0"), val = bool(false)];
+            bool x_491_transpose_y_0 = const()[name = string("x_491_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_3859_to_fp16 = const()[name = string("op_3859_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(336855360)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_45_cast_fp16 = transpose(perm = q_with_bias_v_45_perm_0, x = var_3857_cast_fp16)[name = string("transpose_158")];
+            tensor<fp16, [1, 8, 188, 375]> x_491_cast_fp16 = matmul(transpose_x = x_491_transpose_x_0, transpose_y = x_491_transpose_y_0, x = q_with_bias_v_45_cast_fp16, y = var_3859_to_fp16)[name = string("x_491_cast_fp16")];
+            tensor<int32, [8]> x_493_pad_0 = const()[name = string("x_493_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_493_mode_0 = const()[name = string("x_493_mode_0"), val = string("constant")];
+            fp16 const_234_to_fp16 = const()[name = string("const_234_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_493_cast_fp16 = pad(constant_val = const_234_to_fp16, mode = x_493_mode_0, pad = x_493_pad_0, x = x_491_cast_fp16)[name = string("x_493_cast_fp16")];
+            tensor<int32, [4]> var_3867 = const()[name = string("op_3867"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_495_cast_fp16 = reshape(shape = var_3867, x = x_493_cast_fp16)[name = string("x_495_cast_fp16")];
+            tensor<int32, [4]> var_3871_begin_0 = const()[name = string("op_3871_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3871_end_0 = const()[name = string("op_3871_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_3871_end_mask_0 = const()[name = string("op_3871_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_3871_cast_fp16 = slice_by_index(begin = var_3871_begin_0, end = var_3871_end_0, end_mask = var_3871_end_mask_0, x = x_495_cast_fp16)[name = string("op_3871_cast_fp16")];
+            tensor<int32, [4]> var_3872 = const()[name = string("op_3872"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_89_cast_fp16 = reshape(shape = var_3872, x = var_3871_cast_fp16)[name = string("matrix_bd_89_cast_fp16")];
+            bool matrix_ac_45_transpose_x_0 = const()[name = string("matrix_ac_45_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_45_transpose_y_0 = const()[name = string("matrix_ac_45_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_140_perm_0 = const()[name = string("transpose_140_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_141_perm_0 = const()[name = string("transpose_141_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_141 = transpose(perm = transpose_141_perm_0, x = k_89_cast_fp16)[name = string("transpose_156")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_140 = transpose(perm = transpose_140_perm_0, x = var_3855_cast_fp16)[name = string("transpose_157")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_45_cast_fp16 = matmul(transpose_x = matrix_ac_45_transpose_x_0, transpose_y = matrix_ac_45_transpose_y_0, x = transpose_140, y = transpose_141)[name = string("matrix_ac_45_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_91_begin_0 = const()[name = string("matrix_bd_91_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_91_end_0 = const()[name = string("matrix_bd_91_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_91_end_mask_0 = const()[name = string("matrix_bd_91_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_91_cast_fp16 = slice_by_index(begin = matrix_bd_91_begin_0, end = matrix_bd_91_end_0, end_mask = matrix_bd_91_end_mask_0, x = matrix_bd_89_cast_fp16)[name = string("matrix_bd_91_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_3881_cast_fp16 = add(x = matrix_ac_45_cast_fp16, y = matrix_bd_91_cast_fp16)[name = string("op_3881_cast_fp16")];
+            fp16 _inversed_scores_89_y_0_to_fp16 = const()[name = string("_inversed_scores_89_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_89_cast_fp16 = mul(x = var_3881_cast_fp16, y = _inversed_scores_89_y_0_to_fp16)[name = string("_inversed_scores_89_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_91_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_89_cast_fp16, cond = mask_3)[name = string("scores_91_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_3887_cast_fp16 = softmax(axis = var_30, x = scores_91_cast_fp16)[name = string("op_3887_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_1177_cast_fp16 = select(a = var_11_to_fp16, b = var_3887_cast_fp16, cond = mask_3)[name = string("input_1177_cast_fp16")];
+            bool x_497_transpose_x_0 = const()[name = string("x_497_transpose_x_0"), val = bool(false)];
+            bool x_497_transpose_y_0 = const()[name = string("x_497_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_47_cast_fp16 = transpose(perm = value_47_perm_0, x = v_45_cast_fp16)[name = string("transpose_155")];
+            tensor<fp16, [1, 8, 188, 128]> x_497_cast_fp16 = matmul(transpose_x = x_497_transpose_x_0, transpose_y = x_497_transpose_y_0, x = input_1177_cast_fp16, y = value_47_cast_fp16)[name = string("x_497_cast_fp16")];
+            tensor<int32, [4]> var_3891_perm_0 = const()[name = string("op_3891_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3892 = const()[name = string("op_3892"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_3891_cast_fp16 = transpose(perm = var_3891_perm_0, x = x_497_cast_fp16)[name = string("transpose_154")];
+            tensor<fp16, [1, 188, 1024]> input_1179_cast_fp16 = reshape(shape = var_3892, x = var_3891_cast_fp16)[name = string("input_1179_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523989]> module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288588672))), indices_nonzero_data = tensor<uint6, [523989]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288719808))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289112896))))[name = string("module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_205_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_1179_cast_fp16)[name = string("linear_205_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_1183_cast_fp16 = add(x = input_1175_cast_fp16, y = linear_205_cast_fp16)[name = string("input_1183_cast_fp16")];
+            tensor<int32, [1]> x_501_axes_0 = const()[name = string("x_501_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_22_norm_conv_weight_to_fp16 = const()[name = string("module_layers_22_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289129344)))];
+            tensor<fp16, [1024]> module_layers_22_norm_conv_bias_to_fp16 = const()[name = string("module_layers_22_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289131456)))];
+            tensor<fp16, [1, 188, 1024]> x_501_cast_fp16 = layer_norm(axes = x_501_axes_0, beta = module_layers_22_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_conv_weight_to_fp16, x = input_1183_cast_fp16)[name = string("x_501_cast_fp16")];
+            tensor<int32, [3]> input_1185_perm_0 = const()[name = string("input_1185_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_1187_pad_type_0 = const()[name = string("input_1187_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_1187_strides_0 = const()[name = string("input_1187_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1187_pad_0 = const()[name = string("input_1187_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1187_dilations_0 = const()[name = string("input_1187_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_1187_groups_0 = const()[name = string("input_1187_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1048742]> module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289133568))), indices_nonzero_data = tensor<uint6, [1048742]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289395776))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290182400))))[name = string("module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_1185_cast_fp16 = transpose(perm = input_1185_perm_0, x = x_501_cast_fp16)[name = string("transpose_153")];
+            tensor<fp16, [1, 2048, 188]> input_1187_cast_fp16 = conv(dilations = input_1187_dilations_0, groups = input_1187_groups_0, pad = input_1187_pad_0, pad_type = input_1187_pad_type_0, strides = input_1187_strides_0, weight = module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_1185_cast_fp16)[name = string("input_1187_cast_fp16")];
+            int32 x_503_split_num_splits_0 = const()[name = string("x_503_split_num_splits_0"), val = int32(2)];
+            int32 x_503_split_axis_0 = const()[name = string("x_503_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_503_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_503_split_cast_fp16_1 = split(axis = x_503_split_axis_0, num_splits = x_503_split_num_splits_0, x = input_1187_cast_fp16)[name = string("x_503_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_503_split_1_sigmoid_cast_fp16 = sigmoid(x = x_503_split_cast_fp16_1)[name = string("x_503_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_503_cast_fp16 = mul(x = x_503_split_cast_fp16_0, y = x_503_split_1_sigmoid_cast_fp16)[name = string("x_503_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_1189_cast_fp16 = select(a = var_11_to_fp16, b = x_503_cast_fp16, cond = var_328)[name = string("input_1189_cast_fp16")];
+            tensor<int32, [6]> input_1191_pad_0 = const()[name = string("input_1191_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_1191_mode_0 = const()[name = string("input_1191_mode_0"), val = string("constant")];
+            fp16 const_237_to_fp16 = const()[name = string("const_237_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_1191_cast_fp16 = pad(constant_val = const_237_to_fp16, mode = input_1191_mode_0, pad = input_1191_pad_0, x = input_1189_cast_fp16)[name = string("input_1191_cast_fp16")];
+            string input_1193_pad_type_0 = const()[name = string("input_1193_pad_type_0"), val = string("valid")];
+            int32 input_1193_groups_0 = const()[name = string("input_1193_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_1193_strides_0 = const()[name = string("input_1193_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1193_pad_0 = const()[name = string("input_1193_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1193_dilations_0 = const()[name = string("input_1193_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_292_to_fp16 = const()[name = string("const_292_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290215232)))];
+            tensor<fp16, [1024]> const_293_to_fp16 = const()[name = string("const_293_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290233728)))];
+            tensor<fp16, [1, 1024, 188]> input_1195_cast_fp16 = conv(bias = const_293_to_fp16, dilations = input_1193_dilations_0, groups = input_1193_groups_0, pad = input_1193_pad_0, pad_type = input_1193_pad_type_0, strides = input_1193_strides_0, weight = const_292_to_fp16, x = input_1191_cast_fp16)[name = string("input_1195_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_1197_cast_fp16 = silu(x = input_1195_cast_fp16)[name = string("input_1197_cast_fp16")];
+            string x_505_pad_type_0 = const()[name = string("x_505_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_505_strides_0 = const()[name = string("x_505_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_505_pad_0 = const()[name = string("x_505_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_505_dilations_0 = const()[name = string("x_505_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_505_groups_0 = const()[name = string("x_505_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [518537]> module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290235840))), indices_nonzero_data = tensor<uint6, [518537]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290366976))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290755968))))[name = string("module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_505_cast_fp16 = conv(dilations = x_505_dilations_0, groups = x_505_groups_0, pad = x_505_pad_0, pad_type = x_505_pad_type_0, strides = x_505_strides_0, weight = module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_1197_cast_fp16)[name = string("x_505_cast_fp16")];
+            tensor<int32, [3]> input_1199_perm_0 = const()[name = string("input_1199_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_1199_cast_fp16 = transpose(perm = input_1199_perm_0, x = x_505_cast_fp16)[name = string("transpose_152")];
+            tensor<fp16, [1, 188, 1024]> input_1201_cast_fp16 = add(x = input_1183_cast_fp16, y = input_1199_cast_fp16)[name = string("input_1201_cast_fp16")];
+            tensor<int32, [1]> input_1203_axes_0 = const()[name = string("input_1203_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_22_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_22_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290772416)))];
+            tensor<fp16, [1024]> module_layers_22_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_22_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290774528)))];
+            tensor<fp16, [1, 188, 1024]> input_1203_cast_fp16 = layer_norm(axes = input_1203_axes_0, beta = module_layers_22_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_feed_forward2_weight_to_fp16, x = input_1201_cast_fp16)[name = string("input_1203_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2096859]> module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290776640))), indices_nonzero_data = tensor<uint6, [2096859]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(291300992))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292873728))))[name = string("module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_206_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_1203_cast_fp16)[name = string("linear_206_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_1207_cast_fp16 = silu(x = linear_206_cast_fp16)[name = string("input_1207_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1916371]> module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292939328))), indices_nonzero_data = tensor<uint6, [1916371]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293463680))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294901056))))[name = string("module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_207_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_1207_cast_fp16)[name = string("linear_207_cast_fp16")];
+            fp16 var_3952_to_fp16 = const()[name = string("op_3952_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_3953_cast_fp16 = mul(x = linear_207_cast_fp16, y = var_3952_to_fp16)[name = string("op_3953_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_1213_cast_fp16 = add(x = input_1201_cast_fp16, y = var_3953_cast_fp16)[name = string("input_1213_cast_fp16")];
+            tensor<int32, [1]> input_1215_axes_0 = const()[name = string("input_1215_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_22_norm_out_weight_to_fp16 = const()[name = string("module_layers_22_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294917504)))];
+            tensor<fp16, [1024]> module_layers_22_norm_out_bias_to_fp16 = const()[name = string("module_layers_22_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294919616)))];
+            tensor<fp16, [1, 188, 1024]> input_1215_cast_fp16 = layer_norm(axes = input_1215_axes_0, beta = module_layers_22_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_out_weight_to_fp16, x = input_1213_cast_fp16)[name = string("input_1215_cast_fp16")];
+            tensor<int32, [1]> input_1217_axes_0 = const()[name = string("input_1217_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_23_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_23_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294921728)))];
+            tensor<fp16, [1024]> module_layers_23_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_23_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294923840)))];
+            tensor<fp16, [1, 188, 1024]> input_1217_cast_fp16 = layer_norm(axes = input_1217_axes_0, beta = module_layers_23_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_feed_forward1_weight_to_fp16, x = input_1215_cast_fp16)[name = string("input_1217_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2102731]> module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294925952))), indices_nonzero_data = tensor<uint6, [2102731]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295450304))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297027456))))[name = string("module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_208_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_1217_cast_fp16)[name = string("linear_208_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_1221_cast_fp16 = silu(x = linear_208_cast_fp16)[name = string("input_1221_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1967096]> module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297093056))), indices_nonzero_data = tensor<uint6, [1967096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297617408))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299092800))))[name = string("module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_209_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_1221_cast_fp16)[name = string("linear_209_cast_fp16")];
+            fp16 var_3981_to_fp16 = const()[name = string("op_3981_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_3982_cast_fp16 = mul(x = linear_209_cast_fp16, y = var_3981_to_fp16)[name = string("op_3982_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_1227_cast_fp16 = add(x = input_1215_cast_fp16, y = var_3982_cast_fp16)[name = string("input_1227_cast_fp16")];
+            tensor<int32, [1]> query_axes_0 = const()[name = string("query_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_23_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_23_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299109248)))];
+            tensor<fp16, [1024]> module_layers_23_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_23_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299111360)))];
+            tensor<fp16, [1, 188, 1024]> query_cast_fp16 = layer_norm(axes = query_axes_0, beta = module_layers_23_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_self_att_weight_to_fp16, x = input_1227_cast_fp16)[name = string("query_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [526594]> module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299113472))), indices_nonzero_data = tensor<uint6, [526594]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299244608))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299639680))))[name = string("module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_210_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_cast_fp16)[name = string("linear_210_cast_fp16")];
+            tensor<int32, [4]> var_3998 = const()[name = string("op_3998"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> q_139_cast_fp16 = reshape(shape = var_3998, x = linear_210_cast_fp16)[name = string("q_139_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524871]> module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299656128))), indices_nonzero_data = tensor<uint6, [524871]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299787264))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300180992))))[name = string("module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_211_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_cast_fp16)[name = string("linear_211_cast_fp16")];
+            tensor<int32, [4]> var_4002 = const()[name = string("op_4002"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> k_93_cast_fp16 = reshape(shape = var_4002, x = linear_211_cast_fp16)[name = string("k_93_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525256]> module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300197440))), indices_nonzero_data = tensor<uint6, [525256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300328576))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300722624))))[name = string("module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_212_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_cast_fp16)[name = string("linear_212_cast_fp16")];
+            tensor<int32, [4]> var_4006 = const()[name = string("op_4006"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 188, 8, 128]> v_cast_fp16 = reshape(shape = var_4006, x = linear_212_cast_fp16)[name = string("v_cast_fp16")];
+            tensor<int32, [4]> value_perm_0 = const()[name = string("value_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_23_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_23_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300739072)))];
+            tensor<fp16, [1, 188, 8, 128]> var_4018_cast_fp16 = add(x = q_139_cast_fp16, y = module_layers_23_self_attn_pos_bias_u_to_fp16)[name = string("op_4018_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_23_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_23_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300741184)))];
+            tensor<fp16, [1, 188, 8, 128]> var_4020_cast_fp16 = add(x = q_139_cast_fp16, y = module_layers_23_self_attn_pos_bias_v_to_fp16)[name = string("op_4020_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_perm_0 = const()[name = string("q_with_bias_v_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_513_transpose_x_0 = const()[name = string("x_513_transpose_x_0"), val = bool(false)];
+            bool x_513_transpose_y_0 = const()[name = string("x_513_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 375]> var_4022_to_fp16 = const()[name = string("op_4022_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337623424)))];
+            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_cast_fp16 = transpose(perm = q_with_bias_v_perm_0, x = var_4020_cast_fp16)[name = string("transpose_151")];
+            tensor<fp16, [1, 8, 188, 375]> x_513_cast_fp16 = matmul(transpose_x = x_513_transpose_x_0, transpose_y = x_513_transpose_y_0, x = q_with_bias_v_cast_fp16, y = var_4022_to_fp16)[name = string("x_513_cast_fp16")];
+            tensor<int32, [8]> x_515_pad_0 = const()[name = string("x_515_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_515_mode_0 = const()[name = string("x_515_mode_0"), val = string("constant")];
+            fp16 const_244_to_fp16 = const()[name = string("const_244_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 188, 376]> x_515_cast_fp16 = pad(constant_val = const_244_to_fp16, mode = x_515_mode_0, pad = x_515_pad_0, x = x_513_cast_fp16)[name = string("x_515_cast_fp16")];
+            tensor<int32, [4]> var_4030 = const()[name = string("op_4030"), val = tensor<int32, [4]>([1, 8, -1, 188])];
+            tensor<fp16, [1, 8, 376, 188]> x_517_cast_fp16 = reshape(shape = var_4030, x = x_515_cast_fp16)[name = string("x_517_cast_fp16")];
+            tensor<int32, [4]> var_4034_begin_0 = const()[name = string("op_4034_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_4034_end_0 = const()[name = string("op_4034_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<bool, [4]> var_4034_end_mask_0 = const()[name = string("op_4034_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 375, 188]> var_4034_cast_fp16 = slice_by_index(begin = var_4034_begin_0, end = var_4034_end_0, end_mask = var_4034_end_mask_0, x = x_517_cast_fp16)[name = string("op_4034_cast_fp16")];
+            tensor<int32, [4]> var_4035 = const()[name = string("op_4035"), val = tensor<int32, [4]>([1, 8, 188, 375])];
+            tensor<fp16, [1, 8, 188, 375]> matrix_bd_93_cast_fp16 = reshape(shape = var_4035, x = var_4034_cast_fp16)[name = string("matrix_bd_93_cast_fp16")];
+            bool matrix_ac_transpose_x_0 = const()[name = string("matrix_ac_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_transpose_y_0 = const()[name = string("matrix_ac_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_142_perm_0 = const()[name = string("transpose_142_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_143_perm_0 = const()[name = string("transpose_143_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 188]> transpose_143 = transpose(perm = transpose_143_perm_0, x = k_93_cast_fp16)[name = string("transpose_149")];
+            tensor<fp16, [1, 8, 188, 128]> transpose_142 = transpose(perm = transpose_142_perm_0, x = var_4018_cast_fp16)[name = string("transpose_150")];
+            tensor<fp16, [1, 8, 188, 188]> matrix_ac_cast_fp16 = matmul(transpose_x = matrix_ac_transpose_x_0, transpose_y = matrix_ac_transpose_y_0, x = transpose_142, y = transpose_143)[name = string("matrix_ac_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_begin_0 = const()[name = string("matrix_bd_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_end_0 = const()[name = string("matrix_bd_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<bool, [4]> matrix_bd_end_mask_0 = const()[name = string("matrix_bd_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 188, 188]> matrix_bd_cast_fp16 = slice_by_index(begin = matrix_bd_begin_0, end = matrix_bd_end_0, end_mask = matrix_bd_end_mask_0, x = matrix_bd_93_cast_fp16)[name = string("matrix_bd_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_4044_cast_fp16 = add(x = matrix_ac_cast_fp16, y = matrix_bd_cast_fp16)[name = string("op_4044_cast_fp16")];
+            fp16 _inversed_scores_93_y_0_to_fp16 = const()[name = string("_inversed_scores_93_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_93_cast_fp16 = mul(x = var_4044_cast_fp16, y = _inversed_scores_93_y_0_to_fp16)[name = string("_inversed_scores_93_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> scores_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_93_cast_fp16, cond = mask_3)[name = string("scores_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> var_4050_cast_fp16 = softmax(axis = var_30, x = scores_cast_fp16)[name = string("op_4050_cast_fp16")];
+            tensor<fp16, [1, 8, 188, 188]> input_1229_cast_fp16 = select(a = var_11_to_fp16, b = var_4050_cast_fp16, cond = mask_3)[name = string("input_1229_cast_fp16")];
+            bool x_519_transpose_x_0 = const()[name = string("x_519_transpose_x_0"), val = bool(false)];
+            bool x_519_transpose_y_0 = const()[name = string("x_519_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 188, 128]> value_cast_fp16 = transpose(perm = value_perm_0, x = v_cast_fp16)[name = string("transpose_148")];
+            tensor<fp16, [1, 8, 188, 128]> x_519_cast_fp16 = matmul(transpose_x = x_519_transpose_x_0, transpose_y = x_519_transpose_y_0, x = input_1229_cast_fp16, y = value_cast_fp16)[name = string("x_519_cast_fp16")];
+            tensor<int32, [4]> var_4054_perm_0 = const()[name = string("op_4054_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_4055 = const()[name = string("op_4055"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 188, 8, 128]> var_4054_cast_fp16 = transpose(perm = var_4054_perm_0, x = x_519_cast_fp16)[name = string("transpose_147")];
+            tensor<fp16, [1, 188, 1024]> input_1231_cast_fp16 = reshape(shape = var_4055, x = var_4054_cast_fp16)[name = string("input_1231_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [527827]> module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301003456))), indices_nonzero_data = tensor<uint6, [527827]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301134592))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301530560))))[name = string("module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_214_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_1231_cast_fp16)[name = string("linear_214_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_1235_cast_fp16 = add(x = input_1227_cast_fp16, y = linear_214_cast_fp16)[name = string("input_1235_cast_fp16")];
+            tensor<int32, [1]> x_523_axes_0 = const()[name = string("x_523_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_23_norm_conv_weight_to_fp16 = const()[name = string("module_layers_23_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301547008)))];
+            tensor<fp16, [1024]> module_layers_23_norm_conv_bias_to_fp16 = const()[name = string("module_layers_23_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301549120)))];
+            tensor<fp16, [1, 188, 1024]> x_523_cast_fp16 = layer_norm(axes = x_523_axes_0, beta = module_layers_23_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_conv_weight_to_fp16, x = input_1235_cast_fp16)[name = string("x_523_cast_fp16")];
+            tensor<int32, [3]> input_1237_perm_0 = const()[name = string("input_1237_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_1239_pad_type_0 = const()[name = string("input_1239_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_1239_strides_0 = const()[name = string("input_1239_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1239_pad_0 = const()[name = string("input_1239_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1239_dilations_0 = const()[name = string("input_1239_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_1239_groups_0 = const()[name = string("input_1239_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1051649]> module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301551232))), indices_nonzero_data = tensor<uint6, [1051649]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301813440))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302602304))))[name = string("module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> input_1237_cast_fp16 = transpose(perm = input_1237_perm_0, x = x_523_cast_fp16)[name = string("transpose_146")];
+            tensor<fp16, [1, 2048, 188]> input_1239_cast_fp16 = conv(dilations = input_1239_dilations_0, groups = input_1239_groups_0, pad = input_1239_pad_0, pad_type = input_1239_pad_type_0, strides = input_1239_strides_0, weight = module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_1237_cast_fp16)[name = string("input_1239_cast_fp16")];
+            int32 x_525_split_num_splits_0 = const()[name = string("x_525_split_num_splits_0"), val = int32(2)];
+            int32 x_525_split_axis_0 = const()[name = string("x_525_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 188]> x_525_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_525_split_cast_fp16_1 = split(axis = x_525_split_axis_0, num_splits = x_525_split_num_splits_0, x = input_1239_cast_fp16)[name = string("x_525_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_525_split_1_sigmoid_cast_fp16 = sigmoid(x = x_525_split_cast_fp16_1)[name = string("x_525_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> x_525_cast_fp16 = mul(x = x_525_split_cast_fp16_0, y = x_525_split_1_sigmoid_cast_fp16)[name = string("x_525_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_1241_cast_fp16 = select(a = var_11_to_fp16, b = x_525_cast_fp16, cond = var_328)[name = string("input_1241_cast_fp16")];
+            tensor<int32, [6]> input_1243_pad_0 = const()[name = string("input_1243_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_1243_mode_0 = const()[name = string("input_1243_mode_0"), val = string("constant")];
+            fp16 const_247_to_fp16 = const()[name = string("const_247_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 196]> input_1243_cast_fp16 = pad(constant_val = const_247_to_fp16, mode = input_1243_mode_0, pad = input_1243_pad_0, x = input_1241_cast_fp16)[name = string("input_1243_cast_fp16")];
+            string input_1245_pad_type_0 = const()[name = string("input_1245_pad_type_0"), val = string("valid")];
+            int32 input_1245_groups_0 = const()[name = string("input_1245_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_1245_strides_0 = const()[name = string("input_1245_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1245_pad_0 = const()[name = string("input_1245_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1245_dilations_0 = const()[name = string("input_1245_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_294_to_fp16 = const()[name = string("const_294_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302635136)))];
+            tensor<fp16, [1024]> const_295_to_fp16 = const()[name = string("const_295_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302653632)))];
+            tensor<fp16, [1, 1024, 188]> input_1247_cast_fp16 = conv(bias = const_295_to_fp16, dilations = input_1245_dilations_0, groups = input_1245_groups_0, pad = input_1245_pad_0, pad_type = input_1245_pad_type_0, strides = input_1245_strides_0, weight = const_294_to_fp16, x = input_1243_cast_fp16)[name = string("input_1247_cast_fp16")];
+            tensor<fp16, [1, 1024, 188]> input_1249_cast_fp16 = silu(x = input_1247_cast_fp16)[name = string("input_1249_cast_fp16")];
+            string x_527_pad_type_0 = const()[name = string("x_527_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_527_strides_0 = const()[name = string("x_527_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_527_pad_0 = const()[name = string("x_527_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_527_dilations_0 = const()[name = string("x_527_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_527_groups_0 = const()[name = string("x_527_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519222]> module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302655744))), indices_nonzero_data = tensor<uint6, [519222]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302786880))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303176384))))[name = string("module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 188]> x_527_cast_fp16 = conv(dilations = x_527_dilations_0, groups = x_527_groups_0, pad = x_527_pad_0, pad_type = x_527_pad_type_0, strides = x_527_strides_0, weight = module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_1249_cast_fp16)[name = string("x_527_cast_fp16")];
+            tensor<int32, [3]> input_1251_perm_0 = const()[name = string("input_1251_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 188, 1024]> input_1251_cast_fp16 = transpose(perm = input_1251_perm_0, x = x_527_cast_fp16)[name = string("transpose_145")];
+            tensor<fp16, [1, 188, 1024]> input_1253_cast_fp16 = add(x = input_1235_cast_fp16, y = input_1251_cast_fp16)[name = string("input_1253_cast_fp16")];
+            tensor<int32, [1]> input_1255_axes_0 = const()[name = string("input_1255_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_23_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_23_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303192832)))];
+            tensor<fp16, [1024]> module_layers_23_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_23_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303194944)))];
+            tensor<fp16, [1, 188, 1024]> input_1255_cast_fp16 = layer_norm(axes = input_1255_axes_0, beta = module_layers_23_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_feed_forward2_weight_to_fp16, x = input_1253_cast_fp16)[name = string("input_1255_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2106054]> module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303197056))), indices_nonzero_data = tensor<uint6, [2106054]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303721408))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305301056))))[name = string("module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 4096]> linear_215_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_1255_cast_fp16)[name = string("linear_215_cast_fp16")];
+            tensor<fp16, [1, 188, 4096]> input_1259_cast_fp16 = silu(x = linear_215_cast_fp16)[name = string("input_1259_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2012639]> module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305366656))), indices_nonzero_data = tensor<uint6, [2012639]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305891008))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307400576))))[name = string("module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 188, 1024]> linear_216_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_1259_cast_fp16)[name = string("linear_216_cast_fp16")];
+            fp16 var_4115_to_fp16 = const()[name = string("op_4115_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 188, 1024]> var_4116_cast_fp16 = mul(x = linear_216_cast_fp16, y = var_4115_to_fp16)[name = string("op_4116_cast_fp16")];
+            tensor<fp16, [1, 188, 1024]> input_cast_fp16 = add(x = input_1253_cast_fp16, y = var_4116_cast_fp16)[name = string("input_cast_fp16")];
+            tensor<int32, [1]> audio_signal_axes_0 = const()[name = string("audio_signal_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_23_norm_out_weight_to_fp16 = const()[name = string("module_layers_23_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307417024)))];
+            tensor<fp16, [1024]> module_layers_23_norm_out_bias_to_fp16 = const()[name = string("module_layers_23_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307419136)))];
+            tensor<fp16, [1, 188, 1024]> audio_signal_cast_fp16 = layer_norm(axes = audio_signal_axes_0, beta = module_layers_23_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_out_weight_to_fp16, x = input_cast_fp16)[name = string("audio_signal_cast_fp16")];
+            tensor<int32, [3]> obj_1_perm_0 = const()[name = string("obj_1_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string obj_1_cast_fp16_to_fp32_dtype_0 = const()[name = string("obj_1_cast_fp16_to_fp32_dtype_0"), val = string("fp32")];
+            tensor<fp16, [1, 1024, 188]> obj_1_cast_fp16 = transpose(perm = obj_1_perm_0, x = audio_signal_cast_fp16)[name = string("transpose_144")];
+            tensor<fp32, [1, 1024, 188]> encoder = cast(dtype = obj_1_cast_fp16_to_fp32_dtype_0, x = obj_1_cast_fp16)[name = string("cast_0")];
+        } -> (encoder, encoder_length);
+    func enc_5040ms<ios18>(tensor<fp32, [1, 128, 505]> mel, tensor<int32, [1]> mel_length) {
+            int32 var_30 = const()[name = string("op_30"), val = int32(-1)];
+            tensor<int32, [3]> x_1_perm_0 = const()[name = string("x_1_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string mel_to_fp16_dtype_0 = const()[name = string("mel_to_fp16_dtype_0"), val = string("fp16")];
+            string var_86_to_fp16_dtype_0 = const()[name = string("op_86_to_fp16_dtype_0"), val = string("fp16")];
+            fp16 var_87_promoted_to_fp16 = const()[name = string("op_87_promoted_to_fp16"), val = fp16(-0x1p+0)];
+            tensor<fp16, [1]> mel_length_to_fp16 = cast(dtype = var_86_to_fp16_dtype_0, x = mel_length)[name = string("cast_3")];
+            tensor<fp16, [1]> var_88_cast_fp16 = add(x = mel_length_to_fp16, y = var_87_promoted_to_fp16)[name = string("op_88_cast_fp16")];
+            fp16 _inversed_90_y_0_to_fp16 = const()[name = string("_inversed_90_y_0_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1]> _inversed_90_cast_fp16 = mul(x = var_88_cast_fp16, y = _inversed_90_y_0_to_fp16)[name = string("_inversed_90_cast_fp16")];
+            fp16 var_91_to_fp16 = const()[name = string("op_91_to_fp16"), val = fp16(0x1p+0)];
+            tensor<fp16, [1]> lengths_1_cast_fp16 = add(x = _inversed_90_cast_fp16, y = var_91_to_fp16)[name = string("lengths_1_cast_fp16")];
+            tensor<fp16, [1]> lengths_3_cast_fp16 = floor(x = lengths_1_cast_fp16)[name = string("lengths_3_cast_fp16")];
+            fp16 var_95_promoted_to_fp16 = const()[name = string("op_95_promoted_to_fp16"), val = fp16(-0x1p+0)];
+            tensor<fp16, [1]> var_96_cast_fp16 = add(x = lengths_3_cast_fp16, y = var_95_promoted_to_fp16)[name = string("op_96_cast_fp16")];
+            fp16 _inversed_98_y_0_to_fp16 = const()[name = string("_inversed_98_y_0_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1]> _inversed_98_cast_fp16 = mul(x = var_96_cast_fp16, y = _inversed_98_y_0_to_fp16)[name = string("_inversed_98_cast_fp16")];
+            fp16 var_99_to_fp16 = const()[name = string("op_99_to_fp16"), val = fp16(0x1p+0)];
+            tensor<fp16, [1]> lengths_7_cast_fp16 = add(x = _inversed_98_cast_fp16, y = var_99_to_fp16)[name = string("lengths_7_cast_fp16")];
+            tensor<fp16, [1]> lengths_9_cast_fp16 = floor(x = lengths_7_cast_fp16)[name = string("lengths_9_cast_fp16")];
+            fp16 var_103_promoted_to_fp16 = const()[name = string("op_103_promoted_to_fp16"), val = fp16(-0x1p+0)];
+            tensor<fp16, [1]> var_104_cast_fp16 = add(x = lengths_9_cast_fp16, y = var_103_promoted_to_fp16)[name = string("op_104_cast_fp16")];
+            fp16 _inversed_106_y_0_to_fp16 = const()[name = string("_inversed_106_y_0_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1]> _inversed_106_cast_fp16 = mul(x = var_104_cast_fp16, y = _inversed_106_y_0_to_fp16)[name = string("_inversed_106_cast_fp16")];
+            fp16 var_107_to_fp16 = const()[name = string("op_107_to_fp16"), val = fp16(0x1p+0)];
+            tensor<fp16, [1]> lengths_13_cast_fp16 = add(x = _inversed_106_cast_fp16, y = var_107_to_fp16)[name = string("lengths_13_cast_fp16")];
+            tensor<fp16, [1]> lengths_cast_fp16 = floor(x = lengths_13_cast_fp16)[name = string("lengths_cast_fp16")];
+            tensor<int32, [1]> input_1_axes_0 = const()[name = string("input_1_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1, 128, 505]> mel_to_fp16 = cast(dtype = mel_to_fp16_dtype_0, x = mel)[name = string("cast_2")];
+            tensor<fp16, [1, 505, 128]> x_1_cast_fp16 = transpose(perm = x_1_perm_0, x = mel_to_fp16)[name = string("transpose_315")];
+            tensor<fp16, [1, 1, 505, 128]> input_1_cast_fp16 = expand_dims(axes = input_1_axes_0, x = x_1_cast_fp16)[name = string("input_1_cast_fp16")];
+            string input_3_pad_type_0 = const()[name = string("input_3_pad_type_0"), val = string("custom")];
+            tensor<int32, [4]> input_3_pad_0 = const()[name = string("input_3_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
+            tensor<int32, [2]> input_3_strides_0 = const()[name = string("input_3_strides_0"), val = tensor<int32, [2]>([2, 2])];
+            tensor<int32, [2]> input_3_dilations_0 = const()[name = string("input_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
+            int32 input_3_groups_0 = const()[name = string("input_3_groups_0"), val = int32(1)];
+            tensor<fp16, [256, 1, 3, 3]> module_pre_encode_conv_0_weight_to_fp16 = const()[name = string("module_pre_encode_conv_0_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
+            tensor<fp16, [256]> module_pre_encode_conv_0_bias_to_fp16 = const()[name = string("module_pre_encode_conv_0_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4736)))];
+            tensor<fp16, [1, 256, 253, 64]> input_3_cast_fp16 = conv(bias = module_pre_encode_conv_0_bias_to_fp16, dilations = input_3_dilations_0, groups = input_3_groups_0, pad = input_3_pad_0, pad_type = input_3_pad_type_0, strides = input_3_strides_0, weight = module_pre_encode_conv_0_weight_to_fp16, x = input_1_cast_fp16)[name = string("input_3_cast_fp16")];
+            tensor<fp16, [1, 256, 253, 64]> input_5_cast_fp16 = relu(x = input_3_cast_fp16)[name = string("input_5_cast_fp16")];
+            string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("custom")];
+            tensor<int32, [4]> input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
+            tensor<int32, [2]> input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor<int32, [2]>([2, 2])];
+            int32 input_7_groups_0 = const()[name = string("input_7_groups_0"), val = int32(256)];
+            tensor<int32, [2]> input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<fp16, [256, 1, 3, 3]> module_pre_encode_conv_2_weight_to_fp16 = const()[name = string("module_pre_encode_conv_2_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5312)))];
+            tensor<fp16, [256]> module_pre_encode_conv_2_bias_to_fp16 = const()[name = string("module_pre_encode_conv_2_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9984)))];
+            tensor<fp16, [1, 256, 127, 32]> input_7_cast_fp16 = conv(bias = module_pre_encode_conv_2_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = module_pre_encode_conv_2_weight_to_fp16, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")];
+            string input_9_pad_type_0 = const()[name = string("input_9_pad_type_0"), val = string("valid")];
+            tensor<int32, [2]> input_9_strides_0 = const()[name = string("input_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [4]> input_9_pad_0 = const()[name = string("input_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [2]> input_9_dilations_0 = const()[name = string("input_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
+            int32 input_9_groups_0 = const()[name = string("input_9_groups_0"), val = int32(1)];
+            tensor<fp16, [256, 256, 1, 1]> module_pre_encode_conv_3_weight_to_fp16 = const()[name = string("module_pre_encode_conv_3_weight_to_fp16"), val = tensor<fp16, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10560)))];
+            tensor<fp16, [256]> module_pre_encode_conv_3_bias_to_fp16 = const()[name = string("module_pre_encode_conv_3_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141696)))];
+            tensor<fp16, [1, 256, 127, 32]> input_9_cast_fp16 = conv(bias = module_pre_encode_conv_3_bias_to_fp16, dilations = input_9_dilations_0, groups = input_9_groups_0, pad = input_9_pad_0, pad_type = input_9_pad_type_0, strides = input_9_strides_0, weight = module_pre_encode_conv_3_weight_to_fp16, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")];
+            tensor<fp16, [1, 256, 127, 32]> input_11_cast_fp16 = relu(x = input_9_cast_fp16)[name = string("input_11_cast_fp16")];
+            string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("custom")];
+            tensor<int32, [4]> input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
+            tensor<int32, [2]> input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor<int32, [2]>([2, 2])];
+            int32 input_13_groups_0 = const()[name = string("input_13_groups_0"), val = int32(256)];
+            tensor<int32, [2]> input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<fp16, [256, 1, 3, 3]> module_pre_encode_conv_5_weight_to_fp16 = const()[name = string("module_pre_encode_conv_5_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142272)))];
+            tensor<fp16, [256]> module_pre_encode_conv_5_bias_to_fp16 = const()[name = string("module_pre_encode_conv_5_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146944)))];
+            tensor<fp16, [1, 256, 64, 16]> input_13_cast_fp16 = conv(bias = module_pre_encode_conv_5_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = module_pre_encode_conv_5_weight_to_fp16, x = input_11_cast_fp16)[name = string("input_13_cast_fp16")];
+            string input_15_pad_type_0 = const()[name = string("input_15_pad_type_0"), val = string("valid")];
+            tensor<int32, [2]> input_15_strides_0 = const()[name = string("input_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [4]> input_15_pad_0 = const()[name = string("input_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [2]> input_15_dilations_0 = const()[name = string("input_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
+            int32 input_15_groups_0 = const()[name = string("input_15_groups_0"), val = int32(1)];
+            tensor<fp16, [256, 256, 1, 1]> module_pre_encode_conv_6_weight_to_fp16 = const()[name = string("module_pre_encode_conv_6_weight_to_fp16"), val = tensor<fp16, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147520)))];
+            tensor<fp16, [256]> module_pre_encode_conv_6_bias_to_fp16 = const()[name = string("module_pre_encode_conv_6_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278656)))];
+            tensor<fp16, [1, 256, 64, 16]> input_15_cast_fp16 = conv(bias = module_pre_encode_conv_6_bias_to_fp16, dilations = input_15_dilations_0, groups = input_15_groups_0, pad = input_15_pad_0, pad_type = input_15_pad_type_0, strides = input_15_strides_0, weight = module_pre_encode_conv_6_weight_to_fp16, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")];
+            tensor<fp16, [1, 256, 64, 16]> x_3_cast_fp16 = relu(x = input_15_cast_fp16)[name = string("x_3_cast_fp16")];
+            tensor<int32, [4]> var_157_perm_0 = const()[name = string("op_157_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_158 = const()[name = string("op_158"), val = tensor<int32, [3]>([1, 64, -1])];
+            tensor<fp16, [1, 64, 256, 16]> var_157_cast_fp16 = transpose(perm = var_157_perm_0, x = x_3_cast_fp16)[name = string("transpose_314")];
+            tensor<fp16, [1, 64, 4096]> input_17_cast_fp16 = reshape(shape = var_158, x = var_157_cast_fp16)[name = string("input_17_cast_fp16")];
+            tensor<fp16, [1024, 4096]> module_pre_encode_out_weight_to_fp16 = const()[name = string("module_pre_encode_out_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279232)))];
+            tensor<fp16, [1024]> module_pre_encode_out_bias_to_fp16 = const()[name = string("module_pre_encode_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8667904)))];
+            tensor<fp16, [1, 64, 1024]> linear_0_cast_fp16 = linear(bias = module_pre_encode_out_bias_to_fp16, weight = module_pre_encode_out_weight_to_fp16, x = input_17_cast_fp16)[name = string("linear_0_cast_fp16")];
+            string padding_length_dtype_0 = const()[name = string("padding_length_dtype_0"), val = string("int32")];
+            tensor<int32, [1, 64]> expand_dims_0 = const()[name = string("expand_dims_0"), val = tensor<int32, [1, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8670016)))];
+            tensor<int32, [1]> var_196_axes_0 = const()[name = string("op_196_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<int32, [1]> encoder_length = cast(dtype = padding_length_dtype_0, x = lengths_cast_fp16)[name = string("cast_1")];
+            tensor<int32, [1, 1]> var_196 = expand_dims(axes = var_196_axes_0, x = encoder_length)[name = string("op_196")];
+            tensor<bool, [1, 64]> pad_mask_1 = less(x = expand_dims_0, y = var_196)[name = string("pad_mask_1")];
+            tensor<int32, [1]> var_198_axes_0 = const()[name = string("op_198_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<bool, [1, 1, 64]> var_198 = expand_dims(axes = var_198_axes_0, x = pad_mask_1)[name = string("op_198")];
+            tensor<int32, [3]> var_199 = const()[name = string("op_199"), val = tensor<int32, [3]>([1, 64, 1])];
+            tensor<bool, [1, 64, 64]> pad_mask_for_att_mask_1 = tile(reps = var_199, x = var_198)[name = string("pad_mask_for_att_mask_1")];
+            tensor<int32, [3]> var_201_perm_0 = const()[name = string("op_201_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<bool, [1, 64, 64]> var_201 = transpose(perm = var_201_perm_0, x = pad_mask_for_att_mask_1)[name = string("transpose_313")];
+            tensor<bool, [1, 64, 64]> pad_mask_for_att_mask = logical_and(x = pad_mask_for_att_mask_1, y = var_201)[name = string("pad_mask_for_att_mask")];
+            tensor<bool, [1, 64, 64]> const_7 = const()[name = string("const_7"), val = tensor<bool, [1, 64, 64]>([[[true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true]]])];
+            tensor<bool, [1, 64, 64]> att_mask = logical_and(x = pad_mask_for_att_mask, y = const_7)[name = string("att_mask")];
+            tensor<bool, [1, 64, 64]> mask_1 = logical_not(x = att_mask)[name = string("mask_1")];
+            tensor<bool, [1, 64]> pad_mask = logical_not(x = pad_mask_1)[name = string("pad_mask")];
+            tensor<int32, [1]> input_21_axes_0 = const()[name = string("input_21_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_0_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_0_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8670336)))];
+            tensor<fp16, [1024]> module_layers_0_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_0_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8672448)))];
+            fp16 var_9_to_fp16 = const()[name = string("op_9_to_fp16"), val = fp16(0x1.5p-17)];
+            tensor<fp16, [1, 64, 1024]> input_21_cast_fp16 = layer_norm(axes = input_21_axes_0, beta = module_layers_0_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_feed_forward1_weight_to_fp16, x = linear_0_cast_fp16)[name = string("input_21_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2072390]> module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8674560))), indices_nonzero_data = tensor<uint6, [2072390]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9198912))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10753280))))[name = string("module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [4096]> linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10818880)))];
+            tensor<fp16, [1, 64, 4096]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_0_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_21_cast_fp16)[name = string("linear_1_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_25_cast_fp16 = silu(x = linear_1_cast_fp16)[name = string("input_25_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2038634]> module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10827136))), indices_nonzero_data = tensor<uint6, [2038634]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11351488))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12880576))))[name = string("module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1024]> linear_2_bias_0_to_fp16 = const()[name = string("linear_2_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12897024)))];
+            tensor<fp16, [1, 64, 1024]> linear_2_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_25_cast_fp16)[name = string("linear_2_cast_fp16")];
+            fp16 var_232_to_fp16 = const()[name = string("op_232_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_233_cast_fp16 = mul(x = linear_2_cast_fp16, y = var_232_to_fp16)[name = string("op_233_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_31_cast_fp16 = add(x = linear_0_cast_fp16, y = var_233_cast_fp16)[name = string("input_31_cast_fp16")];
+            tensor<int32, [1]> query_1_axes_0 = const()[name = string("query_1_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_0_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_0_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12899136)))];
+            tensor<fp16, [1024]> module_layers_0_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_0_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12901248)))];
+            tensor<fp16, [1, 64, 1024]> query_1_cast_fp16 = layer_norm(axes = query_1_axes_0, beta = module_layers_0_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_self_att_weight_to_fp16, x = input_31_cast_fp16)[name = string("query_1_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522455]> module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12903360))), indices_nonzero_data = tensor<uint6, [522455]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13034496))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13426432))))[name = string("module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_3_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_1_cast_fp16)[name = string("linear_3_cast_fp16")];
+            tensor<int32, [4]> var_249 = const()[name = string("op_249"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> q_1_cast_fp16 = reshape(shape = var_249, x = linear_3_cast_fp16)[name = string("q_1_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [515024]> module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13442880))), indices_nonzero_data = tensor<uint6, [515024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13574016))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13960384))))[name = string("module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_4_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_1_cast_fp16)[name = string("linear_4_cast_fp16")];
+            tensor<int32, [4]> var_253 = const()[name = string("op_253"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> k_1_cast_fp16 = reshape(shape = var_253, x = linear_4_cast_fp16)[name = string("k_1_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [505889]> module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13976832))), indices_nonzero_data = tensor<uint6, [505889]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14107968))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14487488))))[name = string("module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_5_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_1_cast_fp16)[name = string("linear_5_cast_fp16")];
+            tensor<int32, [4]> var_257 = const()[name = string("op_257"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> v_1_cast_fp16 = reshape(shape = var_257, x = linear_5_cast_fp16)[name = string("v_1_cast_fp16")];
+            tensor<int32, [4]> value_3_perm_0 = const()[name = string("value_3_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_0_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_0_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14503936)))];
+            tensor<fp16, [1, 64, 8, 128]> var_269_cast_fp16 = add(x = q_1_cast_fp16, y = module_layers_0_self_attn_pos_bias_u_to_fp16)[name = string("op_269_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_0_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_0_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14506048)))];
+            tensor<fp16, [1, 64, 8, 128]> var_271_cast_fp16 = add(x = q_1_cast_fp16, y = module_layers_0_self_attn_pos_bias_v_to_fp16)[name = string("op_271_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_1_perm_0 = const()[name = string("q_with_bias_v_1_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_7_transpose_x_0 = const()[name = string("x_7_transpose_x_0"), val = bool(false)];
+            bool x_7_transpose_y_0 = const()[name = string("x_7_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 127]> var_273_to_fp16 = const()[name = string("op_273_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14508160)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_1_cast_fp16 = transpose(perm = q_with_bias_v_1_perm_0, x = var_271_cast_fp16)[name = string("transpose_312")];
+            tensor<fp16, [1, 8, 64, 127]> x_7_cast_fp16 = matmul(transpose_x = x_7_transpose_x_0, transpose_y = x_7_transpose_y_0, x = q_with_bias_v_1_cast_fp16, y = var_273_to_fp16)[name = string("x_7_cast_fp16")];
+            tensor<int32, [8]> x_9_pad_0 = const()[name = string("x_9_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_9_mode_0 = const()[name = string("x_9_mode_0"), val = string("constant")];
+            fp16 const_14_to_fp16 = const()[name = string("const_14_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 64, 128]> x_9_cast_fp16 = pad(constant_val = const_14_to_fp16, mode = x_9_mode_0, pad = x_9_pad_0, x = x_7_cast_fp16)[name = string("x_9_cast_fp16")];
+            tensor<int32, [4]> var_281 = const()[name = string("op_281"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_11_cast_fp16 = reshape(shape = var_281, x = x_9_cast_fp16)[name = string("x_11_cast_fp16")];
+            tensor<int32, [4]> var_285_begin_0 = const()[name = string("op_285_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_285_end_0 = const()[name = string("op_285_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
+            tensor<bool, [4]> var_285_end_mask_0 = const()[name = string("op_285_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 127, 64]> var_285_cast_fp16 = slice_by_index(begin = var_285_begin_0, end = var_285_end_0, end_mask = var_285_end_mask_0, x = x_11_cast_fp16)[name = string("op_285_cast_fp16")];
+            tensor<int32, [4]> var_286 = const()[name = string("op_286"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_1_cast_fp16 = reshape(shape = var_286, x = var_285_cast_fp16)[name = string("matrix_bd_1_cast_fp16")];
+            bool matrix_ac_1_transpose_x_0 = const()[name = string("matrix_ac_1_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_1_transpose_y_0 = const()[name = string("matrix_ac_1_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_96_perm_0 = const()[name = string("transpose_96_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_97_perm_0 = const()[name = string("transpose_97_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 64]> transpose_97 = transpose(perm = transpose_97_perm_0, x = k_1_cast_fp16)[name = string("transpose_310")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_96 = transpose(perm = transpose_96_perm_0, x = var_269_cast_fp16)[name = string("transpose_311")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_1_cast_fp16 = matmul(transpose_x = matrix_ac_1_transpose_x_0, transpose_y = matrix_ac_1_transpose_y_0, x = transpose_96, y = transpose_97)[name = string("matrix_ac_1_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_3_begin_0 = const()[name = string("matrix_bd_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_3_end_0 = const()[name = string("matrix_bd_3_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
+            tensor<bool, [4]> matrix_bd_3_end_mask_0 = const()[name = string("matrix_bd_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_3_cast_fp16 = slice_by_index(begin = matrix_bd_3_begin_0, end = matrix_bd_3_end_0, end_mask = matrix_bd_3_end_mask_0, x = matrix_bd_1_cast_fp16)[name = string("matrix_bd_3_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_295_cast_fp16 = add(x = matrix_ac_1_cast_fp16, y = matrix_bd_3_cast_fp16)[name = string("op_295_cast_fp16")];
+            fp16 _inversed_scores_1_y_0_to_fp16 = const()[name = string("_inversed_scores_1_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_1_cast_fp16 = mul(x = var_295_cast_fp16, y = _inversed_scores_1_y_0_to_fp16)[name = string("_inversed_scores_1_cast_fp16")];
+            tensor<int32, [1]> mask_3_axes_0 = const()[name = string("mask_3_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<bool, [1, 1, 64, 64]> mask_3 = expand_dims(axes = mask_3_axes_0, x = mask_1)[name = string("mask_3")];
+            fp16 var_12_to_fp16 = const()[name = string("op_12_to_fp16"), val = fp16(-0x1.388p+13)];
+            tensor<fp16, [1, 8, 64, 64]> scores_3_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_1_cast_fp16, cond = mask_3)[name = string("scores_3_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_301_cast_fp16 = softmax(axis = var_30, x = scores_3_cast_fp16)[name = string("op_301_cast_fp16")];
+            fp16 var_11_to_fp16 = const()[name = string("op_11_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 64, 64]> input_33_cast_fp16 = select(a = var_11_to_fp16, b = var_301_cast_fp16, cond = mask_3)[name = string("input_33_cast_fp16")];
+            bool x_13_transpose_x_0 = const()[name = string("x_13_transpose_x_0"), val = bool(false)];
+            bool x_13_transpose_y_0 = const()[name = string("x_13_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 64, 128]> value_3_cast_fp16 = transpose(perm = value_3_perm_0, x = v_1_cast_fp16)[name = string("transpose_309")];
+            tensor<fp16, [1, 8, 64, 128]> x_13_cast_fp16 = matmul(transpose_x = x_13_transpose_x_0, transpose_y = x_13_transpose_y_0, x = input_33_cast_fp16, y = value_3_cast_fp16)[name = string("x_13_cast_fp16")];
+            tensor<int32, [4]> var_305_perm_0 = const()[name = string("op_305_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_306 = const()[name = string("op_306"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 64, 8, 128]> var_305_cast_fp16 = transpose(perm = var_305_perm_0, x = x_13_cast_fp16)[name = string("transpose_308")];
+            tensor<fp16, [1, 64, 1024]> input_35_cast_fp16 = reshape(shape = var_306, x = var_305_cast_fp16)[name = string("input_35_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [517310]> module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14768320))), indices_nonzero_data = tensor<uint6, [517310]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14899456))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15287552))))[name = string("module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_7_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_35_cast_fp16)[name = string("linear_7_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_39_cast_fp16 = add(x = input_31_cast_fp16, y = linear_7_cast_fp16)[name = string("input_39_cast_fp16")];
+            tensor<int32, [1]> x_17_axes_0 = const()[name = string("x_17_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_0_norm_conv_weight_to_fp16 = const()[name = string("module_layers_0_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15304000)))];
+            tensor<fp16, [1024]> module_layers_0_norm_conv_bias_to_fp16 = const()[name = string("module_layers_0_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15306112)))];
+            tensor<fp16, [1, 64, 1024]> x_17_cast_fp16 = layer_norm(axes = x_17_axes_0, beta = module_layers_0_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_conv_weight_to_fp16, x = input_39_cast_fp16)[name = string("x_17_cast_fp16")];
+            tensor<int32, [3]> input_41_perm_0 = const()[name = string("input_41_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_43_pad_type_0 = const()[name = string("input_43_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_43_strides_0 = const()[name = string("input_43_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_43_pad_0 = const()[name = string("input_43_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_43_dilations_0 = const()[name = string("input_43_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_43_groups_0 = const()[name = string("input_43_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1033148]> module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15308224))), indices_nonzero_data = tensor<uint6, [1033148]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15570432))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16345408))))[name = string("module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> input_41_cast_fp16 = transpose(perm = input_41_perm_0, x = x_17_cast_fp16)[name = string("transpose_307")];
+            tensor<fp16, [1, 2048, 64]> input_43_cast_fp16 = conv(dilations = input_43_dilations_0, groups = input_43_groups_0, pad = input_43_pad_0, pad_type = input_43_pad_type_0, strides = input_43_strides_0, weight = module_layers_0_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_41_cast_fp16)[name = string("input_43_cast_fp16")];
+            int32 x_19_split_num_splits_0 = const()[name = string("x_19_split_num_splits_0"), val = int32(2)];
+            int32 x_19_split_axis_0 = const()[name = string("x_19_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 64]> x_19_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_19_split_cast_fp16_1 = split(axis = x_19_split_axis_0, num_splits = x_19_split_num_splits_0, x = input_43_cast_fp16)[name = string("x_19_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_19_split_1_sigmoid_cast_fp16 = sigmoid(x = x_19_split_cast_fp16_1)[name = string("x_19_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_19_cast_fp16 = mul(x = x_19_split_cast_fp16_0, y = x_19_split_1_sigmoid_cast_fp16)[name = string("x_19_cast_fp16")];
+            tensor<int32, [1]> var_328_axes_0 = const()[name = string("op_328_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<bool, [1, 1, 64]> var_328 = expand_dims(axes = var_328_axes_0, x = pad_mask)[name = string("op_328")];
+            tensor<fp16, [1, 1024, 64]> input_45_cast_fp16 = select(a = var_11_to_fp16, b = x_19_cast_fp16, cond = var_328)[name = string("input_45_cast_fp16")];
+            tensor<int32, [6]> input_47_pad_0 = const()[name = string("input_47_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_47_mode_0 = const()[name = string("input_47_mode_0"), val = string("constant")];
+            fp16 const_17_to_fp16 = const()[name = string("const_17_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 72]> input_47_cast_fp16 = pad(constant_val = const_17_to_fp16, mode = input_47_mode_0, pad = input_47_pad_0, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")];
+            string input_49_pad_type_0 = const()[name = string("input_49_pad_type_0"), val = string("valid")];
+            int32 input_49_groups_0 = const()[name = string("input_49_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_49_strides_0 = const()[name = string("input_49_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_49_pad_0 = const()[name = string("input_49_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_49_dilations_0 = const()[name = string("input_49_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_248_to_fp16 = const()[name = string("const_248_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16378240)))];
+            tensor<fp16, [1024]> const_249_to_fp16 = const()[name = string("const_249_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16396736)))];
+            tensor<fp16, [1, 1024, 64]> input_51_cast_fp16 = conv(bias = const_249_to_fp16, dilations = input_49_dilations_0, groups = input_49_groups_0, pad = input_49_pad_0, pad_type = input_49_pad_type_0, strides = input_49_strides_0, weight = const_248_to_fp16, x = input_47_cast_fp16)[name = string("input_51_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_53_cast_fp16 = silu(x = input_51_cast_fp16)[name = string("input_53_cast_fp16")];
+            string x_21_pad_type_0 = const()[name = string("x_21_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_21_strides_0 = const()[name = string("x_21_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_21_pad_0 = const()[name = string("x_21_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_21_dilations_0 = const()[name = string("x_21_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_21_groups_0 = const()[name = string("x_21_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [512831]> module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16398848))), indices_nonzero_data = tensor<uint6, [512831]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16529984))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16914688))))[name = string("module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> x_21_cast_fp16 = conv(dilations = x_21_dilations_0, groups = x_21_groups_0, pad = x_21_pad_0, pad_type = x_21_pad_type_0, strides = x_21_strides_0, weight = module_layers_0_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_53_cast_fp16)[name = string("x_21_cast_fp16")];
+            tensor<int32, [3]> input_55_perm_0 = const()[name = string("input_55_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 64, 1024]> input_55_cast_fp16 = transpose(perm = input_55_perm_0, x = x_21_cast_fp16)[name = string("transpose_306")];
+            tensor<fp16, [1, 64, 1024]> input_57_cast_fp16 = add(x = input_39_cast_fp16, y = input_55_cast_fp16)[name = string("input_57_cast_fp16")];
+            tensor<int32, [1]> input_59_axes_0 = const()[name = string("input_59_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_0_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_0_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16931136)))];
+            tensor<fp16, [1024]> module_layers_0_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_0_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16933248)))];
+            tensor<fp16, [1, 64, 1024]> input_59_cast_fp16 = layer_norm(axes = input_59_axes_0, beta = module_layers_0_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_feed_forward2_weight_to_fp16, x = input_57_cast_fp16)[name = string("input_59_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2071218]> module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16935360))), indices_nonzero_data = tensor<uint6, [2071218]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17459712))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19013248))))[name = string("module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_8_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_0_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_59_cast_fp16)[name = string("linear_8_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_63_cast_fp16 = silu(x = linear_8_cast_fp16)[name = string("input_63_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2031949]> module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19078848))), indices_nonzero_data = tensor<uint6, [2031949]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19603200))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21127232))))[name = string("module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_9_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_0_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_63_cast_fp16)[name = string("linear_9_cast_fp16")];
+            fp16 var_366_to_fp16 = const()[name = string("op_366_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_367_cast_fp16 = mul(x = linear_9_cast_fp16, y = var_366_to_fp16)[name = string("op_367_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_69_cast_fp16 = add(x = input_57_cast_fp16, y = var_367_cast_fp16)[name = string("input_69_cast_fp16")];
+            tensor<int32, [1]> input_71_axes_0 = const()[name = string("input_71_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_0_norm_out_weight_to_fp16 = const()[name = string("module_layers_0_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21143680)))];
+            tensor<fp16, [1024]> module_layers_0_norm_out_bias_to_fp16 = const()[name = string("module_layers_0_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21145792)))];
+            tensor<fp16, [1, 64, 1024]> input_71_cast_fp16 = layer_norm(axes = input_71_axes_0, beta = module_layers_0_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_out_weight_to_fp16, x = input_69_cast_fp16)[name = string("input_71_cast_fp16")];
+            tensor<int32, [1]> input_73_axes_0 = const()[name = string("input_73_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_1_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_1_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21147904)))];
+            tensor<fp16, [1024]> module_layers_1_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_1_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21150016)))];
+            tensor<fp16, [1, 64, 1024]> input_73_cast_fp16 = layer_norm(axes = input_73_axes_0, beta = module_layers_1_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_feed_forward1_weight_to_fp16, x = input_71_cast_fp16)[name = string("input_73_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2039600]> module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21152128))), indices_nonzero_data = tensor<uint6, [2039600]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21676480))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23206272))))[name = string("module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_10_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_1_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_73_cast_fp16)[name = string("linear_10_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_77_cast_fp16 = silu(x = linear_10_cast_fp16)[name = string("input_77_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1944084]> module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23271872))), indices_nonzero_data = tensor<uint6, [1944084]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23796224))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25254400))))[name = string("module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_11_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_77_cast_fp16)[name = string("linear_11_cast_fp16")];
+            fp16 var_395_to_fp16 = const()[name = string("op_395_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_396_cast_fp16 = mul(x = linear_11_cast_fp16, y = var_395_to_fp16)[name = string("op_396_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_83_cast_fp16 = add(x = input_71_cast_fp16, y = var_396_cast_fp16)[name = string("input_83_cast_fp16")];
+            tensor<int32, [1]> query_3_axes_0 = const()[name = string("query_3_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_1_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_1_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25270848)))];
+            tensor<fp16, [1024]> module_layers_1_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_1_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25272960)))];
+            tensor<fp16, [1, 64, 1024]> query_3_cast_fp16 = layer_norm(axes = query_3_axes_0, beta = module_layers_1_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_self_att_weight_to_fp16, x = input_83_cast_fp16)[name = string("query_3_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523158]> module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25275072))), indices_nonzero_data = tensor<uint6, [523158]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25406208))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25798656))))[name = string("module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_12_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_3_cast_fp16)[name = string("linear_12_cast_fp16")];
+            tensor<int32, [4]> var_412 = const()[name = string("op_412"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> q_7_cast_fp16 = reshape(shape = var_412, x = linear_12_cast_fp16)[name = string("q_7_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [515603]> module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25815104))), indices_nonzero_data = tensor<uint6, [515603]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25946240))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26333056))))[name = string("module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_13_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_3_cast_fp16)[name = string("linear_13_cast_fp16")];
+            tensor<int32, [4]> var_416 = const()[name = string("op_416"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> k_5_cast_fp16 = reshape(shape = var_416, x = linear_13_cast_fp16)[name = string("k_5_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [511483]> module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26349504))), indices_nonzero_data = tensor<uint6, [511483]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26480640))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26864320))))[name = string("module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_14_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_3_cast_fp16)[name = string("linear_14_cast_fp16")];
+            tensor<int32, [4]> var_420 = const()[name = string("op_420"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> v_3_cast_fp16 = reshape(shape = var_420, x = linear_14_cast_fp16)[name = string("v_3_cast_fp16")];
+            tensor<int32, [4]> value_5_perm_0 = const()[name = string("value_5_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_1_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_1_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26880768)))];
+            tensor<fp16, [1, 64, 8, 128]> var_432_cast_fp16 = add(x = q_7_cast_fp16, y = module_layers_1_self_attn_pos_bias_u_to_fp16)[name = string("op_432_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_1_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_1_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26882880)))];
+            tensor<fp16, [1, 64, 8, 128]> var_434_cast_fp16 = add(x = q_7_cast_fp16, y = module_layers_1_self_attn_pos_bias_v_to_fp16)[name = string("op_434_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_3_perm_0 = const()[name = string("q_with_bias_v_3_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_29_transpose_x_0 = const()[name = string("x_29_transpose_x_0"), val = bool(false)];
+            bool x_29_transpose_y_0 = const()[name = string("x_29_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 127]> var_436_to_fp16 = const()[name = string("op_436_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26884992)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_3_cast_fp16 = transpose(perm = q_with_bias_v_3_perm_0, x = var_434_cast_fp16)[name = string("transpose_305")];
+            tensor<fp16, [1, 8, 64, 127]> x_29_cast_fp16 = matmul(transpose_x = x_29_transpose_x_0, transpose_y = x_29_transpose_y_0, x = q_with_bias_v_3_cast_fp16, y = var_436_to_fp16)[name = string("x_29_cast_fp16")];
+            tensor<int32, [8]> x_31_pad_0 = const()[name = string("x_31_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_31_mode_0 = const()[name = string("x_31_mode_0"), val = string("constant")];
+            fp16 const_24_to_fp16 = const()[name = string("const_24_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 64, 128]> x_31_cast_fp16 = pad(constant_val = const_24_to_fp16, mode = x_31_mode_0, pad = x_31_pad_0, x = x_29_cast_fp16)[name = string("x_31_cast_fp16")];
+            tensor<int32, [4]> var_444 = const()[name = string("op_444"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_33_cast_fp16 = reshape(shape = var_444, x = x_31_cast_fp16)[name = string("x_33_cast_fp16")];
+            tensor<int32, [4]> var_448_begin_0 = const()[name = string("op_448_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_448_end_0 = const()[name = string("op_448_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
+            tensor<bool, [4]> var_448_end_mask_0 = const()[name = string("op_448_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 127, 64]> var_448_cast_fp16 = slice_by_index(begin = var_448_begin_0, end = var_448_end_0, end_mask = var_448_end_mask_0, x = x_33_cast_fp16)[name = string("op_448_cast_fp16")];
+            tensor<int32, [4]> var_449 = const()[name = string("op_449"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_5_cast_fp16 = reshape(shape = var_449, x = var_448_cast_fp16)[name = string("matrix_bd_5_cast_fp16")];
+            bool matrix_ac_3_transpose_x_0 = const()[name = string("matrix_ac_3_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_3_transpose_y_0 = const()[name = string("matrix_ac_3_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_98_perm_0 = const()[name = string("transpose_98_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_99_perm_0 = const()[name = string("transpose_99_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 64]> transpose_99 = transpose(perm = transpose_99_perm_0, x = k_5_cast_fp16)[name = string("transpose_303")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_98 = transpose(perm = transpose_98_perm_0, x = var_432_cast_fp16)[name = string("transpose_304")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_3_cast_fp16 = matmul(transpose_x = matrix_ac_3_transpose_x_0, transpose_y = matrix_ac_3_transpose_y_0, x = transpose_98, y = transpose_99)[name = string("matrix_ac_3_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_7_begin_0 = const()[name = string("matrix_bd_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_7_end_0 = const()[name = string("matrix_bd_7_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
+            tensor<bool, [4]> matrix_bd_7_end_mask_0 = const()[name = string("matrix_bd_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_7_cast_fp16 = slice_by_index(begin = matrix_bd_7_begin_0, end = matrix_bd_7_end_0, end_mask = matrix_bd_7_end_mask_0, x = matrix_bd_5_cast_fp16)[name = string("matrix_bd_7_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_458_cast_fp16 = add(x = matrix_ac_3_cast_fp16, y = matrix_bd_7_cast_fp16)[name = string("op_458_cast_fp16")];
+            fp16 _inversed_scores_5_y_0_to_fp16 = const()[name = string("_inversed_scores_5_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_5_cast_fp16 = mul(x = var_458_cast_fp16, y = _inversed_scores_5_y_0_to_fp16)[name = string("_inversed_scores_5_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_7_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_5_cast_fp16, cond = mask_3)[name = string("scores_7_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_464_cast_fp16 = softmax(axis = var_30, x = scores_7_cast_fp16)[name = string("op_464_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_85_cast_fp16 = select(a = var_11_to_fp16, b = var_464_cast_fp16, cond = mask_3)[name = string("input_85_cast_fp16")];
+            bool x_35_transpose_x_0 = const()[name = string("x_35_transpose_x_0"), val = bool(false)];
+            bool x_35_transpose_y_0 = const()[name = string("x_35_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 64, 128]> value_5_cast_fp16 = transpose(perm = value_5_perm_0, x = v_3_cast_fp16)[name = string("transpose_302")];
+            tensor<fp16, [1, 8, 64, 128]> x_35_cast_fp16 = matmul(transpose_x = x_35_transpose_x_0, transpose_y = x_35_transpose_y_0, x = input_85_cast_fp16, y = value_5_cast_fp16)[name = string("x_35_cast_fp16")];
+            tensor<int32, [4]> var_468_perm_0 = const()[name = string("op_468_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_469 = const()[name = string("op_469"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 64, 8, 128]> var_468_cast_fp16 = transpose(perm = var_468_perm_0, x = x_35_cast_fp16)[name = string("transpose_301")];
+            tensor<fp16, [1, 64, 1024]> input_87_cast_fp16 = reshape(shape = var_469, x = var_468_cast_fp16)[name = string("input_87_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523005]> module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27145152))), indices_nonzero_data = tensor<uint6, [523005]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27276288))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27668608))))[name = string("module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_16_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_87_cast_fp16)[name = string("linear_16_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_91_cast_fp16 = add(x = input_83_cast_fp16, y = linear_16_cast_fp16)[name = string("input_91_cast_fp16")];
+            tensor<int32, [1]> x_39_axes_0 = const()[name = string("x_39_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_1_norm_conv_weight_to_fp16 = const()[name = string("module_layers_1_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27685056)))];
+            tensor<fp16, [1024]> module_layers_1_norm_conv_bias_to_fp16 = const()[name = string("module_layers_1_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27687168)))];
+            tensor<fp16, [1, 64, 1024]> x_39_cast_fp16 = layer_norm(axes = x_39_axes_0, beta = module_layers_1_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_conv_weight_to_fp16, x = input_91_cast_fp16)[name = string("x_39_cast_fp16")];
+            tensor<int32, [3]> input_93_perm_0 = const()[name = string("input_93_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_95_pad_type_0 = const()[name = string("input_95_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_95_strides_0 = const()[name = string("input_95_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_95_pad_0 = const()[name = string("input_95_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_95_dilations_0 = const()[name = string("input_95_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_95_groups_0 = const()[name = string("input_95_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1039540]> module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27689280))), indices_nonzero_data = tensor<uint6, [1039540]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27951488))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28731264))))[name = string("module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> input_93_cast_fp16 = transpose(perm = input_93_perm_0, x = x_39_cast_fp16)[name = string("transpose_300")];
+            tensor<fp16, [1, 2048, 64]> input_95_cast_fp16 = conv(dilations = input_95_dilations_0, groups = input_95_groups_0, pad = input_95_pad_0, pad_type = input_95_pad_type_0, strides = input_95_strides_0, weight = module_layers_1_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_93_cast_fp16)[name = string("input_95_cast_fp16")];
+            int32 x_41_split_num_splits_0 = const()[name = string("x_41_split_num_splits_0"), val = int32(2)];
+            int32 x_41_split_axis_0 = const()[name = string("x_41_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 64]> x_41_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_41_split_cast_fp16_1 = split(axis = x_41_split_axis_0, num_splits = x_41_split_num_splits_0, x = input_95_cast_fp16)[name = string("x_41_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_41_split_1_sigmoid_cast_fp16 = sigmoid(x = x_41_split_cast_fp16_1)[name = string("x_41_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_41_cast_fp16 = mul(x = x_41_split_cast_fp16_0, y = x_41_split_1_sigmoid_cast_fp16)[name = string("x_41_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_97_cast_fp16 = select(a = var_11_to_fp16, b = x_41_cast_fp16, cond = var_328)[name = string("input_97_cast_fp16")];
+            tensor<int32, [6]> input_99_pad_0 = const()[name = string("input_99_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_99_mode_0 = const()[name = string("input_99_mode_0"), val = string("constant")];
+            fp16 const_27_to_fp16 = const()[name = string("const_27_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 72]> input_99_cast_fp16 = pad(constant_val = const_27_to_fp16, mode = input_99_mode_0, pad = input_99_pad_0, x = input_97_cast_fp16)[name = string("input_99_cast_fp16")];
+            string input_101_pad_type_0 = const()[name = string("input_101_pad_type_0"), val = string("valid")];
+            int32 input_101_groups_0 = const()[name = string("input_101_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_101_strides_0 = const()[name = string("input_101_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_101_pad_0 = const()[name = string("input_101_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_101_dilations_0 = const()[name = string("input_101_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_250_to_fp16 = const()[name = string("const_250_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28764096)))];
+            tensor<fp16, [1024]> const_251_to_fp16 = const()[name = string("const_251_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28782592)))];
+            tensor<fp16, [1, 1024, 64]> input_103_cast_fp16 = conv(bias = const_251_to_fp16, dilations = input_101_dilations_0, groups = input_101_groups_0, pad = input_101_pad_0, pad_type = input_101_pad_type_0, strides = input_101_strides_0, weight = const_250_to_fp16, x = input_99_cast_fp16)[name = string("input_103_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_105_cast_fp16 = silu(x = input_103_cast_fp16)[name = string("input_105_cast_fp16")];
+            string x_43_pad_type_0 = const()[name = string("x_43_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_43_strides_0 = const()[name = string("x_43_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_43_pad_0 = const()[name = string("x_43_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_43_dilations_0 = const()[name = string("x_43_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_43_groups_0 = const()[name = string("x_43_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [515869]> module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28784704))), indices_nonzero_data = tensor<uint6, [515869]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28915840))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29302848))))[name = string("module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> x_43_cast_fp16 = conv(dilations = x_43_dilations_0, groups = x_43_groups_0, pad = x_43_pad_0, pad_type = x_43_pad_type_0, strides = x_43_strides_0, weight = module_layers_1_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_105_cast_fp16)[name = string("x_43_cast_fp16")];
+            tensor<int32, [3]> input_107_perm_0 = const()[name = string("input_107_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 64, 1024]> input_107_cast_fp16 = transpose(perm = input_107_perm_0, x = x_43_cast_fp16)[name = string("transpose_299")];
+            tensor<fp16, [1, 64, 1024]> input_109_cast_fp16 = add(x = input_91_cast_fp16, y = input_107_cast_fp16)[name = string("input_109_cast_fp16")];
+            tensor<int32, [1]> input_111_axes_0 = const()[name = string("input_111_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_1_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_1_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29319296)))];
+            tensor<fp16, [1024]> module_layers_1_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_1_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29321408)))];
+            tensor<fp16, [1, 64, 1024]> input_111_cast_fp16 = layer_norm(axes = input_111_axes_0, beta = module_layers_1_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_feed_forward2_weight_to_fp16, x = input_109_cast_fp16)[name = string("input_111_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2057265]> module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29323520))), indices_nonzero_data = tensor<uint6, [2057265]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29847872))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31390912))))[name = string("module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_17_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_1_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_111_cast_fp16)[name = string("linear_17_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_115_cast_fp16 = silu(x = linear_17_cast_fp16)[name = string("input_115_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1937635]> module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31456512))), indices_nonzero_data = tensor<uint6, [1937635]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31980864))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33434176))))[name = string("module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_18_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_1_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_115_cast_fp16)[name = string("linear_18_cast_fp16")];
+            fp16 var_529_to_fp16 = const()[name = string("op_529_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_530_cast_fp16 = mul(x = linear_18_cast_fp16, y = var_529_to_fp16)[name = string("op_530_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_121_cast_fp16 = add(x = input_109_cast_fp16, y = var_530_cast_fp16)[name = string("input_121_cast_fp16")];
+            tensor<int32, [1]> input_123_axes_0 = const()[name = string("input_123_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_1_norm_out_weight_to_fp16 = const()[name = string("module_layers_1_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33450624)))];
+            tensor<fp16, [1024]> module_layers_1_norm_out_bias_to_fp16 = const()[name = string("module_layers_1_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33452736)))];
+            tensor<fp16, [1, 64, 1024]> input_123_cast_fp16 = layer_norm(axes = input_123_axes_0, beta = module_layers_1_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_out_weight_to_fp16, x = input_121_cast_fp16)[name = string("input_123_cast_fp16")];
+            tensor<int32, [1]> input_125_axes_0 = const()[name = string("input_125_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_2_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_2_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33454848)))];
+            tensor<fp16, [1024]> module_layers_2_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_2_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33456960)))];
+            tensor<fp16, [1, 64, 1024]> input_125_cast_fp16 = layer_norm(axes = input_125_axes_0, beta = module_layers_2_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_feed_forward1_weight_to_fp16, x = input_123_cast_fp16)[name = string("input_125_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2079554]> module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33459072))), indices_nonzero_data = tensor<uint6, [2079554]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33983424))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35543168))))[name = string("module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_2_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_125_cast_fp16)[name = string("linear_19_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_129_cast_fp16 = silu(x = linear_19_cast_fp16)[name = string("input_129_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1998259]> module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35608768))), indices_nonzero_data = tensor<uint6, [1998259]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36133120))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37631936))))[name = string("module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_20_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_129_cast_fp16)[name = string("linear_20_cast_fp16")];
+            fp16 var_558_to_fp16 = const()[name = string("op_558_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_559_cast_fp16 = mul(x = linear_20_cast_fp16, y = var_558_to_fp16)[name = string("op_559_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_135_cast_fp16 = add(x = input_123_cast_fp16, y = var_559_cast_fp16)[name = string("input_135_cast_fp16")];
+            tensor<int32, [1]> query_5_axes_0 = const()[name = string("query_5_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_2_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_2_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37648384)))];
+            tensor<fp16, [1024]> module_layers_2_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_2_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37650496)))];
+            tensor<fp16, [1, 64, 1024]> query_5_cast_fp16 = layer_norm(axes = query_5_axes_0, beta = module_layers_2_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_self_att_weight_to_fp16, x = input_135_cast_fp16)[name = string("query_5_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522858]> module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37652608))), indices_nonzero_data = tensor<uint6, [522858]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37783744))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38176000))))[name = string("module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_21_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_5_cast_fp16)[name = string("linear_21_cast_fp16")];
+            tensor<int32, [4]> var_575 = const()[name = string("op_575"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> q_13_cast_fp16 = reshape(shape = var_575, x = linear_21_cast_fp16)[name = string("q_13_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521195]> module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38192448))), indices_nonzero_data = tensor<uint6, [521195]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38323584))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38714560))))[name = string("module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_22_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_5_cast_fp16)[name = string("linear_22_cast_fp16")];
+            tensor<int32, [4]> var_579 = const()[name = string("op_579"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> k_9_cast_fp16 = reshape(shape = var_579, x = linear_22_cast_fp16)[name = string("k_9_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521755]> module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38731008))), indices_nonzero_data = tensor<uint6, [521755]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38862144))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39253568))))[name = string("module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_23_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_5_cast_fp16)[name = string("linear_23_cast_fp16")];
+            tensor<int32, [4]> var_583 = const()[name = string("op_583"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> v_5_cast_fp16 = reshape(shape = var_583, x = linear_23_cast_fp16)[name = string("v_5_cast_fp16")];
+            tensor<int32, [4]> value_7_perm_0 = const()[name = string("value_7_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_2_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_2_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39270016)))];
+            tensor<fp16, [1, 64, 8, 128]> var_595_cast_fp16 = add(x = q_13_cast_fp16, y = module_layers_2_self_attn_pos_bias_u_to_fp16)[name = string("op_595_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_2_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_2_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39272128)))];
+            tensor<fp16, [1, 64, 8, 128]> var_597_cast_fp16 = add(x = q_13_cast_fp16, y = module_layers_2_self_attn_pos_bias_v_to_fp16)[name = string("op_597_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_5_perm_0 = const()[name = string("q_with_bias_v_5_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_51_transpose_x_0 = const()[name = string("x_51_transpose_x_0"), val = bool(false)];
+            bool x_51_transpose_y_0 = const()[name = string("x_51_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 127]> var_599_to_fp16 = const()[name = string("op_599_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39274240)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_5_cast_fp16 = transpose(perm = q_with_bias_v_5_perm_0, x = var_597_cast_fp16)[name = string("transpose_298")];
+            tensor<fp16, [1, 8, 64, 127]> x_51_cast_fp16 = matmul(transpose_x = x_51_transpose_x_0, transpose_y = x_51_transpose_y_0, x = q_with_bias_v_5_cast_fp16, y = var_599_to_fp16)[name = string("x_51_cast_fp16")];
+            tensor<int32, [8]> x_53_pad_0 = const()[name = string("x_53_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_53_mode_0 = const()[name = string("x_53_mode_0"), val = string("constant")];
+            fp16 const_34_to_fp16 = const()[name = string("const_34_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 64, 128]> x_53_cast_fp16 = pad(constant_val = const_34_to_fp16, mode = x_53_mode_0, pad = x_53_pad_0, x = x_51_cast_fp16)[name = string("x_53_cast_fp16")];
+            tensor<int32, [4]> var_607 = const()[name = string("op_607"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_55_cast_fp16 = reshape(shape = var_607, x = x_53_cast_fp16)[name = string("x_55_cast_fp16")];
+            tensor<int32, [4]> var_611_begin_0 = const()[name = string("op_611_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_611_end_0 = const()[name = string("op_611_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
+            tensor<bool, [4]> var_611_end_mask_0 = const()[name = string("op_611_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 127, 64]> var_611_cast_fp16 = slice_by_index(begin = var_611_begin_0, end = var_611_end_0, end_mask = var_611_end_mask_0, x = x_55_cast_fp16)[name = string("op_611_cast_fp16")];
+            tensor<int32, [4]> var_612 = const()[name = string("op_612"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_9_cast_fp16 = reshape(shape = var_612, x = var_611_cast_fp16)[name = string("matrix_bd_9_cast_fp16")];
+            bool matrix_ac_5_transpose_x_0 = const()[name = string("matrix_ac_5_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_5_transpose_y_0 = const()[name = string("matrix_ac_5_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_100_perm_0 = const()[name = string("transpose_100_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_101_perm_0 = const()[name = string("transpose_101_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 64]> transpose_101 = transpose(perm = transpose_101_perm_0, x = k_9_cast_fp16)[name = string("transpose_296")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_100 = transpose(perm = transpose_100_perm_0, x = var_595_cast_fp16)[name = string("transpose_297")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_5_cast_fp16 = matmul(transpose_x = matrix_ac_5_transpose_x_0, transpose_y = matrix_ac_5_transpose_y_0, x = transpose_100, y = transpose_101)[name = string("matrix_ac_5_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_11_begin_0 = const()[name = string("matrix_bd_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_11_end_0 = const()[name = string("matrix_bd_11_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
+            tensor<bool, [4]> matrix_bd_11_end_mask_0 = const()[name = string("matrix_bd_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_11_cast_fp16 = slice_by_index(begin = matrix_bd_11_begin_0, end = matrix_bd_11_end_0, end_mask = matrix_bd_11_end_mask_0, x = matrix_bd_9_cast_fp16)[name = string("matrix_bd_11_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_621_cast_fp16 = add(x = matrix_ac_5_cast_fp16, y = matrix_bd_11_cast_fp16)[name = string("op_621_cast_fp16")];
+            fp16 _inversed_scores_9_y_0_to_fp16 = const()[name = string("_inversed_scores_9_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_9_cast_fp16 = mul(x = var_621_cast_fp16, y = _inversed_scores_9_y_0_to_fp16)[name = string("_inversed_scores_9_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_11_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_9_cast_fp16, cond = mask_3)[name = string("scores_11_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_627_cast_fp16 = softmax(axis = var_30, x = scores_11_cast_fp16)[name = string("op_627_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_137_cast_fp16 = select(a = var_11_to_fp16, b = var_627_cast_fp16, cond = mask_3)[name = string("input_137_cast_fp16")];
+            bool x_57_transpose_x_0 = const()[name = string("x_57_transpose_x_0"), val = bool(false)];
+            bool x_57_transpose_y_0 = const()[name = string("x_57_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 64, 128]> value_7_cast_fp16 = transpose(perm = value_7_perm_0, x = v_5_cast_fp16)[name = string("transpose_295")];
+            tensor<fp16, [1, 8, 64, 128]> x_57_cast_fp16 = matmul(transpose_x = x_57_transpose_x_0, transpose_y = x_57_transpose_y_0, x = input_137_cast_fp16, y = value_7_cast_fp16)[name = string("x_57_cast_fp16")];
+            tensor<int32, [4]> var_631_perm_0 = const()[name = string("op_631_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_632 = const()[name = string("op_632"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 64, 8, 128]> var_631_cast_fp16 = transpose(perm = var_631_perm_0, x = x_57_cast_fp16)[name = string("transpose_294")];
+            tensor<fp16, [1, 64, 1024]> input_139_cast_fp16 = reshape(shape = var_632, x = var_631_cast_fp16)[name = string("input_139_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525056]> module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39534400))), indices_nonzero_data = tensor<uint6, [525056]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39665536))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40059392))))[name = string("module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_25_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_139_cast_fp16)[name = string("linear_25_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_143_cast_fp16 = add(x = input_135_cast_fp16, y = linear_25_cast_fp16)[name = string("input_143_cast_fp16")];
+            tensor<int32, [1]> x_61_axes_0 = const()[name = string("x_61_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_2_norm_conv_weight_to_fp16 = const()[name = string("module_layers_2_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40075840)))];
+            tensor<fp16, [1024]> module_layers_2_norm_conv_bias_to_fp16 = const()[name = string("module_layers_2_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40077952)))];
+            tensor<fp16, [1, 64, 1024]> x_61_cast_fp16 = layer_norm(axes = x_61_axes_0, beta = module_layers_2_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_conv_weight_to_fp16, x = input_143_cast_fp16)[name = string("x_61_cast_fp16")];
+            tensor<int32, [3]> input_145_perm_0 = const()[name = string("input_145_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_147_pad_type_0 = const()[name = string("input_147_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_147_strides_0 = const()[name = string("input_147_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_147_pad_0 = const()[name = string("input_147_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_147_dilations_0 = const()[name = string("input_147_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_147_groups_0 = const()[name = string("input_147_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1042766]> module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40080064))), indices_nonzero_data = tensor<uint6, [1042766]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40342272))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41124416))))[name = string("module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> input_145_cast_fp16 = transpose(perm = input_145_perm_0, x = x_61_cast_fp16)[name = string("transpose_293")];
+            tensor<fp16, [1, 2048, 64]> input_147_cast_fp16 = conv(dilations = input_147_dilations_0, groups = input_147_groups_0, pad = input_147_pad_0, pad_type = input_147_pad_type_0, strides = input_147_strides_0, weight = module_layers_2_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_145_cast_fp16)[name = string("input_147_cast_fp16")];
+            int32 x_63_split_num_splits_0 = const()[name = string("x_63_split_num_splits_0"), val = int32(2)];
+            int32 x_63_split_axis_0 = const()[name = string("x_63_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 64]> x_63_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_63_split_cast_fp16_1 = split(axis = x_63_split_axis_0, num_splits = x_63_split_num_splits_0, x = input_147_cast_fp16)[name = string("x_63_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_63_split_1_sigmoid_cast_fp16 = sigmoid(x = x_63_split_cast_fp16_1)[name = string("x_63_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_63_cast_fp16 = mul(x = x_63_split_cast_fp16_0, y = x_63_split_1_sigmoid_cast_fp16)[name = string("x_63_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_149_cast_fp16 = select(a = var_11_to_fp16, b = x_63_cast_fp16, cond = var_328)[name = string("input_149_cast_fp16")];
+            tensor<int32, [6]> input_151_pad_0 = const()[name = string("input_151_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_151_mode_0 = const()[name = string("input_151_mode_0"), val = string("constant")];
+            fp16 const_37_to_fp16 = const()[name = string("const_37_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 72]> input_151_cast_fp16 = pad(constant_val = const_37_to_fp16, mode = input_151_mode_0, pad = input_151_pad_0, x = input_149_cast_fp16)[name = string("input_151_cast_fp16")];
+            string input_153_pad_type_0 = const()[name = string("input_153_pad_type_0"), val = string("valid")];
+            int32 input_153_groups_0 = const()[name = string("input_153_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_153_strides_0 = const()[name = string("input_153_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_153_pad_0 = const()[name = string("input_153_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_153_dilations_0 = const()[name = string("input_153_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_252_to_fp16 = const()[name = string("const_252_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41157248)))];
+            tensor<fp16, [1024]> const_253_to_fp16 = const()[name = string("const_253_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41175744)))];
+            tensor<fp16, [1, 1024, 64]> input_155_cast_fp16 = conv(bias = const_253_to_fp16, dilations = input_153_dilations_0, groups = input_153_groups_0, pad = input_153_pad_0, pad_type = input_153_pad_type_0, strides = input_153_strides_0, weight = const_252_to_fp16, x = input_151_cast_fp16)[name = string("input_155_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_157_cast_fp16 = silu(x = input_155_cast_fp16)[name = string("input_157_cast_fp16")];
+            string x_65_pad_type_0 = const()[name = string("x_65_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_65_strides_0 = const()[name = string("x_65_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_65_pad_0 = const()[name = string("x_65_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_65_dilations_0 = const()[name = string("x_65_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_65_groups_0 = const()[name = string("x_65_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519385]> module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41177856))), indices_nonzero_data = tensor<uint6, [519385]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41308992))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41698624))))[name = string("module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> x_65_cast_fp16 = conv(dilations = x_65_dilations_0, groups = x_65_groups_0, pad = x_65_pad_0, pad_type = x_65_pad_type_0, strides = x_65_strides_0, weight = module_layers_2_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_157_cast_fp16)[name = string("x_65_cast_fp16")];
+            tensor<int32, [3]> input_159_perm_0 = const()[name = string("input_159_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 64, 1024]> input_159_cast_fp16 = transpose(perm = input_159_perm_0, x = x_65_cast_fp16)[name = string("transpose_292")];
+            tensor<fp16, [1, 64, 1024]> input_161_cast_fp16 = add(x = input_143_cast_fp16, y = input_159_cast_fp16)[name = string("input_161_cast_fp16")];
+            tensor<int32, [1]> input_163_axes_0 = const()[name = string("input_163_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_2_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_2_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41715072)))];
+            tensor<fp16, [1024]> module_layers_2_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_2_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41717184)))];
+            tensor<fp16, [1, 64, 1024]> input_163_cast_fp16 = layer_norm(axes = input_163_axes_0, beta = module_layers_2_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_feed_forward2_weight_to_fp16, x = input_161_cast_fp16)[name = string("input_163_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2065499]> module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41719296))), indices_nonzero_data = tensor<uint6, [2065499]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42243648))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43792896))))[name = string("module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_26_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_2_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_163_cast_fp16)[name = string("linear_26_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_167_cast_fp16 = silu(x = linear_26_cast_fp16)[name = string("input_167_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1965997]> module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43858496))), indices_nonzero_data = tensor<uint6, [1965997]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44382848))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45857472))))[name = string("module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_27_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_2_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_167_cast_fp16)[name = string("linear_27_cast_fp16")];
+            fp16 var_692_to_fp16 = const()[name = string("op_692_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_693_cast_fp16 = mul(x = linear_27_cast_fp16, y = var_692_to_fp16)[name = string("op_693_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_173_cast_fp16 = add(x = input_161_cast_fp16, y = var_693_cast_fp16)[name = string("input_173_cast_fp16")];
+            tensor<int32, [1]> input_175_axes_0 = const()[name = string("input_175_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_2_norm_out_weight_to_fp16 = const()[name = string("module_layers_2_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45873920)))];
+            tensor<fp16, [1024]> module_layers_2_norm_out_bias_to_fp16 = const()[name = string("module_layers_2_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45876032)))];
+            tensor<fp16, [1, 64, 1024]> input_175_cast_fp16 = layer_norm(axes = input_175_axes_0, beta = module_layers_2_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_out_weight_to_fp16, x = input_173_cast_fp16)[name = string("input_175_cast_fp16")];
+            tensor<int32, [1]> input_177_axes_0 = const()[name = string("input_177_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_3_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_3_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45878144)))];
+            tensor<fp16, [1024]> module_layers_3_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_3_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45880256)))];
+            tensor<fp16, [1, 64, 1024]> input_177_cast_fp16 = layer_norm(axes = input_177_axes_0, beta = module_layers_3_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_feed_forward1_weight_to_fp16, x = input_175_cast_fp16)[name = string("input_177_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2075730]> module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45882368))), indices_nonzero_data = tensor<uint6, [2075730]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46406720))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47963584))))[name = string("module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_28_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_3_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_177_cast_fp16)[name = string("linear_28_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_181_cast_fp16 = silu(x = linear_28_cast_fp16)[name = string("input_181_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1953322]> module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48029184))), indices_nonzero_data = tensor<uint6, [1953322]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48553536))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50018624))))[name = string("module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_29_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_181_cast_fp16)[name = string("linear_29_cast_fp16")];
+            fp16 var_721_to_fp16 = const()[name = string("op_721_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_722_cast_fp16 = mul(x = linear_29_cast_fp16, y = var_721_to_fp16)[name = string("op_722_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_187_cast_fp16 = add(x = input_175_cast_fp16, y = var_722_cast_fp16)[name = string("input_187_cast_fp16")];
+            tensor<int32, [1]> query_7_axes_0 = const()[name = string("query_7_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_3_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_3_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50035072)))];
+            tensor<fp16, [1024]> module_layers_3_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_3_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50037184)))];
+            tensor<fp16, [1, 64, 1024]> query_7_cast_fp16 = layer_norm(axes = query_7_axes_0, beta = module_layers_3_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_self_att_weight_to_fp16, x = input_187_cast_fp16)[name = string("query_7_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522877]> module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50039296))), indices_nonzero_data = tensor<uint6, [522877]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50170432))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50562688))))[name = string("module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_30_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_7_cast_fp16)[name = string("linear_30_cast_fp16")];
+            tensor<int32, [4]> var_738 = const()[name = string("op_738"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> q_19_cast_fp16 = reshape(shape = var_738, x = linear_30_cast_fp16)[name = string("q_19_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520403]> module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50579136))), indices_nonzero_data = tensor<uint6, [520403]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50710272))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51100672))))[name = string("module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_31_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_7_cast_fp16)[name = string("linear_31_cast_fp16")];
+            tensor<int32, [4]> var_742 = const()[name = string("op_742"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> k_13_cast_fp16 = reshape(shape = var_742, x = linear_31_cast_fp16)[name = string("k_13_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521127]> module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51117120))), indices_nonzero_data = tensor<uint6, [521127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51248256))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51639168))))[name = string("module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_32_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_7_cast_fp16)[name = string("linear_32_cast_fp16")];
+            tensor<int32, [4]> var_746 = const()[name = string("op_746"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> v_7_cast_fp16 = reshape(shape = var_746, x = linear_32_cast_fp16)[name = string("v_7_cast_fp16")];
+            tensor<int32, [4]> value_9_perm_0 = const()[name = string("value_9_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_3_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_3_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51655616)))];
+            tensor<fp16, [1, 64, 8, 128]> var_758_cast_fp16 = add(x = q_19_cast_fp16, y = module_layers_3_self_attn_pos_bias_u_to_fp16)[name = string("op_758_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_3_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_3_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51657728)))];
+            tensor<fp16, [1, 64, 8, 128]> var_760_cast_fp16 = add(x = q_19_cast_fp16, y = module_layers_3_self_attn_pos_bias_v_to_fp16)[name = string("op_760_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_7_perm_0 = const()[name = string("q_with_bias_v_7_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_73_transpose_x_0 = const()[name = string("x_73_transpose_x_0"), val = bool(false)];
+            bool x_73_transpose_y_0 = const()[name = string("x_73_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 127]> var_762_to_fp16 = const()[name = string("op_762_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51659840)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_7_cast_fp16 = transpose(perm = q_with_bias_v_7_perm_0, x = var_760_cast_fp16)[name = string("transpose_291")];
+            tensor<fp16, [1, 8, 64, 127]> x_73_cast_fp16 = matmul(transpose_x = x_73_transpose_x_0, transpose_y = x_73_transpose_y_0, x = q_with_bias_v_7_cast_fp16, y = var_762_to_fp16)[name = string("x_73_cast_fp16")];
+            tensor<int32, [8]> x_75_pad_0 = const()[name = string("x_75_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_75_mode_0 = const()[name = string("x_75_mode_0"), val = string("constant")];
+            fp16 const_44_to_fp16 = const()[name = string("const_44_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 64, 128]> x_75_cast_fp16 = pad(constant_val = const_44_to_fp16, mode = x_75_mode_0, pad = x_75_pad_0, x = x_73_cast_fp16)[name = string("x_75_cast_fp16")];
+            tensor<int32, [4]> var_770 = const()[name = string("op_770"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_77_cast_fp16 = reshape(shape = var_770, x = x_75_cast_fp16)[name = string("x_77_cast_fp16")];
+            tensor<int32, [4]> var_774_begin_0 = const()[name = string("op_774_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_774_end_0 = const()[name = string("op_774_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
+            tensor<bool, [4]> var_774_end_mask_0 = const()[name = string("op_774_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 127, 64]> var_774_cast_fp16 = slice_by_index(begin = var_774_begin_0, end = var_774_end_0, end_mask = var_774_end_mask_0, x = x_77_cast_fp16)[name = string("op_774_cast_fp16")];
+            tensor<int32, [4]> var_775 = const()[name = string("op_775"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_13_cast_fp16 = reshape(shape = var_775, x = var_774_cast_fp16)[name = string("matrix_bd_13_cast_fp16")];
+            bool matrix_ac_7_transpose_x_0 = const()[name = string("matrix_ac_7_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_7_transpose_y_0 = const()[name = string("matrix_ac_7_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_102_perm_0 = const()[name = string("transpose_102_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_103_perm_0 = const()[name = string("transpose_103_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 64]> transpose_103 = transpose(perm = transpose_103_perm_0, x = k_13_cast_fp16)[name = string("transpose_289")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_102 = transpose(perm = transpose_102_perm_0, x = var_758_cast_fp16)[name = string("transpose_290")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_7_cast_fp16 = matmul(transpose_x = matrix_ac_7_transpose_x_0, transpose_y = matrix_ac_7_transpose_y_0, x = transpose_102, y = transpose_103)[name = string("matrix_ac_7_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_15_begin_0 = const()[name = string("matrix_bd_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_15_end_0 = const()[name = string("matrix_bd_15_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
+            tensor<bool, [4]> matrix_bd_15_end_mask_0 = const()[name = string("matrix_bd_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_15_cast_fp16 = slice_by_index(begin = matrix_bd_15_begin_0, end = matrix_bd_15_end_0, end_mask = matrix_bd_15_end_mask_0, x = matrix_bd_13_cast_fp16)[name = string("matrix_bd_15_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_784_cast_fp16 = add(x = matrix_ac_7_cast_fp16, y = matrix_bd_15_cast_fp16)[name = string("op_784_cast_fp16")];
+            fp16 _inversed_scores_13_y_0_to_fp16 = const()[name = string("_inversed_scores_13_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_13_cast_fp16 = mul(x = var_784_cast_fp16, y = _inversed_scores_13_y_0_to_fp16)[name = string("_inversed_scores_13_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_15_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_13_cast_fp16, cond = mask_3)[name = string("scores_15_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_790_cast_fp16 = softmax(axis = var_30, x = scores_15_cast_fp16)[name = string("op_790_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_189_cast_fp16 = select(a = var_11_to_fp16, b = var_790_cast_fp16, cond = mask_3)[name = string("input_189_cast_fp16")];
+            bool x_79_transpose_x_0 = const()[name = string("x_79_transpose_x_0"), val = bool(false)];
+            bool x_79_transpose_y_0 = const()[name = string("x_79_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 64, 128]> value_9_cast_fp16 = transpose(perm = value_9_perm_0, x = v_7_cast_fp16)[name = string("transpose_288")];
+            tensor<fp16, [1, 8, 64, 128]> x_79_cast_fp16 = matmul(transpose_x = x_79_transpose_x_0, transpose_y = x_79_transpose_y_0, x = input_189_cast_fp16, y = value_9_cast_fp16)[name = string("x_79_cast_fp16")];
+            tensor<int32, [4]> var_794_perm_0 = const()[name = string("op_794_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_795 = const()[name = string("op_795"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 64, 8, 128]> var_794_cast_fp16 = transpose(perm = var_794_perm_0, x = x_79_cast_fp16)[name = string("transpose_287")];
+            tensor<fp16, [1, 64, 1024]> input_191_cast_fp16 = reshape(shape = var_795, x = var_794_cast_fp16)[name = string("input_191_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522589]> module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51920000))), indices_nonzero_data = tensor<uint6, [522589]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52051136))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52443200))))[name = string("module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_34_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_191_cast_fp16)[name = string("linear_34_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_195_cast_fp16 = add(x = input_187_cast_fp16, y = linear_34_cast_fp16)[name = string("input_195_cast_fp16")];
+            tensor<int32, [1]> x_83_axes_0 = const()[name = string("x_83_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_3_norm_conv_weight_to_fp16 = const()[name = string("module_layers_3_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52459648)))];
+            tensor<fp16, [1024]> module_layers_3_norm_conv_bias_to_fp16 = const()[name = string("module_layers_3_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52461760)))];
+            tensor<fp16, [1, 64, 1024]> x_83_cast_fp16 = layer_norm(axes = x_83_axes_0, beta = module_layers_3_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_conv_weight_to_fp16, x = input_195_cast_fp16)[name = string("x_83_cast_fp16")];
+            tensor<int32, [3]> input_197_perm_0 = const()[name = string("input_197_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_199_pad_type_0 = const()[name = string("input_199_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_199_strides_0 = const()[name = string("input_199_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_199_pad_0 = const()[name = string("input_199_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_199_dilations_0 = const()[name = string("input_199_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_199_groups_0 = const()[name = string("input_199_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1044665]> module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52463872))), indices_nonzero_data = tensor<uint6, [1044665]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52726080))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53509696))))[name = string("module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> input_197_cast_fp16 = transpose(perm = input_197_perm_0, x = x_83_cast_fp16)[name = string("transpose_286")];
+            tensor<fp16, [1, 2048, 64]> input_199_cast_fp16 = conv(dilations = input_199_dilations_0, groups = input_199_groups_0, pad = input_199_pad_0, pad_type = input_199_pad_type_0, strides = input_199_strides_0, weight = module_layers_3_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_197_cast_fp16)[name = string("input_199_cast_fp16")];
+            int32 x_85_split_num_splits_0 = const()[name = string("x_85_split_num_splits_0"), val = int32(2)];
+            int32 x_85_split_axis_0 = const()[name = string("x_85_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 64]> x_85_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_85_split_cast_fp16_1 = split(axis = x_85_split_axis_0, num_splits = x_85_split_num_splits_0, x = input_199_cast_fp16)[name = string("x_85_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_85_split_1_sigmoid_cast_fp16 = sigmoid(x = x_85_split_cast_fp16_1)[name = string("x_85_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_85_cast_fp16 = mul(x = x_85_split_cast_fp16_0, y = x_85_split_1_sigmoid_cast_fp16)[name = string("x_85_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_201_cast_fp16 = select(a = var_11_to_fp16, b = x_85_cast_fp16, cond = var_328)[name = string("input_201_cast_fp16")];
+            tensor<int32, [6]> input_203_pad_0 = const()[name = string("input_203_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_203_mode_0 = const()[name = string("input_203_mode_0"), val = string("constant")];
+            fp16 const_47_to_fp16 = const()[name = string("const_47_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 72]> input_203_cast_fp16 = pad(constant_val = const_47_to_fp16, mode = input_203_mode_0, pad = input_203_pad_0, x = input_201_cast_fp16)[name = string("input_203_cast_fp16")];
+            string input_205_pad_type_0 = const()[name = string("input_205_pad_type_0"), val = string("valid")];
+            int32 input_205_groups_0 = const()[name = string("input_205_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_205_strides_0 = const()[name = string("input_205_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_205_pad_0 = const()[name = string("input_205_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_205_dilations_0 = const()[name = string("input_205_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_254_to_fp16 = const()[name = string("const_254_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53542528)))];
+            tensor<fp16, [1024]> const_255_to_fp16 = const()[name = string("const_255_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53561024)))];
+            tensor<fp16, [1, 1024, 64]> input_207_cast_fp16 = conv(bias = const_255_to_fp16, dilations = input_205_dilations_0, groups = input_205_groups_0, pad = input_205_pad_0, pad_type = input_205_pad_type_0, strides = input_205_strides_0, weight = const_254_to_fp16, x = input_203_cast_fp16)[name = string("input_207_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_209_cast_fp16 = silu(x = input_207_cast_fp16)[name = string("input_209_cast_fp16")];
+            string x_87_pad_type_0 = const()[name = string("x_87_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_87_strides_0 = const()[name = string("x_87_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_87_pad_0 = const()[name = string("x_87_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_87_dilations_0 = const()[name = string("x_87_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_87_groups_0 = const()[name = string("x_87_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519219]> module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53563136))), indices_nonzero_data = tensor<uint6, [519219]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53694272))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54083776))))[name = string("module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> x_87_cast_fp16 = conv(dilations = x_87_dilations_0, groups = x_87_groups_0, pad = x_87_pad_0, pad_type = x_87_pad_type_0, strides = x_87_strides_0, weight = module_layers_3_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_209_cast_fp16)[name = string("x_87_cast_fp16")];
+            tensor<int32, [3]> input_211_perm_0 = const()[name = string("input_211_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 64, 1024]> input_211_cast_fp16 = transpose(perm = input_211_perm_0, x = x_87_cast_fp16)[name = string("transpose_285")];
+            tensor<fp16, [1, 64, 1024]> input_213_cast_fp16 = add(x = input_195_cast_fp16, y = input_211_cast_fp16)[name = string("input_213_cast_fp16")];
+            tensor<int32, [1]> input_215_axes_0 = const()[name = string("input_215_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_3_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_3_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54100224)))];
+            tensor<fp16, [1024]> module_layers_3_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_3_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54102336)))];
+            tensor<fp16, [1, 64, 1024]> input_215_cast_fp16 = layer_norm(axes = input_215_axes_0, beta = module_layers_3_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_feed_forward2_weight_to_fp16, x = input_213_cast_fp16)[name = string("input_215_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2059316]> module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54104448))), indices_nonzero_data = tensor<uint6, [2059316]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54628800))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56173376))))[name = string("module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_35_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_3_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_215_cast_fp16)[name = string("linear_35_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_219_cast_fp16 = silu(x = linear_35_cast_fp16)[name = string("input_219_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1938642]> module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56238976))), indices_nonzero_data = tensor<uint6, [1938642]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56763328))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58217408))))[name = string("module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_36_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_3_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_219_cast_fp16)[name = string("linear_36_cast_fp16")];
+            fp16 var_855_to_fp16 = const()[name = string("op_855_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_856_cast_fp16 = mul(x = linear_36_cast_fp16, y = var_855_to_fp16)[name = string("op_856_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_225_cast_fp16 = add(x = input_213_cast_fp16, y = var_856_cast_fp16)[name = string("input_225_cast_fp16")];
+            tensor<int32, [1]> input_227_axes_0 = const()[name = string("input_227_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_3_norm_out_weight_to_fp16 = const()[name = string("module_layers_3_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58233856)))];
+            tensor<fp16, [1024]> module_layers_3_norm_out_bias_to_fp16 = const()[name = string("module_layers_3_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58235968)))];
+            tensor<fp16, [1, 64, 1024]> input_227_cast_fp16 = layer_norm(axes = input_227_axes_0, beta = module_layers_3_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_out_weight_to_fp16, x = input_225_cast_fp16)[name = string("input_227_cast_fp16")];
+            tensor<int32, [1]> input_229_axes_0 = const()[name = string("input_229_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_4_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_4_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58238080)))];
+            tensor<fp16, [1024]> module_layers_4_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_4_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58240192)))];
+            tensor<fp16, [1, 64, 1024]> input_229_cast_fp16 = layer_norm(axes = input_229_axes_0, beta = module_layers_4_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_feed_forward1_weight_to_fp16, x = input_227_cast_fp16)[name = string("input_229_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2071628]> module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58242304))), indices_nonzero_data = tensor<uint6, [2071628]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58766656))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60320448))))[name = string("module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_37_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_4_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_229_cast_fp16)[name = string("linear_37_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_233_cast_fp16 = silu(x = linear_37_cast_fp16)[name = string("input_233_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1939129]> module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60386048))), indices_nonzero_data = tensor<uint6, [1939129]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60910400))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62364864))))[name = string("module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_38_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_233_cast_fp16)[name = string("linear_38_cast_fp16")];
+            fp16 var_884_to_fp16 = const()[name = string("op_884_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_885_cast_fp16 = mul(x = linear_38_cast_fp16, y = var_884_to_fp16)[name = string("op_885_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_239_cast_fp16 = add(x = input_227_cast_fp16, y = var_885_cast_fp16)[name = string("input_239_cast_fp16")];
+            tensor<int32, [1]> query_9_axes_0 = const()[name = string("query_9_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_4_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_4_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62381312)))];
+            tensor<fp16, [1024]> module_layers_4_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_4_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62383424)))];
+            tensor<fp16, [1, 64, 1024]> query_9_cast_fp16 = layer_norm(axes = query_9_axes_0, beta = module_layers_4_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_self_att_weight_to_fp16, x = input_239_cast_fp16)[name = string("query_9_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523421]> module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62385536))), indices_nonzero_data = tensor<uint6, [523421]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62516672))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62909312))))[name = string("module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_39_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_9_cast_fp16)[name = string("linear_39_cast_fp16")];
+            tensor<int32, [4]> var_901 = const()[name = string("op_901"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> q_25_cast_fp16 = reshape(shape = var_901, x = linear_39_cast_fp16)[name = string("q_25_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521081]> module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62925760))), indices_nonzero_data = tensor<uint6, [521081]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63056896))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63447808))))[name = string("module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_40_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_9_cast_fp16)[name = string("linear_40_cast_fp16")];
+            tensor<int32, [4]> var_905 = const()[name = string("op_905"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> k_17_cast_fp16 = reshape(shape = var_905, x = linear_40_cast_fp16)[name = string("k_17_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520224]> module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63464256))), indices_nonzero_data = tensor<uint6, [520224]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63595392))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63985664))))[name = string("module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_41_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_9_cast_fp16)[name = string("linear_41_cast_fp16")];
+            tensor<int32, [4]> var_909 = const()[name = string("op_909"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> v_9_cast_fp16 = reshape(shape = var_909, x = linear_41_cast_fp16)[name = string("v_9_cast_fp16")];
+            tensor<int32, [4]> value_11_perm_0 = const()[name = string("value_11_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_4_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_4_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64002112)))];
+            tensor<fp16, [1, 64, 8, 128]> var_921_cast_fp16 = add(x = q_25_cast_fp16, y = module_layers_4_self_attn_pos_bias_u_to_fp16)[name = string("op_921_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_4_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_4_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64004224)))];
+            tensor<fp16, [1, 64, 8, 128]> var_923_cast_fp16 = add(x = q_25_cast_fp16, y = module_layers_4_self_attn_pos_bias_v_to_fp16)[name = string("op_923_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_9_perm_0 = const()[name = string("q_with_bias_v_9_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_95_transpose_x_0 = const()[name = string("x_95_transpose_x_0"), val = bool(false)];
+            bool x_95_transpose_y_0 = const()[name = string("x_95_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 127]> var_925_to_fp16 = const()[name = string("op_925_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64006336)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_9_cast_fp16 = transpose(perm = q_with_bias_v_9_perm_0, x = var_923_cast_fp16)[name = string("transpose_284")];
+            tensor<fp16, [1, 8, 64, 127]> x_95_cast_fp16 = matmul(transpose_x = x_95_transpose_x_0, transpose_y = x_95_transpose_y_0, x = q_with_bias_v_9_cast_fp16, y = var_925_to_fp16)[name = string("x_95_cast_fp16")];
+            tensor<int32, [8]> x_97_pad_0 = const()[name = string("x_97_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_97_mode_0 = const()[name = string("x_97_mode_0"), val = string("constant")];
+            fp16 const_54_to_fp16 = const()[name = string("const_54_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 64, 128]> x_97_cast_fp16 = pad(constant_val = const_54_to_fp16, mode = x_97_mode_0, pad = x_97_pad_0, x = x_95_cast_fp16)[name = string("x_97_cast_fp16")];
+            tensor<int32, [4]> var_933 = const()[name = string("op_933"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_99_cast_fp16 = reshape(shape = var_933, x = x_97_cast_fp16)[name = string("x_99_cast_fp16")];
+            tensor<int32, [4]> var_937_begin_0 = const()[name = string("op_937_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_937_end_0 = const()[name = string("op_937_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
+            tensor<bool, [4]> var_937_end_mask_0 = const()[name = string("op_937_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 127, 64]> var_937_cast_fp16 = slice_by_index(begin = var_937_begin_0, end = var_937_end_0, end_mask = var_937_end_mask_0, x = x_99_cast_fp16)[name = string("op_937_cast_fp16")];
+            tensor<int32, [4]> var_938 = const()[name = string("op_938"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_17_cast_fp16 = reshape(shape = var_938, x = var_937_cast_fp16)[name = string("matrix_bd_17_cast_fp16")];
+            bool matrix_ac_9_transpose_x_0 = const()[name = string("matrix_ac_9_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_9_transpose_y_0 = const()[name = string("matrix_ac_9_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_104_perm_0 = const()[name = string("transpose_104_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_105_perm_0 = const()[name = string("transpose_105_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 64]> transpose_105 = transpose(perm = transpose_105_perm_0, x = k_17_cast_fp16)[name = string("transpose_282")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_104 = transpose(perm = transpose_104_perm_0, x = var_921_cast_fp16)[name = string("transpose_283")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_9_cast_fp16 = matmul(transpose_x = matrix_ac_9_transpose_x_0, transpose_y = matrix_ac_9_transpose_y_0, x = transpose_104, y = transpose_105)[name = string("matrix_ac_9_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_19_begin_0 = const()[name = string("matrix_bd_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_19_end_0 = const()[name = string("matrix_bd_19_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
+            tensor<bool, [4]> matrix_bd_19_end_mask_0 = const()[name = string("matrix_bd_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_19_cast_fp16 = slice_by_index(begin = matrix_bd_19_begin_0, end = matrix_bd_19_end_0, end_mask = matrix_bd_19_end_mask_0, x = matrix_bd_17_cast_fp16)[name = string("matrix_bd_19_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_947_cast_fp16 = add(x = matrix_ac_9_cast_fp16, y = matrix_bd_19_cast_fp16)[name = string("op_947_cast_fp16")];
+            fp16 _inversed_scores_17_y_0_to_fp16 = const()[name = string("_inversed_scores_17_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_17_cast_fp16 = mul(x = var_947_cast_fp16, y = _inversed_scores_17_y_0_to_fp16)[name = string("_inversed_scores_17_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_19_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_17_cast_fp16, cond = mask_3)[name = string("scores_19_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_953_cast_fp16 = softmax(axis = var_30, x = scores_19_cast_fp16)[name = string("op_953_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_241_cast_fp16 = select(a = var_11_to_fp16, b = var_953_cast_fp16, cond = mask_3)[name = string("input_241_cast_fp16")];
+            bool x_101_transpose_x_0 = const()[name = string("x_101_transpose_x_0"), val = bool(false)];
+            bool x_101_transpose_y_0 = const()[name = string("x_101_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 64, 128]> value_11_cast_fp16 = transpose(perm = value_11_perm_0, x = v_9_cast_fp16)[name = string("transpose_281")];
+            tensor<fp16, [1, 8, 64, 128]> x_101_cast_fp16 = matmul(transpose_x = x_101_transpose_x_0, transpose_y = x_101_transpose_y_0, x = input_241_cast_fp16, y = value_11_cast_fp16)[name = string("x_101_cast_fp16")];
+            tensor<int32, [4]> var_957_perm_0 = const()[name = string("op_957_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_958 = const()[name = string("op_958"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 64, 8, 128]> var_957_cast_fp16 = transpose(perm = var_957_perm_0, x = x_101_cast_fp16)[name = string("transpose_280")];
+            tensor<fp16, [1, 64, 1024]> input_243_cast_fp16 = reshape(shape = var_958, x = var_957_cast_fp16)[name = string("input_243_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523771]> module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64266496))), indices_nonzero_data = tensor<uint6, [523771]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64397632))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64790528))))[name = string("module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_43_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_243_cast_fp16)[name = string("linear_43_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_247_cast_fp16 = add(x = input_239_cast_fp16, y = linear_43_cast_fp16)[name = string("input_247_cast_fp16")];
+            tensor<int32, [1]> x_105_axes_0 = const()[name = string("x_105_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_4_norm_conv_weight_to_fp16 = const()[name = string("module_layers_4_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64806976)))];
+            tensor<fp16, [1024]> module_layers_4_norm_conv_bias_to_fp16 = const()[name = string("module_layers_4_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64809088)))];
+            tensor<fp16, [1, 64, 1024]> x_105_cast_fp16 = layer_norm(axes = x_105_axes_0, beta = module_layers_4_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_conv_weight_to_fp16, x = input_247_cast_fp16)[name = string("x_105_cast_fp16")];
+            tensor<int32, [3]> input_249_perm_0 = const()[name = string("input_249_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_251_pad_type_0 = const()[name = string("input_251_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_251_strides_0 = const()[name = string("input_251_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_251_pad_0 = const()[name = string("input_251_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_251_dilations_0 = const()[name = string("input_251_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_251_groups_0 = const()[name = string("input_251_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1046993]> module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64811200))), indices_nonzero_data = tensor<uint6, [1046993]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65073408))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65858752))))[name = string("module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> input_249_cast_fp16 = transpose(perm = input_249_perm_0, x = x_105_cast_fp16)[name = string("transpose_279")];
+            tensor<fp16, [1, 2048, 64]> input_251_cast_fp16 = conv(dilations = input_251_dilations_0, groups = input_251_groups_0, pad = input_251_pad_0, pad_type = input_251_pad_type_0, strides = input_251_strides_0, weight = module_layers_4_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_249_cast_fp16)[name = string("input_251_cast_fp16")];
+            int32 x_107_split_num_splits_0 = const()[name = string("x_107_split_num_splits_0"), val = int32(2)];
+            int32 x_107_split_axis_0 = const()[name = string("x_107_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 64]> x_107_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_107_split_cast_fp16_1 = split(axis = x_107_split_axis_0, num_splits = x_107_split_num_splits_0, x = input_251_cast_fp16)[name = string("x_107_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_107_split_1_sigmoid_cast_fp16 = sigmoid(x = x_107_split_cast_fp16_1)[name = string("x_107_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_107_cast_fp16 = mul(x = x_107_split_cast_fp16_0, y = x_107_split_1_sigmoid_cast_fp16)[name = string("x_107_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_253_cast_fp16 = select(a = var_11_to_fp16, b = x_107_cast_fp16, cond = var_328)[name = string("input_253_cast_fp16")];
+            tensor<int32, [6]> input_255_pad_0 = const()[name = string("input_255_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_255_mode_0 = const()[name = string("input_255_mode_0"), val = string("constant")];
+            fp16 const_57_to_fp16 = const()[name = string("const_57_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 72]> input_255_cast_fp16 = pad(constant_val = const_57_to_fp16, mode = input_255_mode_0, pad = input_255_pad_0, x = input_253_cast_fp16)[name = string("input_255_cast_fp16")];
+            string input_257_pad_type_0 = const()[name = string("input_257_pad_type_0"), val = string("valid")];
+            int32 input_257_groups_0 = const()[name = string("input_257_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_257_strides_0 = const()[name = string("input_257_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_257_pad_0 = const()[name = string("input_257_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_257_dilations_0 = const()[name = string("input_257_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_256_to_fp16 = const()[name = string("const_256_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65891584)))];
+            tensor<fp16, [1024]> const_257_to_fp16 = const()[name = string("const_257_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65910080)))];
+            tensor<fp16, [1, 1024, 64]> input_259_cast_fp16 = conv(bias = const_257_to_fp16, dilations = input_257_dilations_0, groups = input_257_groups_0, pad = input_257_pad_0, pad_type = input_257_pad_type_0, strides = input_257_strides_0, weight = const_256_to_fp16, x = input_255_cast_fp16)[name = string("input_259_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_261_cast_fp16 = silu(x = input_259_cast_fp16)[name = string("input_261_cast_fp16")];
+            string x_109_pad_type_0 = const()[name = string("x_109_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_109_strides_0 = const()[name = string("x_109_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_109_pad_0 = const()[name = string("x_109_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_109_dilations_0 = const()[name = string("x_109_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_109_groups_0 = const()[name = string("x_109_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520275]> module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65912192))), indices_nonzero_data = tensor<uint6, [520275]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66043328))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66433600))))[name = string("module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> x_109_cast_fp16 = conv(dilations = x_109_dilations_0, groups = x_109_groups_0, pad = x_109_pad_0, pad_type = x_109_pad_type_0, strides = x_109_strides_0, weight = module_layers_4_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_261_cast_fp16)[name = string("x_109_cast_fp16")];
+            tensor<int32, [3]> input_263_perm_0 = const()[name = string("input_263_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 64, 1024]> input_263_cast_fp16 = transpose(perm = input_263_perm_0, x = x_109_cast_fp16)[name = string("transpose_278")];
+            tensor<fp16, [1, 64, 1024]> input_265_cast_fp16 = add(x = input_247_cast_fp16, y = input_263_cast_fp16)[name = string("input_265_cast_fp16")];
+            tensor<int32, [1]> input_267_axes_0 = const()[name = string("input_267_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_4_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_4_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66450048)))];
+            tensor<fp16, [1024]> module_layers_4_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_4_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66452160)))];
+            tensor<fp16, [1, 64, 1024]> input_267_cast_fp16 = layer_norm(axes = input_267_axes_0, beta = module_layers_4_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_feed_forward2_weight_to_fp16, x = input_265_cast_fp16)[name = string("input_267_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2060977]> module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66454272))), indices_nonzero_data = tensor<uint6, [2060977]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66978624))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68524480))))[name = string("module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_44_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_4_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_267_cast_fp16)[name = string("linear_44_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_271_cast_fp16 = silu(x = linear_44_cast_fp16)[name = string("input_271_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1931171]> module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68590080))), indices_nonzero_data = tensor<uint6, [1931171]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69114432))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70562880))))[name = string("module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_45_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_4_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_271_cast_fp16)[name = string("linear_45_cast_fp16")];
+            fp16 var_1018_to_fp16 = const()[name = string("op_1018_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_1019_cast_fp16 = mul(x = linear_45_cast_fp16, y = var_1018_to_fp16)[name = string("op_1019_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_277_cast_fp16 = add(x = input_265_cast_fp16, y = var_1019_cast_fp16)[name = string("input_277_cast_fp16")];
+            tensor<int32, [1]> input_279_axes_0 = const()[name = string("input_279_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_4_norm_out_weight_to_fp16 = const()[name = string("module_layers_4_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70579328)))];
+            tensor<fp16, [1024]> module_layers_4_norm_out_bias_to_fp16 = const()[name = string("module_layers_4_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70581440)))];
+            tensor<fp16, [1, 64, 1024]> input_279_cast_fp16 = layer_norm(axes = input_279_axes_0, beta = module_layers_4_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_out_weight_to_fp16, x = input_277_cast_fp16)[name = string("input_279_cast_fp16")];
+            tensor<int32, [1]> input_281_axes_0 = const()[name = string("input_281_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_5_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_5_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70583552)))];
+            tensor<fp16, [1024]> module_layers_5_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_5_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70585664)))];
+            tensor<fp16, [1, 64, 1024]> input_281_cast_fp16 = layer_norm(axes = input_281_axes_0, beta = module_layers_5_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_feed_forward1_weight_to_fp16, x = input_279_cast_fp16)[name = string("input_281_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2075089]> module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70587776))), indices_nonzero_data = tensor<uint6, [2075089]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71112128))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72668544))))[name = string("module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_46_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_5_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_281_cast_fp16)[name = string("linear_46_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_285_cast_fp16 = silu(x = linear_46_cast_fp16)[name = string("input_285_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1956387]> module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72734144))), indices_nonzero_data = tensor<uint6, [1956387]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73258496))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74725888))))[name = string("module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_47_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_285_cast_fp16)[name = string("linear_47_cast_fp16")];
+            fp16 var_1047_to_fp16 = const()[name = string("op_1047_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_1048_cast_fp16 = mul(x = linear_47_cast_fp16, y = var_1047_to_fp16)[name = string("op_1048_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_291_cast_fp16 = add(x = input_279_cast_fp16, y = var_1048_cast_fp16)[name = string("input_291_cast_fp16")];
+            tensor<int32, [1]> query_11_axes_0 = const()[name = string("query_11_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_5_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_5_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74742336)))];
+            tensor<fp16, [1024]> module_layers_5_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_5_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74744448)))];
+            tensor<fp16, [1, 64, 1024]> query_11_cast_fp16 = layer_norm(axes = query_11_axes_0, beta = module_layers_5_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_self_att_weight_to_fp16, x = input_291_cast_fp16)[name = string("query_11_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521972]> module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74746560))), indices_nonzero_data = tensor<uint6, [521972]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74877696))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75269248))))[name = string("module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_48_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_11_cast_fp16)[name = string("linear_48_cast_fp16")];
+            tensor<int32, [4]> var_1064 = const()[name = string("op_1064"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> q_31_cast_fp16 = reshape(shape = var_1064, x = linear_48_cast_fp16)[name = string("q_31_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522028]> module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75285696))), indices_nonzero_data = tensor<uint6, [522028]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75416832))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75808448))))[name = string("module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_49_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_11_cast_fp16)[name = string("linear_49_cast_fp16")];
+            tensor<int32, [4]> var_1068 = const()[name = string("op_1068"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> k_21_cast_fp16 = reshape(shape = var_1068, x = linear_49_cast_fp16)[name = string("k_21_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519496]> module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75824896))), indices_nonzero_data = tensor<uint6, [519496]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75956032))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76345728))))[name = string("module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_50_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_11_cast_fp16)[name = string("linear_50_cast_fp16")];
+            tensor<int32, [4]> var_1072 = const()[name = string("op_1072"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> v_11_cast_fp16 = reshape(shape = var_1072, x = linear_50_cast_fp16)[name = string("v_11_cast_fp16")];
+            tensor<int32, [4]> value_13_perm_0 = const()[name = string("value_13_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_5_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_5_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76362176)))];
+            tensor<fp16, [1, 64, 8, 128]> var_1084_cast_fp16 = add(x = q_31_cast_fp16, y = module_layers_5_self_attn_pos_bias_u_to_fp16)[name = string("op_1084_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_5_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_5_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76364288)))];
+            tensor<fp16, [1, 64, 8, 128]> var_1086_cast_fp16 = add(x = q_31_cast_fp16, y = module_layers_5_self_attn_pos_bias_v_to_fp16)[name = string("op_1086_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_11_perm_0 = const()[name = string("q_with_bias_v_11_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_117_transpose_x_0 = const()[name = string("x_117_transpose_x_0"), val = bool(false)];
+            bool x_117_transpose_y_0 = const()[name = string("x_117_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 127]> var_1088_to_fp16 = const()[name = string("op_1088_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76366400)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_11_cast_fp16 = transpose(perm = q_with_bias_v_11_perm_0, x = var_1086_cast_fp16)[name = string("transpose_277")];
+            tensor<fp16, [1, 8, 64, 127]> x_117_cast_fp16 = matmul(transpose_x = x_117_transpose_x_0, transpose_y = x_117_transpose_y_0, x = q_with_bias_v_11_cast_fp16, y = var_1088_to_fp16)[name = string("x_117_cast_fp16")];
+            tensor<int32, [8]> x_119_pad_0 = const()[name = string("x_119_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_119_mode_0 = const()[name = string("x_119_mode_0"), val = string("constant")];
+            fp16 const_64_to_fp16 = const()[name = string("const_64_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 64, 128]> x_119_cast_fp16 = pad(constant_val = const_64_to_fp16, mode = x_119_mode_0, pad = x_119_pad_0, x = x_117_cast_fp16)[name = string("x_119_cast_fp16")];
+            tensor<int32, [4]> var_1096 = const()[name = string("op_1096"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_121_cast_fp16 = reshape(shape = var_1096, x = x_119_cast_fp16)[name = string("x_121_cast_fp16")];
+            tensor<int32, [4]> var_1100_begin_0 = const()[name = string("op_1100_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1100_end_0 = const()[name = string("op_1100_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
+            tensor<bool, [4]> var_1100_end_mask_0 = const()[name = string("op_1100_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 127, 64]> var_1100_cast_fp16 = slice_by_index(begin = var_1100_begin_0, end = var_1100_end_0, end_mask = var_1100_end_mask_0, x = x_121_cast_fp16)[name = string("op_1100_cast_fp16")];
+            tensor<int32, [4]> var_1101 = const()[name = string("op_1101"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_21_cast_fp16 = reshape(shape = var_1101, x = var_1100_cast_fp16)[name = string("matrix_bd_21_cast_fp16")];
+            bool matrix_ac_11_transpose_x_0 = const()[name = string("matrix_ac_11_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_11_transpose_y_0 = const()[name = string("matrix_ac_11_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_106_perm_0 = const()[name = string("transpose_106_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_107_perm_0 = const()[name = string("transpose_107_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 64]> transpose_107 = transpose(perm = transpose_107_perm_0, x = k_21_cast_fp16)[name = string("transpose_275")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_106 = transpose(perm = transpose_106_perm_0, x = var_1084_cast_fp16)[name = string("transpose_276")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_11_cast_fp16 = matmul(transpose_x = matrix_ac_11_transpose_x_0, transpose_y = matrix_ac_11_transpose_y_0, x = transpose_106, y = transpose_107)[name = string("matrix_ac_11_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_23_begin_0 = const()[name = string("matrix_bd_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_23_end_0 = const()[name = string("matrix_bd_23_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
+            tensor<bool, [4]> matrix_bd_23_end_mask_0 = const()[name = string("matrix_bd_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_23_cast_fp16 = slice_by_index(begin = matrix_bd_23_begin_0, end = matrix_bd_23_end_0, end_mask = matrix_bd_23_end_mask_0, x = matrix_bd_21_cast_fp16)[name = string("matrix_bd_23_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_1110_cast_fp16 = add(x = matrix_ac_11_cast_fp16, y = matrix_bd_23_cast_fp16)[name = string("op_1110_cast_fp16")];
+            fp16 _inversed_scores_21_y_0_to_fp16 = const()[name = string("_inversed_scores_21_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_21_cast_fp16 = mul(x = var_1110_cast_fp16, y = _inversed_scores_21_y_0_to_fp16)[name = string("_inversed_scores_21_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_23_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_21_cast_fp16, cond = mask_3)[name = string("scores_23_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_1116_cast_fp16 = softmax(axis = var_30, x = scores_23_cast_fp16)[name = string("op_1116_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_293_cast_fp16 = select(a = var_11_to_fp16, b = var_1116_cast_fp16, cond = mask_3)[name = string("input_293_cast_fp16")];
+            bool x_123_transpose_x_0 = const()[name = string("x_123_transpose_x_0"), val = bool(false)];
+            bool x_123_transpose_y_0 = const()[name = string("x_123_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 64, 128]> value_13_cast_fp16 = transpose(perm = value_13_perm_0, x = v_11_cast_fp16)[name = string("transpose_274")];
+            tensor<fp16, [1, 8, 64, 128]> x_123_cast_fp16 = matmul(transpose_x = x_123_transpose_x_0, transpose_y = x_123_transpose_y_0, x = input_293_cast_fp16, y = value_13_cast_fp16)[name = string("x_123_cast_fp16")];
+            tensor<int32, [4]> var_1120_perm_0 = const()[name = string("op_1120_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1121 = const()[name = string("op_1121"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 64, 8, 128]> var_1120_cast_fp16 = transpose(perm = var_1120_perm_0, x = x_123_cast_fp16)[name = string("transpose_273")];
+            tensor<fp16, [1, 64, 1024]> input_295_cast_fp16 = reshape(shape = var_1121, x = var_1120_cast_fp16)[name = string("input_295_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523851]> module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76626560))), indices_nonzero_data = tensor<uint6, [523851]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76757696))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77150656))))[name = string("module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_52_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_295_cast_fp16)[name = string("linear_52_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_299_cast_fp16 = add(x = input_291_cast_fp16, y = linear_52_cast_fp16)[name = string("input_299_cast_fp16")];
+            tensor<int32, [1]> x_127_axes_0 = const()[name = string("x_127_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_5_norm_conv_weight_to_fp16 = const()[name = string("module_layers_5_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77167104)))];
+            tensor<fp16, [1024]> module_layers_5_norm_conv_bias_to_fp16 = const()[name = string("module_layers_5_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77169216)))];
+            tensor<fp16, [1, 64, 1024]> x_127_cast_fp16 = layer_norm(axes = x_127_axes_0, beta = module_layers_5_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_conv_weight_to_fp16, x = input_299_cast_fp16)[name = string("x_127_cast_fp16")];
+            tensor<int32, [3]> input_301_perm_0 = const()[name = string("input_301_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_303_pad_type_0 = const()[name = string("input_303_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_303_strides_0 = const()[name = string("input_303_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_303_pad_0 = const()[name = string("input_303_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_303_dilations_0 = const()[name = string("input_303_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_303_groups_0 = const()[name = string("input_303_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047810]> module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77171328))), indices_nonzero_data = tensor<uint6, [1047810]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77433536))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78219520))))[name = string("module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> input_301_cast_fp16 = transpose(perm = input_301_perm_0, x = x_127_cast_fp16)[name = string("transpose_272")];
+            tensor<fp16, [1, 2048, 64]> input_303_cast_fp16 = conv(dilations = input_303_dilations_0, groups = input_303_groups_0, pad = input_303_pad_0, pad_type = input_303_pad_type_0, strides = input_303_strides_0, weight = module_layers_5_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_301_cast_fp16)[name = string("input_303_cast_fp16")];
+            int32 x_129_split_num_splits_0 = const()[name = string("x_129_split_num_splits_0"), val = int32(2)];
+            int32 x_129_split_axis_0 = const()[name = string("x_129_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 64]> x_129_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_129_split_cast_fp16_1 = split(axis = x_129_split_axis_0, num_splits = x_129_split_num_splits_0, x = input_303_cast_fp16)[name = string("x_129_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_129_split_1_sigmoid_cast_fp16 = sigmoid(x = x_129_split_cast_fp16_1)[name = string("x_129_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_129_cast_fp16 = mul(x = x_129_split_cast_fp16_0, y = x_129_split_1_sigmoid_cast_fp16)[name = string("x_129_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_305_cast_fp16 = select(a = var_11_to_fp16, b = x_129_cast_fp16, cond = var_328)[name = string("input_305_cast_fp16")];
+            tensor<int32, [6]> input_307_pad_0 = const()[name = string("input_307_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_307_mode_0 = const()[name = string("input_307_mode_0"), val = string("constant")];
+            fp16 const_67_to_fp16 = const()[name = string("const_67_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 72]> input_307_cast_fp16 = pad(constant_val = const_67_to_fp16, mode = input_307_mode_0, pad = input_307_pad_0, x = input_305_cast_fp16)[name = string("input_307_cast_fp16")];
+            string input_309_pad_type_0 = const()[name = string("input_309_pad_type_0"), val = string("valid")];
+            int32 input_309_groups_0 = const()[name = string("input_309_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_309_strides_0 = const()[name = string("input_309_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_309_pad_0 = const()[name = string("input_309_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_309_dilations_0 = const()[name = string("input_309_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_258_to_fp16 = const()[name = string("const_258_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78252352)))];
+            tensor<fp16, [1024]> const_259_to_fp16 = const()[name = string("const_259_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78270848)))];
+            tensor<fp16, [1, 1024, 64]> input_311_cast_fp16 = conv(bias = const_259_to_fp16, dilations = input_309_dilations_0, groups = input_309_groups_0, pad = input_309_pad_0, pad_type = input_309_pad_type_0, strides = input_309_strides_0, weight = const_258_to_fp16, x = input_307_cast_fp16)[name = string("input_311_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_313_cast_fp16 = silu(x = input_311_cast_fp16)[name = string("input_313_cast_fp16")];
+            string x_131_pad_type_0 = const()[name = string("x_131_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_131_strides_0 = const()[name = string("x_131_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_131_pad_0 = const()[name = string("x_131_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_131_dilations_0 = const()[name = string("x_131_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_131_groups_0 = const()[name = string("x_131_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520509]> module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78272960))), indices_nonzero_data = tensor<uint6, [520509]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78404096))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78794560))))[name = string("module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> x_131_cast_fp16 = conv(dilations = x_131_dilations_0, groups = x_131_groups_0, pad = x_131_pad_0, pad_type = x_131_pad_type_0, strides = x_131_strides_0, weight = module_layers_5_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_313_cast_fp16)[name = string("x_131_cast_fp16")];
+            tensor<int32, [3]> input_315_perm_0 = const()[name = string("input_315_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 64, 1024]> input_315_cast_fp16 = transpose(perm = input_315_perm_0, x = x_131_cast_fp16)[name = string("transpose_271")];
+            tensor<fp16, [1, 64, 1024]> input_317_cast_fp16 = add(x = input_299_cast_fp16, y = input_315_cast_fp16)[name = string("input_317_cast_fp16")];
+            tensor<int32, [1]> input_319_axes_0 = const()[name = string("input_319_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_5_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_5_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78811008)))];
+            tensor<fp16, [1024]> module_layers_5_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_5_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78813120)))];
+            tensor<fp16, [1, 64, 1024]> input_319_cast_fp16 = layer_norm(axes = input_319_axes_0, beta = module_layers_5_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_feed_forward2_weight_to_fp16, x = input_317_cast_fp16)[name = string("input_319_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2063618]> module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78815232))), indices_nonzero_data = tensor<uint6, [2063618]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79339584))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80887424))))[name = string("module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_53_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_5_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_319_cast_fp16)[name = string("linear_53_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_323_cast_fp16 = silu(x = linear_53_cast_fp16)[name = string("input_323_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1985406]> module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80953024))), indices_nonzero_data = tensor<uint6, [1985406]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81477376))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82966528))))[name = string("module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_54_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_5_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_323_cast_fp16)[name = string("linear_54_cast_fp16")];
+            fp16 var_1181_to_fp16 = const()[name = string("op_1181_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_1182_cast_fp16 = mul(x = linear_54_cast_fp16, y = var_1181_to_fp16)[name = string("op_1182_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_329_cast_fp16 = add(x = input_317_cast_fp16, y = var_1182_cast_fp16)[name = string("input_329_cast_fp16")];
+            tensor<int32, [1]> input_331_axes_0 = const()[name = string("input_331_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_5_norm_out_weight_to_fp16 = const()[name = string("module_layers_5_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82982976)))];
+            tensor<fp16, [1024]> module_layers_5_norm_out_bias_to_fp16 = const()[name = string("module_layers_5_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82985088)))];
+            tensor<fp16, [1, 64, 1024]> input_331_cast_fp16 = layer_norm(axes = input_331_axes_0, beta = module_layers_5_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_out_weight_to_fp16, x = input_329_cast_fp16)[name = string("input_331_cast_fp16")];
+            tensor<int32, [1]> input_333_axes_0 = const()[name = string("input_333_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_6_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_6_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82987200)))];
+            tensor<fp16, [1024]> module_layers_6_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_6_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82989312)))];
+            tensor<fp16, [1, 64, 1024]> input_333_cast_fp16 = layer_norm(axes = input_333_axes_0, beta = module_layers_6_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_feed_forward1_weight_to_fp16, x = input_331_cast_fp16)[name = string("input_333_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2073214]> module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82991424))), indices_nonzero_data = tensor<uint6, [2073214]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83515776))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85070784))))[name = string("module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_6_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_333_cast_fp16)[name = string("linear_55_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_337_cast_fp16 = silu(x = linear_55_cast_fp16)[name = string("input_337_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1956749]> module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85136384))), indices_nonzero_data = tensor<uint6, [1956749]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85660736))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87128384))))[name = string("module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_56_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_337_cast_fp16)[name = string("linear_56_cast_fp16")];
+            fp16 var_1210_to_fp16 = const()[name = string("op_1210_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_1211_cast_fp16 = mul(x = linear_56_cast_fp16, y = var_1210_to_fp16)[name = string("op_1211_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_343_cast_fp16 = add(x = input_331_cast_fp16, y = var_1211_cast_fp16)[name = string("input_343_cast_fp16")];
+            tensor<int32, [1]> query_13_axes_0 = const()[name = string("query_13_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_6_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_6_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87144832)))];
+            tensor<fp16, [1024]> module_layers_6_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_6_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87146944)))];
+            tensor<fp16, [1, 64, 1024]> query_13_cast_fp16 = layer_norm(axes = query_13_axes_0, beta = module_layers_6_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_self_att_weight_to_fp16, x = input_343_cast_fp16)[name = string("query_13_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522973]> module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87149056))), indices_nonzero_data = tensor<uint6, [522973]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87280192))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87672512))))[name = string("module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_57_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_13_cast_fp16)[name = string("linear_57_cast_fp16")];
+            tensor<int32, [4]> var_1227 = const()[name = string("op_1227"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> q_37_cast_fp16 = reshape(shape = var_1227, x = linear_57_cast_fp16)[name = string("q_37_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521918]> module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87688960))), indices_nonzero_data = tensor<uint6, [521918]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87820096))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88211648))))[name = string("module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_58_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_13_cast_fp16)[name = string("linear_58_cast_fp16")];
+            tensor<int32, [4]> var_1231 = const()[name = string("op_1231"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> k_25_cast_fp16 = reshape(shape = var_1231, x = linear_58_cast_fp16)[name = string("k_25_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520459]> module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88228096))), indices_nonzero_data = tensor<uint6, [520459]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88359232))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88749696))))[name = string("module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_59_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_13_cast_fp16)[name = string("linear_59_cast_fp16")];
+            tensor<int32, [4]> var_1235 = const()[name = string("op_1235"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> v_13_cast_fp16 = reshape(shape = var_1235, x = linear_59_cast_fp16)[name = string("v_13_cast_fp16")];
+            tensor<int32, [4]> value_15_perm_0 = const()[name = string("value_15_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_6_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_6_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88766144)))];
+            tensor<fp16, [1, 64, 8, 128]> var_1247_cast_fp16 = add(x = q_37_cast_fp16, y = module_layers_6_self_attn_pos_bias_u_to_fp16)[name = string("op_1247_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_6_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_6_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88768256)))];
+            tensor<fp16, [1, 64, 8, 128]> var_1249_cast_fp16 = add(x = q_37_cast_fp16, y = module_layers_6_self_attn_pos_bias_v_to_fp16)[name = string("op_1249_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_13_perm_0 = const()[name = string("q_with_bias_v_13_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_139_transpose_x_0 = const()[name = string("x_139_transpose_x_0"), val = bool(false)];
+            bool x_139_transpose_y_0 = const()[name = string("x_139_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 127]> var_1251_to_fp16 = const()[name = string("op_1251_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88770368)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_13_cast_fp16 = transpose(perm = q_with_bias_v_13_perm_0, x = var_1249_cast_fp16)[name = string("transpose_270")];
+            tensor<fp16, [1, 8, 64, 127]> x_139_cast_fp16 = matmul(transpose_x = x_139_transpose_x_0, transpose_y = x_139_transpose_y_0, x = q_with_bias_v_13_cast_fp16, y = var_1251_to_fp16)[name = string("x_139_cast_fp16")];
+            tensor<int32, [8]> x_141_pad_0 = const()[name = string("x_141_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_141_mode_0 = const()[name = string("x_141_mode_0"), val = string("constant")];
+            fp16 const_74_to_fp16 = const()[name = string("const_74_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 64, 128]> x_141_cast_fp16 = pad(constant_val = const_74_to_fp16, mode = x_141_mode_0, pad = x_141_pad_0, x = x_139_cast_fp16)[name = string("x_141_cast_fp16")];
+            tensor<int32, [4]> var_1259 = const()[name = string("op_1259"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_143_cast_fp16 = reshape(shape = var_1259, x = x_141_cast_fp16)[name = string("x_143_cast_fp16")];
+            tensor<int32, [4]> var_1263_begin_0 = const()[name = string("op_1263_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1263_end_0 = const()[name = string("op_1263_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
+            tensor<bool, [4]> var_1263_end_mask_0 = const()[name = string("op_1263_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 127, 64]> var_1263_cast_fp16 = slice_by_index(begin = var_1263_begin_0, end = var_1263_end_0, end_mask = var_1263_end_mask_0, x = x_143_cast_fp16)[name = string("op_1263_cast_fp16")];
+            tensor<int32, [4]> var_1264 = const()[name = string("op_1264"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_25_cast_fp16 = reshape(shape = var_1264, x = var_1263_cast_fp16)[name = string("matrix_bd_25_cast_fp16")];
+            bool matrix_ac_13_transpose_x_0 = const()[name = string("matrix_ac_13_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_13_transpose_y_0 = const()[name = string("matrix_ac_13_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_108_perm_0 = const()[name = string("transpose_108_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_109_perm_0 = const()[name = string("transpose_109_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 64]> transpose_109 = transpose(perm = transpose_109_perm_0, x = k_25_cast_fp16)[name = string("transpose_268")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_108 = transpose(perm = transpose_108_perm_0, x = var_1247_cast_fp16)[name = string("transpose_269")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_13_cast_fp16 = matmul(transpose_x = matrix_ac_13_transpose_x_0, transpose_y = matrix_ac_13_transpose_y_0, x = transpose_108, y = transpose_109)[name = string("matrix_ac_13_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_27_begin_0 = const()[name = string("matrix_bd_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_27_end_0 = const()[name = string("matrix_bd_27_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
+            tensor<bool, [4]> matrix_bd_27_end_mask_0 = const()[name = string("matrix_bd_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_27_cast_fp16 = slice_by_index(begin = matrix_bd_27_begin_0, end = matrix_bd_27_end_0, end_mask = matrix_bd_27_end_mask_0, x = matrix_bd_25_cast_fp16)[name = string("matrix_bd_27_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_1273_cast_fp16 = add(x = matrix_ac_13_cast_fp16, y = matrix_bd_27_cast_fp16)[name = string("op_1273_cast_fp16")];
+            fp16 _inversed_scores_25_y_0_to_fp16 = const()[name = string("_inversed_scores_25_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_25_cast_fp16 = mul(x = var_1273_cast_fp16, y = _inversed_scores_25_y_0_to_fp16)[name = string("_inversed_scores_25_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_27_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_25_cast_fp16, cond = mask_3)[name = string("scores_27_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_1279_cast_fp16 = softmax(axis = var_30, x = scores_27_cast_fp16)[name = string("op_1279_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_345_cast_fp16 = select(a = var_11_to_fp16, b = var_1279_cast_fp16, cond = mask_3)[name = string("input_345_cast_fp16")];
+            bool x_145_transpose_x_0 = const()[name = string("x_145_transpose_x_0"), val = bool(false)];
+            bool x_145_transpose_y_0 = const()[name = string("x_145_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 64, 128]> value_15_cast_fp16 = transpose(perm = value_15_perm_0, x = v_13_cast_fp16)[name = string("transpose_267")];
+            tensor<fp16, [1, 8, 64, 128]> x_145_cast_fp16 = matmul(transpose_x = x_145_transpose_x_0, transpose_y = x_145_transpose_y_0, x = input_345_cast_fp16, y = value_15_cast_fp16)[name = string("x_145_cast_fp16")];
+            tensor<int32, [4]> var_1283_perm_0 = const()[name = string("op_1283_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1284 = const()[name = string("op_1284"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 64, 8, 128]> var_1283_cast_fp16 = transpose(perm = var_1283_perm_0, x = x_145_cast_fp16)[name = string("transpose_266")];
+            tensor<fp16, [1, 64, 1024]> input_347_cast_fp16 = reshape(shape = var_1284, x = var_1283_cast_fp16)[name = string("input_347_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524874]> module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89030528))), indices_nonzero_data = tensor<uint6, [524874]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89161664))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89555392))))[name = string("module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_61_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_347_cast_fp16)[name = string("linear_61_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_351_cast_fp16 = add(x = input_343_cast_fp16, y = linear_61_cast_fp16)[name = string("input_351_cast_fp16")];
+            tensor<int32, [1]> x_149_axes_0 = const()[name = string("x_149_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_6_norm_conv_weight_to_fp16 = const()[name = string("module_layers_6_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89571840)))];
+            tensor<fp16, [1024]> module_layers_6_norm_conv_bias_to_fp16 = const()[name = string("module_layers_6_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89573952)))];
+            tensor<fp16, [1, 64, 1024]> x_149_cast_fp16 = layer_norm(axes = x_149_axes_0, beta = module_layers_6_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_conv_weight_to_fp16, x = input_351_cast_fp16)[name = string("x_149_cast_fp16")];
+            tensor<int32, [3]> input_353_perm_0 = const()[name = string("input_353_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_355_pad_type_0 = const()[name = string("input_355_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_355_strides_0 = const()[name = string("input_355_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_355_pad_0 = const()[name = string("input_355_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_355_dilations_0 = const()[name = string("input_355_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_355_groups_0 = const()[name = string("input_355_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1048536]> module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89576064))), indices_nonzero_data = tensor<uint6, [1048536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89838272))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90624768))))[name = string("module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> input_353_cast_fp16 = transpose(perm = input_353_perm_0, x = x_149_cast_fp16)[name = string("transpose_265")];
+            tensor<fp16, [1, 2048, 64]> input_355_cast_fp16 = conv(dilations = input_355_dilations_0, groups = input_355_groups_0, pad = input_355_pad_0, pad_type = input_355_pad_type_0, strides = input_355_strides_0, weight = module_layers_6_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_353_cast_fp16)[name = string("input_355_cast_fp16")];
+            int32 x_151_split_num_splits_0 = const()[name = string("x_151_split_num_splits_0"), val = int32(2)];
+            int32 x_151_split_axis_0 = const()[name = string("x_151_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 64]> x_151_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_151_split_cast_fp16_1 = split(axis = x_151_split_axis_0, num_splits = x_151_split_num_splits_0, x = input_355_cast_fp16)[name = string("x_151_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_151_split_1_sigmoid_cast_fp16 = sigmoid(x = x_151_split_cast_fp16_1)[name = string("x_151_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_151_cast_fp16 = mul(x = x_151_split_cast_fp16_0, y = x_151_split_1_sigmoid_cast_fp16)[name = string("x_151_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_357_cast_fp16 = select(a = var_11_to_fp16, b = x_151_cast_fp16, cond = var_328)[name = string("input_357_cast_fp16")];
+            tensor<int32, [6]> input_359_pad_0 = const()[name = string("input_359_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_359_mode_0 = const()[name = string("input_359_mode_0"), val = string("constant")];
+            fp16 const_77_to_fp16 = const()[name = string("const_77_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 72]> input_359_cast_fp16 = pad(constant_val = const_77_to_fp16, mode = input_359_mode_0, pad = input_359_pad_0, x = input_357_cast_fp16)[name = string("input_359_cast_fp16")];
+            string input_361_pad_type_0 = const()[name = string("input_361_pad_type_0"), val = string("valid")];
+            int32 input_361_groups_0 = const()[name = string("input_361_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_361_strides_0 = const()[name = string("input_361_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_361_pad_0 = const()[name = string("input_361_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_361_dilations_0 = const()[name = string("input_361_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_260_to_fp16 = const()[name = string("const_260_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90657600)))];
+            tensor<fp16, [1024]> const_261_to_fp16 = const()[name = string("const_261_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90676096)))];
+            tensor<fp16, [1, 1024, 64]> input_363_cast_fp16 = conv(bias = const_261_to_fp16, dilations = input_361_dilations_0, groups = input_361_groups_0, pad = input_361_pad_0, pad_type = input_361_pad_type_0, strides = input_361_strides_0, weight = const_260_to_fp16, x = input_359_cast_fp16)[name = string("input_363_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_365_cast_fp16 = silu(x = input_363_cast_fp16)[name = string("input_365_cast_fp16")];
+            string x_153_pad_type_0 = const()[name = string("x_153_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_153_strides_0 = const()[name = string("x_153_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_153_pad_0 = const()[name = string("x_153_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_153_dilations_0 = const()[name = string("x_153_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_153_groups_0 = const()[name = string("x_153_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519645]> module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90678208))), indices_nonzero_data = tensor<uint6, [519645]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90809344))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91199168))))[name = string("module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> x_153_cast_fp16 = conv(dilations = x_153_dilations_0, groups = x_153_groups_0, pad = x_153_pad_0, pad_type = x_153_pad_type_0, strides = x_153_strides_0, weight = module_layers_6_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_365_cast_fp16)[name = string("x_153_cast_fp16")];
+            tensor<int32, [3]> input_367_perm_0 = const()[name = string("input_367_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 64, 1024]> input_367_cast_fp16 = transpose(perm = input_367_perm_0, x = x_153_cast_fp16)[name = string("transpose_264")];
+            tensor<fp16, [1, 64, 1024]> input_369_cast_fp16 = add(x = input_351_cast_fp16, y = input_367_cast_fp16)[name = string("input_369_cast_fp16")];
+            tensor<int32, [1]> input_371_axes_0 = const()[name = string("input_371_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_6_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_6_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91215616)))];
+            tensor<fp16, [1024]> module_layers_6_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_6_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91217728)))];
+            tensor<fp16, [1, 64, 1024]> input_371_cast_fp16 = layer_norm(axes = input_371_axes_0, beta = module_layers_6_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_feed_forward2_weight_to_fp16, x = input_369_cast_fp16)[name = string("input_371_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2063870]> module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91219840))), indices_nonzero_data = tensor<uint6, [2063870]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91744192))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93292160))))[name = string("module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_62_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_6_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_371_cast_fp16)[name = string("linear_62_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_375_cast_fp16 = silu(x = linear_62_cast_fp16)[name = string("input_375_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2019355]> module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93357760))), indices_nonzero_data = tensor<uint6, [2019355]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93882112))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95396736))))[name = string("module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_63_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_6_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_375_cast_fp16)[name = string("linear_63_cast_fp16")];
+            fp16 var_1344_to_fp16 = const()[name = string("op_1344_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_1345_cast_fp16 = mul(x = linear_63_cast_fp16, y = var_1344_to_fp16)[name = string("op_1345_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_381_cast_fp16 = add(x = input_369_cast_fp16, y = var_1345_cast_fp16)[name = string("input_381_cast_fp16")];
+            tensor<int32, [1]> input_383_axes_0 = const()[name = string("input_383_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_6_norm_out_weight_to_fp16 = const()[name = string("module_layers_6_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95413184)))];
+            tensor<fp16, [1024]> module_layers_6_norm_out_bias_to_fp16 = const()[name = string("module_layers_6_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95415296)))];
+            tensor<fp16, [1, 64, 1024]> input_383_cast_fp16 = layer_norm(axes = input_383_axes_0, beta = module_layers_6_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_out_weight_to_fp16, x = input_381_cast_fp16)[name = string("input_383_cast_fp16")];
+            tensor<int32, [1]> input_385_axes_0 = const()[name = string("input_385_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_7_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_7_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95417408)))];
+            tensor<fp16, [1024]> module_layers_7_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_7_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95419520)))];
+            tensor<fp16, [1, 64, 1024]> input_385_cast_fp16 = layer_norm(axes = input_385_axes_0, beta = module_layers_7_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_feed_forward1_weight_to_fp16, x = input_383_cast_fp16)[name = string("input_385_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2072252]> module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95421632))), indices_nonzero_data = tensor<uint6, [2072252]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95945984))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97500288))))[name = string("module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_64_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_7_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_385_cast_fp16)[name = string("linear_64_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_389_cast_fp16 = silu(x = linear_64_cast_fp16)[name = string("input_389_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1978853]> module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97565888))), indices_nonzero_data = tensor<uint6, [1978853]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98090240))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99574464))))[name = string("module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_65_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_389_cast_fp16)[name = string("linear_65_cast_fp16")];
+            fp16 var_1373_to_fp16 = const()[name = string("op_1373_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_1374_cast_fp16 = mul(x = linear_65_cast_fp16, y = var_1373_to_fp16)[name = string("op_1374_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_395_cast_fp16 = add(x = input_383_cast_fp16, y = var_1374_cast_fp16)[name = string("input_395_cast_fp16")];
+            tensor<int32, [1]> query_15_axes_0 = const()[name = string("query_15_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_7_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_7_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99590912)))];
+            tensor<fp16, [1024]> module_layers_7_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_7_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99593024)))];
+            tensor<fp16, [1, 64, 1024]> query_15_cast_fp16 = layer_norm(axes = query_15_axes_0, beta = module_layers_7_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_self_att_weight_to_fp16, x = input_395_cast_fp16)[name = string("query_15_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522949]> module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99595136))), indices_nonzero_data = tensor<uint6, [522949]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99726272))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100118592))))[name = string("module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_66_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_15_cast_fp16)[name = string("linear_66_cast_fp16")];
+            tensor<int32, [4]> var_1390 = const()[name = string("op_1390"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> q_43_cast_fp16 = reshape(shape = var_1390, x = linear_66_cast_fp16)[name = string("q_43_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [515713]> module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100135040))), indices_nonzero_data = tensor<uint6, [515713]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100266176))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100653056))))[name = string("module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_67_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_15_cast_fp16)[name = string("linear_67_cast_fp16")];
+            tensor<int32, [4]> var_1394 = const()[name = string("op_1394"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> k_29_cast_fp16 = reshape(shape = var_1394, x = linear_67_cast_fp16)[name = string("k_29_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [507764]> module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100669504))), indices_nonzero_data = tensor<uint6, [507764]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100800640))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101181568))))[name = string("module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_68_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_15_cast_fp16)[name = string("linear_68_cast_fp16")];
+            tensor<int32, [4]> var_1398 = const()[name = string("op_1398"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> v_15_cast_fp16 = reshape(shape = var_1398, x = linear_68_cast_fp16)[name = string("v_15_cast_fp16")];
+            tensor<int32, [4]> value_17_perm_0 = const()[name = string("value_17_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_7_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_7_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101198016)))];
+            tensor<fp16, [1, 64, 8, 128]> var_1410_cast_fp16 = add(x = q_43_cast_fp16, y = module_layers_7_self_attn_pos_bias_u_to_fp16)[name = string("op_1410_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_7_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_7_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101200128)))];
+            tensor<fp16, [1, 64, 8, 128]> var_1412_cast_fp16 = add(x = q_43_cast_fp16, y = module_layers_7_self_attn_pos_bias_v_to_fp16)[name = string("op_1412_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_15_perm_0 = const()[name = string("q_with_bias_v_15_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_161_transpose_x_0 = const()[name = string("x_161_transpose_x_0"), val = bool(false)];
+            bool x_161_transpose_y_0 = const()[name = string("x_161_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 127]> var_1414_to_fp16 = const()[name = string("op_1414_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101202240)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_15_cast_fp16 = transpose(perm = q_with_bias_v_15_perm_0, x = var_1412_cast_fp16)[name = string("transpose_263")];
+            tensor<fp16, [1, 8, 64, 127]> x_161_cast_fp16 = matmul(transpose_x = x_161_transpose_x_0, transpose_y = x_161_transpose_y_0, x = q_with_bias_v_15_cast_fp16, y = var_1414_to_fp16)[name = string("x_161_cast_fp16")];
+            tensor<int32, [8]> x_163_pad_0 = const()[name = string("x_163_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_163_mode_0 = const()[name = string("x_163_mode_0"), val = string("constant")];
+            fp16 const_84_to_fp16 = const()[name = string("const_84_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 64, 128]> x_163_cast_fp16 = pad(constant_val = const_84_to_fp16, mode = x_163_mode_0, pad = x_163_pad_0, x = x_161_cast_fp16)[name = string("x_163_cast_fp16")];
+            tensor<int32, [4]> var_1422 = const()[name = string("op_1422"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_165_cast_fp16 = reshape(shape = var_1422, x = x_163_cast_fp16)[name = string("x_165_cast_fp16")];
+            tensor<int32, [4]> var_1426_begin_0 = const()[name = string("op_1426_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1426_end_0 = const()[name = string("op_1426_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
+            tensor<bool, [4]> var_1426_end_mask_0 = const()[name = string("op_1426_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 127, 64]> var_1426_cast_fp16 = slice_by_index(begin = var_1426_begin_0, end = var_1426_end_0, end_mask = var_1426_end_mask_0, x = x_165_cast_fp16)[name = string("op_1426_cast_fp16")];
+            tensor<int32, [4]> var_1427 = const()[name = string("op_1427"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_29_cast_fp16 = reshape(shape = var_1427, x = var_1426_cast_fp16)[name = string("matrix_bd_29_cast_fp16")];
+            bool matrix_ac_15_transpose_x_0 = const()[name = string("matrix_ac_15_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_15_transpose_y_0 = const()[name = string("matrix_ac_15_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_110_perm_0 = const()[name = string("transpose_110_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_111_perm_0 = const()[name = string("transpose_111_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 64]> transpose_111 = transpose(perm = transpose_111_perm_0, x = k_29_cast_fp16)[name = string("transpose_261")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_110 = transpose(perm = transpose_110_perm_0, x = var_1410_cast_fp16)[name = string("transpose_262")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_15_cast_fp16 = matmul(transpose_x = matrix_ac_15_transpose_x_0, transpose_y = matrix_ac_15_transpose_y_0, x = transpose_110, y = transpose_111)[name = string("matrix_ac_15_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_31_begin_0 = const()[name = string("matrix_bd_31_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_31_end_0 = const()[name = string("matrix_bd_31_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
+            tensor<bool, [4]> matrix_bd_31_end_mask_0 = const()[name = string("matrix_bd_31_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_31_cast_fp16 = slice_by_index(begin = matrix_bd_31_begin_0, end = matrix_bd_31_end_0, end_mask = matrix_bd_31_end_mask_0, x = matrix_bd_29_cast_fp16)[name = string("matrix_bd_31_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_1436_cast_fp16 = add(x = matrix_ac_15_cast_fp16, y = matrix_bd_31_cast_fp16)[name = string("op_1436_cast_fp16")];
+            fp16 _inversed_scores_29_y_0_to_fp16 = const()[name = string("_inversed_scores_29_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_29_cast_fp16 = mul(x = var_1436_cast_fp16, y = _inversed_scores_29_y_0_to_fp16)[name = string("_inversed_scores_29_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_31_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_29_cast_fp16, cond = mask_3)[name = string("scores_31_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_1442_cast_fp16 = softmax(axis = var_30, x = scores_31_cast_fp16)[name = string("op_1442_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_397_cast_fp16 = select(a = var_11_to_fp16, b = var_1442_cast_fp16, cond = mask_3)[name = string("input_397_cast_fp16")];
+            bool x_167_transpose_x_0 = const()[name = string("x_167_transpose_x_0"), val = bool(false)];
+            bool x_167_transpose_y_0 = const()[name = string("x_167_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 64, 128]> value_17_cast_fp16 = transpose(perm = value_17_perm_0, x = v_15_cast_fp16)[name = string("transpose_260")];
+            tensor<fp16, [1, 8, 64, 128]> x_167_cast_fp16 = matmul(transpose_x = x_167_transpose_x_0, transpose_y = x_167_transpose_y_0, x = input_397_cast_fp16, y = value_17_cast_fp16)[name = string("x_167_cast_fp16")];
+            tensor<int32, [4]> var_1446_perm_0 = const()[name = string("op_1446_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1447 = const()[name = string("op_1447"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 64, 8, 128]> var_1446_cast_fp16 = transpose(perm = var_1446_perm_0, x = x_167_cast_fp16)[name = string("transpose_259")];
+            tensor<fp16, [1, 64, 1024]> input_399_cast_fp16 = reshape(shape = var_1447, x = var_1446_cast_fp16)[name = string("input_399_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525262]> module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101462400))), indices_nonzero_data = tensor<uint6, [525262]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101593536))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101987584))))[name = string("module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_70_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_399_cast_fp16)[name = string("linear_70_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_403_cast_fp16 = add(x = input_395_cast_fp16, y = linear_70_cast_fp16)[name = string("input_403_cast_fp16")];
+            tensor<int32, [1]> x_171_axes_0 = const()[name = string("x_171_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_7_norm_conv_weight_to_fp16 = const()[name = string("module_layers_7_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102004032)))];
+            tensor<fp16, [1024]> module_layers_7_norm_conv_bias_to_fp16 = const()[name = string("module_layers_7_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102006144)))];
+            tensor<fp16, [1, 64, 1024]> x_171_cast_fp16 = layer_norm(axes = x_171_axes_0, beta = module_layers_7_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_conv_weight_to_fp16, x = input_403_cast_fp16)[name = string("x_171_cast_fp16")];
+            tensor<int32, [3]> input_405_perm_0 = const()[name = string("input_405_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_407_pad_type_0 = const()[name = string("input_407_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_407_strides_0 = const()[name = string("input_407_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_407_pad_0 = const()[name = string("input_407_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_407_dilations_0 = const()[name = string("input_407_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_407_groups_0 = const()[name = string("input_407_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047562]> module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102008256))), indices_nonzero_data = tensor<uint6, [1047562]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102270464))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103056256))))[name = string("module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> input_405_cast_fp16 = transpose(perm = input_405_perm_0, x = x_171_cast_fp16)[name = string("transpose_258")];
+            tensor<fp16, [1, 2048, 64]> input_407_cast_fp16 = conv(dilations = input_407_dilations_0, groups = input_407_groups_0, pad = input_407_pad_0, pad_type = input_407_pad_type_0, strides = input_407_strides_0, weight = module_layers_7_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_405_cast_fp16)[name = string("input_407_cast_fp16")];
+            int32 x_173_split_num_splits_0 = const()[name = string("x_173_split_num_splits_0"), val = int32(2)];
+            int32 x_173_split_axis_0 = const()[name = string("x_173_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 64]> x_173_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_173_split_cast_fp16_1 = split(axis = x_173_split_axis_0, num_splits = x_173_split_num_splits_0, x = input_407_cast_fp16)[name = string("x_173_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_173_split_1_sigmoid_cast_fp16 = sigmoid(x = x_173_split_cast_fp16_1)[name = string("x_173_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_173_cast_fp16 = mul(x = x_173_split_cast_fp16_0, y = x_173_split_1_sigmoid_cast_fp16)[name = string("x_173_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_409_cast_fp16 = select(a = var_11_to_fp16, b = x_173_cast_fp16, cond = var_328)[name = string("input_409_cast_fp16")];
+            tensor<int32, [6]> input_411_pad_0 = const()[name = string("input_411_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_411_mode_0 = const()[name = string("input_411_mode_0"), val = string("constant")];
+            fp16 const_87_to_fp16 = const()[name = string("const_87_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 72]> input_411_cast_fp16 = pad(constant_val = const_87_to_fp16, mode = input_411_mode_0, pad = input_411_pad_0, x = input_409_cast_fp16)[name = string("input_411_cast_fp16")];
+            string input_413_pad_type_0 = const()[name = string("input_413_pad_type_0"), val = string("valid")];
+            int32 input_413_groups_0 = const()[name = string("input_413_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_413_strides_0 = const()[name = string("input_413_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_413_pad_0 = const()[name = string("input_413_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_413_dilations_0 = const()[name = string("input_413_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_262_to_fp16 = const()[name = string("const_262_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103089088)))];
+            tensor<fp16, [1024]> const_263_to_fp16 = const()[name = string("const_263_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103107584)))];
+            tensor<fp16, [1, 1024, 64]> input_415_cast_fp16 = conv(bias = const_263_to_fp16, dilations = input_413_dilations_0, groups = input_413_groups_0, pad = input_413_pad_0, pad_type = input_413_pad_type_0, strides = input_413_strides_0, weight = const_262_to_fp16, x = input_411_cast_fp16)[name = string("input_415_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_417_cast_fp16 = silu(x = input_415_cast_fp16)[name = string("input_417_cast_fp16")];
+            string x_175_pad_type_0 = const()[name = string("x_175_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_175_strides_0 = const()[name = string("x_175_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_175_pad_0 = const()[name = string("x_175_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_175_dilations_0 = const()[name = string("x_175_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_175_groups_0 = const()[name = string("x_175_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519522]> module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103109696))), indices_nonzero_data = tensor<uint6, [519522]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103240832))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103630592))))[name = string("module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> x_175_cast_fp16 = conv(dilations = x_175_dilations_0, groups = x_175_groups_0, pad = x_175_pad_0, pad_type = x_175_pad_type_0, strides = x_175_strides_0, weight = module_layers_7_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_417_cast_fp16)[name = string("x_175_cast_fp16")];
+            tensor<int32, [3]> input_419_perm_0 = const()[name = string("input_419_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 64, 1024]> input_419_cast_fp16 = transpose(perm = input_419_perm_0, x = x_175_cast_fp16)[name = string("transpose_257")];
+            tensor<fp16, [1, 64, 1024]> input_421_cast_fp16 = add(x = input_403_cast_fp16, y = input_419_cast_fp16)[name = string("input_421_cast_fp16")];
+            tensor<int32, [1]> input_423_axes_0 = const()[name = string("input_423_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_7_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_7_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103647040)))];
+            tensor<fp16, [1024]> module_layers_7_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_7_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103649152)))];
+            tensor<fp16, [1, 64, 1024]> input_423_cast_fp16 = layer_norm(axes = input_423_axes_0, beta = module_layers_7_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_feed_forward2_weight_to_fp16, x = input_421_cast_fp16)[name = string("input_423_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2062964]> module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103651264))), indices_nonzero_data = tensor<uint6, [2062964]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104175616))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105722944))))[name = string("module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_71_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_7_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_423_cast_fp16)[name = string("linear_71_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_427_cast_fp16 = silu(x = linear_71_cast_fp16)[name = string("input_427_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2021791]> module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105788544))), indices_nonzero_data = tensor<uint6, [2021791]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106312896))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107829312))))[name = string("module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_72_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_7_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_427_cast_fp16)[name = string("linear_72_cast_fp16")];
+            fp16 var_1507_to_fp16 = const()[name = string("op_1507_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_1508_cast_fp16 = mul(x = linear_72_cast_fp16, y = var_1507_to_fp16)[name = string("op_1508_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_433_cast_fp16 = add(x = input_421_cast_fp16, y = var_1508_cast_fp16)[name = string("input_433_cast_fp16")];
+            tensor<int32, [1]> input_435_axes_0 = const()[name = string("input_435_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_7_norm_out_weight_to_fp16 = const()[name = string("module_layers_7_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107845760)))];
+            tensor<fp16, [1024]> module_layers_7_norm_out_bias_to_fp16 = const()[name = string("module_layers_7_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107847872)))];
+            tensor<fp16, [1, 64, 1024]> input_435_cast_fp16 = layer_norm(axes = input_435_axes_0, beta = module_layers_7_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_out_weight_to_fp16, x = input_433_cast_fp16)[name = string("input_435_cast_fp16")];
+            tensor<int32, [1]> input_437_axes_0 = const()[name = string("input_437_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_8_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_8_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107849984)))];
+            tensor<fp16, [1024]> module_layers_8_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_8_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107852096)))];
+            tensor<fp16, [1, 64, 1024]> input_437_cast_fp16 = layer_norm(axes = input_437_axes_0, beta = module_layers_8_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_feed_forward1_weight_to_fp16, x = input_435_cast_fp16)[name = string("input_437_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2070011]> module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107854208))), indices_nonzero_data = tensor<uint6, [2070011]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108378560))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109931136))))[name = string("module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_73_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_8_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_437_cast_fp16)[name = string("linear_73_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_441_cast_fp16 = silu(x = linear_73_cast_fp16)[name = string("input_441_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1995292]> module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109996736))), indices_nonzero_data = tensor<uint6, [1995292]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110521088))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112017664))))[name = string("module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_74_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_441_cast_fp16)[name = string("linear_74_cast_fp16")];
+            fp16 var_1536_to_fp16 = const()[name = string("op_1536_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_1537_cast_fp16 = mul(x = linear_74_cast_fp16, y = var_1536_to_fp16)[name = string("op_1537_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_447_cast_fp16 = add(x = input_435_cast_fp16, y = var_1537_cast_fp16)[name = string("input_447_cast_fp16")];
+            tensor<int32, [1]> query_17_axes_0 = const()[name = string("query_17_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_8_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_8_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112034112)))];
+            tensor<fp16, [1024]> module_layers_8_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_8_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112036224)))];
+            tensor<fp16, [1, 64, 1024]> query_17_cast_fp16 = layer_norm(axes = query_17_axes_0, beta = module_layers_8_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_self_att_weight_to_fp16, x = input_447_cast_fp16)[name = string("query_17_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522805]> module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112038336))), indices_nonzero_data = tensor<uint6, [522805]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112169472))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112561664))))[name = string("module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_75_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_17_cast_fp16)[name = string("linear_75_cast_fp16")];
+            tensor<int32, [4]> var_1553 = const()[name = string("op_1553"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> q_49_cast_fp16 = reshape(shape = var_1553, x = linear_75_cast_fp16)[name = string("q_49_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521269]> module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112578112))), indices_nonzero_data = tensor<uint6, [521269]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112709248))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113100288))))[name = string("module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_76_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_17_cast_fp16)[name = string("linear_76_cast_fp16")];
+            tensor<int32, [4]> var_1557 = const()[name = string("op_1557"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> k_33_cast_fp16 = reshape(shape = var_1557, x = linear_76_cast_fp16)[name = string("k_33_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [517626]> module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113116736))), indices_nonzero_data = tensor<uint6, [517626]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113247872))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113636160))))[name = string("module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_77_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_17_cast_fp16)[name = string("linear_77_cast_fp16")];
+            tensor<int32, [4]> var_1561 = const()[name = string("op_1561"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> v_17_cast_fp16 = reshape(shape = var_1561, x = linear_77_cast_fp16)[name = string("v_17_cast_fp16")];
+            tensor<int32, [4]> value_19_perm_0 = const()[name = string("value_19_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_8_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_8_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113652608)))];
+            tensor<fp16, [1, 64, 8, 128]> var_1573_cast_fp16 = add(x = q_49_cast_fp16, y = module_layers_8_self_attn_pos_bias_u_to_fp16)[name = string("op_1573_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_8_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_8_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113654720)))];
+            tensor<fp16, [1, 64, 8, 128]> var_1575_cast_fp16 = add(x = q_49_cast_fp16, y = module_layers_8_self_attn_pos_bias_v_to_fp16)[name = string("op_1575_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_17_perm_0 = const()[name = string("q_with_bias_v_17_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_183_transpose_x_0 = const()[name = string("x_183_transpose_x_0"), val = bool(false)];
+            bool x_183_transpose_y_0 = const()[name = string("x_183_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 127]> var_1577_to_fp16 = const()[name = string("op_1577_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113656832)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_17_cast_fp16 = transpose(perm = q_with_bias_v_17_perm_0, x = var_1575_cast_fp16)[name = string("transpose_256")];
+            tensor<fp16, [1, 8, 64, 127]> x_183_cast_fp16 = matmul(transpose_x = x_183_transpose_x_0, transpose_y = x_183_transpose_y_0, x = q_with_bias_v_17_cast_fp16, y = var_1577_to_fp16)[name = string("x_183_cast_fp16")];
+            tensor<int32, [8]> x_185_pad_0 = const()[name = string("x_185_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_185_mode_0 = const()[name = string("x_185_mode_0"), val = string("constant")];
+            fp16 const_94_to_fp16 = const()[name = string("const_94_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 64, 128]> x_185_cast_fp16 = pad(constant_val = const_94_to_fp16, mode = x_185_mode_0, pad = x_185_pad_0, x = x_183_cast_fp16)[name = string("x_185_cast_fp16")];
+            tensor<int32, [4]> var_1585 = const()[name = string("op_1585"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_187_cast_fp16 = reshape(shape = var_1585, x = x_185_cast_fp16)[name = string("x_187_cast_fp16")];
+            tensor<int32, [4]> var_1589_begin_0 = const()[name = string("op_1589_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1589_end_0 = const()[name = string("op_1589_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
+            tensor<bool, [4]> var_1589_end_mask_0 = const()[name = string("op_1589_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 127, 64]> var_1589_cast_fp16 = slice_by_index(begin = var_1589_begin_0, end = var_1589_end_0, end_mask = var_1589_end_mask_0, x = x_187_cast_fp16)[name = string("op_1589_cast_fp16")];
+            tensor<int32, [4]> var_1590 = const()[name = string("op_1590"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_33_cast_fp16 = reshape(shape = var_1590, x = var_1589_cast_fp16)[name = string("matrix_bd_33_cast_fp16")];
+            bool matrix_ac_17_transpose_x_0 = const()[name = string("matrix_ac_17_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_17_transpose_y_0 = const()[name = string("matrix_ac_17_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_112_perm_0 = const()[name = string("transpose_112_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_113_perm_0 = const()[name = string("transpose_113_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 64]> transpose_113 = transpose(perm = transpose_113_perm_0, x = k_33_cast_fp16)[name = string("transpose_254")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_112 = transpose(perm = transpose_112_perm_0, x = var_1573_cast_fp16)[name = string("transpose_255")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_17_cast_fp16 = matmul(transpose_x = matrix_ac_17_transpose_x_0, transpose_y = matrix_ac_17_transpose_y_0, x = transpose_112, y = transpose_113)[name = string("matrix_ac_17_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_35_begin_0 = const()[name = string("matrix_bd_35_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_35_end_0 = const()[name = string("matrix_bd_35_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
+            tensor<bool, [4]> matrix_bd_35_end_mask_0 = const()[name = string("matrix_bd_35_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_35_cast_fp16 = slice_by_index(begin = matrix_bd_35_begin_0, end = matrix_bd_35_end_0, end_mask = matrix_bd_35_end_mask_0, x = matrix_bd_33_cast_fp16)[name = string("matrix_bd_35_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_1599_cast_fp16 = add(x = matrix_ac_17_cast_fp16, y = matrix_bd_35_cast_fp16)[name = string("op_1599_cast_fp16")];
+            fp16 _inversed_scores_33_y_0_to_fp16 = const()[name = string("_inversed_scores_33_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_33_cast_fp16 = mul(x = var_1599_cast_fp16, y = _inversed_scores_33_y_0_to_fp16)[name = string("_inversed_scores_33_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_35_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_33_cast_fp16, cond = mask_3)[name = string("scores_35_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_1605_cast_fp16 = softmax(axis = var_30, x = scores_35_cast_fp16)[name = string("op_1605_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_449_cast_fp16 = select(a = var_11_to_fp16, b = var_1605_cast_fp16, cond = mask_3)[name = string("input_449_cast_fp16")];
+            bool x_189_transpose_x_0 = const()[name = string("x_189_transpose_x_0"), val = bool(false)];
+            bool x_189_transpose_y_0 = const()[name = string("x_189_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 64, 128]> value_19_cast_fp16 = transpose(perm = value_19_perm_0, x = v_17_cast_fp16)[name = string("transpose_253")];
+            tensor<fp16, [1, 8, 64, 128]> x_189_cast_fp16 = matmul(transpose_x = x_189_transpose_x_0, transpose_y = x_189_transpose_y_0, x = input_449_cast_fp16, y = value_19_cast_fp16)[name = string("x_189_cast_fp16")];
+            tensor<int32, [4]> var_1609_perm_0 = const()[name = string("op_1609_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1610 = const()[name = string("op_1610"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 64, 8, 128]> var_1609_cast_fp16 = transpose(perm = var_1609_perm_0, x = x_189_cast_fp16)[name = string("transpose_252")];
+            tensor<fp16, [1, 64, 1024]> input_451_cast_fp16 = reshape(shape = var_1610, x = var_1609_cast_fp16)[name = string("input_451_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522953]> module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113916992))), indices_nonzero_data = tensor<uint6, [522953]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114048128))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114440448))))[name = string("module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_79_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_451_cast_fp16)[name = string("linear_79_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_455_cast_fp16 = add(x = input_447_cast_fp16, y = linear_79_cast_fp16)[name = string("input_455_cast_fp16")];
+            tensor<int32, [1]> x_193_axes_0 = const()[name = string("x_193_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_8_norm_conv_weight_to_fp16 = const()[name = string("module_layers_8_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114456896)))];
+            tensor<fp16, [1024]> module_layers_8_norm_conv_bias_to_fp16 = const()[name = string("module_layers_8_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114459008)))];
+            tensor<fp16, [1, 64, 1024]> x_193_cast_fp16 = layer_norm(axes = x_193_axes_0, beta = module_layers_8_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_conv_weight_to_fp16, x = input_455_cast_fp16)[name = string("x_193_cast_fp16")];
+            tensor<int32, [3]> input_457_perm_0 = const()[name = string("input_457_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_459_pad_type_0 = const()[name = string("input_459_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_459_strides_0 = const()[name = string("input_459_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_459_pad_0 = const()[name = string("input_459_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_459_dilations_0 = const()[name = string("input_459_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_459_groups_0 = const()[name = string("input_459_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1050224]> module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114461120))), indices_nonzero_data = tensor<uint6, [1050224]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114723328))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115511104))))[name = string("module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> input_457_cast_fp16 = transpose(perm = input_457_perm_0, x = x_193_cast_fp16)[name = string("transpose_251")];
+            tensor<fp16, [1, 2048, 64]> input_459_cast_fp16 = conv(dilations = input_459_dilations_0, groups = input_459_groups_0, pad = input_459_pad_0, pad_type = input_459_pad_type_0, strides = input_459_strides_0, weight = module_layers_8_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_457_cast_fp16)[name = string("input_459_cast_fp16")];
+            int32 x_195_split_num_splits_0 = const()[name = string("x_195_split_num_splits_0"), val = int32(2)];
+            int32 x_195_split_axis_0 = const()[name = string("x_195_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 64]> x_195_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_195_split_cast_fp16_1 = split(axis = x_195_split_axis_0, num_splits = x_195_split_num_splits_0, x = input_459_cast_fp16)[name = string("x_195_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_195_split_1_sigmoid_cast_fp16 = sigmoid(x = x_195_split_cast_fp16_1)[name = string("x_195_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_195_cast_fp16 = mul(x = x_195_split_cast_fp16_0, y = x_195_split_1_sigmoid_cast_fp16)[name = string("x_195_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_461_cast_fp16 = select(a = var_11_to_fp16, b = x_195_cast_fp16, cond = var_328)[name = string("input_461_cast_fp16")];
+            tensor<int32, [6]> input_463_pad_0 = const()[name = string("input_463_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_463_mode_0 = const()[name = string("input_463_mode_0"), val = string("constant")];
+            fp16 const_97_to_fp16 = const()[name = string("const_97_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 72]> input_463_cast_fp16 = pad(constant_val = const_97_to_fp16, mode = input_463_mode_0, pad = input_463_pad_0, x = input_461_cast_fp16)[name = string("input_463_cast_fp16")];
+            string input_465_pad_type_0 = const()[name = string("input_465_pad_type_0"), val = string("valid")];
+            int32 input_465_groups_0 = const()[name = string("input_465_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_465_strides_0 = const()[name = string("input_465_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_465_pad_0 = const()[name = string("input_465_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_465_dilations_0 = const()[name = string("input_465_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_264_to_fp16 = const()[name = string("const_264_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115543936)))];
+            tensor<fp16, [1024]> const_265_to_fp16 = const()[name = string("const_265_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115562432)))];
+            tensor<fp16, [1, 1024, 64]> input_467_cast_fp16 = conv(bias = const_265_to_fp16, dilations = input_465_dilations_0, groups = input_465_groups_0, pad = input_465_pad_0, pad_type = input_465_pad_type_0, strides = input_465_strides_0, weight = const_264_to_fp16, x = input_463_cast_fp16)[name = string("input_467_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_469_cast_fp16 = silu(x = input_467_cast_fp16)[name = string("input_469_cast_fp16")];
+            string x_197_pad_type_0 = const()[name = string("x_197_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_197_strides_0 = const()[name = string("x_197_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_197_pad_0 = const()[name = string("x_197_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_197_dilations_0 = const()[name = string("x_197_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_197_groups_0 = const()[name = string("x_197_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521521]> module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115564544))), indices_nonzero_data = tensor<uint6, [521521]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115695680))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116086912))))[name = string("module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> x_197_cast_fp16 = conv(dilations = x_197_dilations_0, groups = x_197_groups_0, pad = x_197_pad_0, pad_type = x_197_pad_type_0, strides = x_197_strides_0, weight = module_layers_8_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_469_cast_fp16)[name = string("x_197_cast_fp16")];
+            tensor<int32, [3]> input_471_perm_0 = const()[name = string("input_471_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 64, 1024]> input_471_cast_fp16 = transpose(perm = input_471_perm_0, x = x_197_cast_fp16)[name = string("transpose_250")];
+            tensor<fp16, [1, 64, 1024]> input_473_cast_fp16 = add(x = input_455_cast_fp16, y = input_471_cast_fp16)[name = string("input_473_cast_fp16")];
+            tensor<int32, [1]> input_475_axes_0 = const()[name = string("input_475_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_8_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_8_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116103360)))];
+            tensor<fp16, [1024]> module_layers_8_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_8_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116105472)))];
+            tensor<fp16, [1, 64, 1024]> input_475_cast_fp16 = layer_norm(axes = input_475_axes_0, beta = module_layers_8_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_feed_forward2_weight_to_fp16, x = input_473_cast_fp16)[name = string("input_475_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2065024]> module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116107584))), indices_nonzero_data = tensor<uint6, [2065024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116631936))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118180800))))[name = string("module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_80_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_8_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_475_cast_fp16)[name = string("linear_80_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_479_cast_fp16 = silu(x = linear_80_cast_fp16)[name = string("input_479_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2041094]> module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118246400))), indices_nonzero_data = tensor<uint6, [2041094]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118770752))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120301696))))[name = string("module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_81_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_8_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_479_cast_fp16)[name = string("linear_81_cast_fp16")];
+            fp16 var_1670_to_fp16 = const()[name = string("op_1670_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_1671_cast_fp16 = mul(x = linear_81_cast_fp16, y = var_1670_to_fp16)[name = string("op_1671_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_485_cast_fp16 = add(x = input_473_cast_fp16, y = var_1671_cast_fp16)[name = string("input_485_cast_fp16")];
+            tensor<int32, [1]> input_487_axes_0 = const()[name = string("input_487_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_8_norm_out_weight_to_fp16 = const()[name = string("module_layers_8_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120318144)))];
+            tensor<fp16, [1024]> module_layers_8_norm_out_bias_to_fp16 = const()[name = string("module_layers_8_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120320256)))];
+            tensor<fp16, [1, 64, 1024]> input_487_cast_fp16 = layer_norm(axes = input_487_axes_0, beta = module_layers_8_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_out_weight_to_fp16, x = input_485_cast_fp16)[name = string("input_487_cast_fp16")];
+            tensor<int32, [1]> input_489_axes_0 = const()[name = string("input_489_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_9_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_9_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120322368)))];
+            tensor<fp16, [1024]> module_layers_9_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_9_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120324480)))];
+            tensor<fp16, [1, 64, 1024]> input_489_cast_fp16 = layer_norm(axes = input_489_axes_0, beta = module_layers_9_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_feed_forward1_weight_to_fp16, x = input_487_cast_fp16)[name = string("input_489_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2056248]> module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120326592))), indices_nonzero_data = tensor<uint6, [2056248]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120850944))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122393216))))[name = string("module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_82_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_9_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_489_cast_fp16)[name = string("linear_82_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_493_cast_fp16 = silu(x = linear_82_cast_fp16)[name = string("input_493_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1988988]> module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122458816))), indices_nonzero_data = tensor<uint6, [1988988]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122983168))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124475008))))[name = string("module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_83_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_493_cast_fp16)[name = string("linear_83_cast_fp16")];
+            fp16 var_1699_to_fp16 = const()[name = string("op_1699_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_1700_cast_fp16 = mul(x = linear_83_cast_fp16, y = var_1699_to_fp16)[name = string("op_1700_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_499_cast_fp16 = add(x = input_487_cast_fp16, y = var_1700_cast_fp16)[name = string("input_499_cast_fp16")];
+            tensor<int32, [1]> query_19_axes_0 = const()[name = string("query_19_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_9_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_9_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124491456)))];
+            tensor<fp16, [1024]> module_layers_9_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_9_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124493568)))];
+            tensor<fp16, [1, 64, 1024]> query_19_cast_fp16 = layer_norm(axes = query_19_axes_0, beta = module_layers_9_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_self_att_weight_to_fp16, x = input_499_cast_fp16)[name = string("query_19_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524317]> module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124495680))), indices_nonzero_data = tensor<uint6, [524317]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124626816))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125020160))))[name = string("module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_84_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_19_cast_fp16)[name = string("linear_84_cast_fp16")];
+            tensor<int32, [4]> var_1716 = const()[name = string("op_1716"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> q_55_cast_fp16 = reshape(shape = var_1716, x = linear_84_cast_fp16)[name = string("q_55_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524008]> module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125036608))), indices_nonzero_data = tensor<uint6, [524008]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125167744))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125560832))))[name = string("module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_85_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_19_cast_fp16)[name = string("linear_85_cast_fp16")];
+            tensor<int32, [4]> var_1720 = const()[name = string("op_1720"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> k_37_cast_fp16 = reshape(shape = var_1720, x = linear_85_cast_fp16)[name = string("k_37_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524728]> module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125577280))), indices_nonzero_data = tensor<uint6, [524728]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125708416))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126102080))))[name = string("module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_86_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_19_cast_fp16)[name = string("linear_86_cast_fp16")];
+            tensor<int32, [4]> var_1724 = const()[name = string("op_1724"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> v_19_cast_fp16 = reshape(shape = var_1724, x = linear_86_cast_fp16)[name = string("v_19_cast_fp16")];
+            tensor<int32, [4]> value_21_perm_0 = const()[name = string("value_21_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_9_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_9_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126118528)))];
+            tensor<fp16, [1, 64, 8, 128]> var_1736_cast_fp16 = add(x = q_55_cast_fp16, y = module_layers_9_self_attn_pos_bias_u_to_fp16)[name = string("op_1736_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_9_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_9_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126120640)))];
+            tensor<fp16, [1, 64, 8, 128]> var_1738_cast_fp16 = add(x = q_55_cast_fp16, y = module_layers_9_self_attn_pos_bias_v_to_fp16)[name = string("op_1738_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_19_perm_0 = const()[name = string("q_with_bias_v_19_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_205_transpose_x_0 = const()[name = string("x_205_transpose_x_0"), val = bool(false)];
+            bool x_205_transpose_y_0 = const()[name = string("x_205_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 127]> var_1740_to_fp16 = const()[name = string("op_1740_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126122752)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_19_cast_fp16 = transpose(perm = q_with_bias_v_19_perm_0, x = var_1738_cast_fp16)[name = string("transpose_249")];
+            tensor<fp16, [1, 8, 64, 127]> x_205_cast_fp16 = matmul(transpose_x = x_205_transpose_x_0, transpose_y = x_205_transpose_y_0, x = q_with_bias_v_19_cast_fp16, y = var_1740_to_fp16)[name = string("x_205_cast_fp16")];
+            tensor<int32, [8]> x_207_pad_0 = const()[name = string("x_207_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_207_mode_0 = const()[name = string("x_207_mode_0"), val = string("constant")];
+            fp16 const_104_to_fp16 = const()[name = string("const_104_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 64, 128]> x_207_cast_fp16 = pad(constant_val = const_104_to_fp16, mode = x_207_mode_0, pad = x_207_pad_0, x = x_205_cast_fp16)[name = string("x_207_cast_fp16")];
+            tensor<int32, [4]> var_1748 = const()[name = string("op_1748"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_209_cast_fp16 = reshape(shape = var_1748, x = x_207_cast_fp16)[name = string("x_209_cast_fp16")];
+            tensor<int32, [4]> var_1752_begin_0 = const()[name = string("op_1752_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1752_end_0 = const()[name = string("op_1752_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
+            tensor<bool, [4]> var_1752_end_mask_0 = const()[name = string("op_1752_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 127, 64]> var_1752_cast_fp16 = slice_by_index(begin = var_1752_begin_0, end = var_1752_end_0, end_mask = var_1752_end_mask_0, x = x_209_cast_fp16)[name = string("op_1752_cast_fp16")];
+            tensor<int32, [4]> var_1753 = const()[name = string("op_1753"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_37_cast_fp16 = reshape(shape = var_1753, x = var_1752_cast_fp16)[name = string("matrix_bd_37_cast_fp16")];
+            bool matrix_ac_19_transpose_x_0 = const()[name = string("matrix_ac_19_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_19_transpose_y_0 = const()[name = string("matrix_ac_19_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_114_perm_0 = const()[name = string("transpose_114_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_115_perm_0 = const()[name = string("transpose_115_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 64]> transpose_115 = transpose(perm = transpose_115_perm_0, x = k_37_cast_fp16)[name = string("transpose_247")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_114 = transpose(perm = transpose_114_perm_0, x = var_1736_cast_fp16)[name = string("transpose_248")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_19_cast_fp16 = matmul(transpose_x = matrix_ac_19_transpose_x_0, transpose_y = matrix_ac_19_transpose_y_0, x = transpose_114, y = transpose_115)[name = string("matrix_ac_19_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_39_begin_0 = const()[name = string("matrix_bd_39_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_39_end_0 = const()[name = string("matrix_bd_39_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
+            tensor<bool, [4]> matrix_bd_39_end_mask_0 = const()[name = string("matrix_bd_39_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_39_cast_fp16 = slice_by_index(begin = matrix_bd_39_begin_0, end = matrix_bd_39_end_0, end_mask = matrix_bd_39_end_mask_0, x = matrix_bd_37_cast_fp16)[name = string("matrix_bd_39_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_1762_cast_fp16 = add(x = matrix_ac_19_cast_fp16, y = matrix_bd_39_cast_fp16)[name = string("op_1762_cast_fp16")];
+            fp16 _inversed_scores_37_y_0_to_fp16 = const()[name = string("_inversed_scores_37_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_37_cast_fp16 = mul(x = var_1762_cast_fp16, y = _inversed_scores_37_y_0_to_fp16)[name = string("_inversed_scores_37_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_39_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_37_cast_fp16, cond = mask_3)[name = string("scores_39_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_1768_cast_fp16 = softmax(axis = var_30, x = scores_39_cast_fp16)[name = string("op_1768_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_501_cast_fp16 = select(a = var_11_to_fp16, b = var_1768_cast_fp16, cond = mask_3)[name = string("input_501_cast_fp16")];
+            bool x_211_transpose_x_0 = const()[name = string("x_211_transpose_x_0"), val = bool(false)];
+            bool x_211_transpose_y_0 = const()[name = string("x_211_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 64, 128]> value_21_cast_fp16 = transpose(perm = value_21_perm_0, x = v_19_cast_fp16)[name = string("transpose_246")];
+            tensor<fp16, [1, 8, 64, 128]> x_211_cast_fp16 = matmul(transpose_x = x_211_transpose_x_0, transpose_y = x_211_transpose_y_0, x = input_501_cast_fp16, y = value_21_cast_fp16)[name = string("x_211_cast_fp16")];
+            tensor<int32, [4]> var_1772_perm_0 = const()[name = string("op_1772_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1773 = const()[name = string("op_1773"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 64, 8, 128]> var_1772_cast_fp16 = transpose(perm = var_1772_perm_0, x = x_211_cast_fp16)[name = string("transpose_245")];
+            tensor<fp16, [1, 64, 1024]> input_503_cast_fp16 = reshape(shape = var_1773, x = var_1772_cast_fp16)[name = string("input_503_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524470]> module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126382912))), indices_nonzero_data = tensor<uint6, [524470]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126514048))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126907520))))[name = string("module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_88_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_503_cast_fp16)[name = string("linear_88_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_507_cast_fp16 = add(x = input_499_cast_fp16, y = linear_88_cast_fp16)[name = string("input_507_cast_fp16")];
+            tensor<int32, [1]> x_215_axes_0 = const()[name = string("x_215_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_9_norm_conv_weight_to_fp16 = const()[name = string("module_layers_9_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126923968)))];
+            tensor<fp16, [1024]> module_layers_9_norm_conv_bias_to_fp16 = const()[name = string("module_layers_9_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126926080)))];
+            tensor<fp16, [1, 64, 1024]> x_215_cast_fp16 = layer_norm(axes = x_215_axes_0, beta = module_layers_9_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_conv_weight_to_fp16, x = input_507_cast_fp16)[name = string("x_215_cast_fp16")];
+            tensor<int32, [3]> input_509_perm_0 = const()[name = string("input_509_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_511_pad_type_0 = const()[name = string("input_511_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_511_strides_0 = const()[name = string("input_511_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_511_pad_0 = const()[name = string("input_511_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_511_dilations_0 = const()[name = string("input_511_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_511_groups_0 = const()[name = string("input_511_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1050959]> module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126928192))), indices_nonzero_data = tensor<uint6, [1050959]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127190400))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127978688))))[name = string("module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> input_509_cast_fp16 = transpose(perm = input_509_perm_0, x = x_215_cast_fp16)[name = string("transpose_244")];
+            tensor<fp16, [1, 2048, 64]> input_511_cast_fp16 = conv(dilations = input_511_dilations_0, groups = input_511_groups_0, pad = input_511_pad_0, pad_type = input_511_pad_type_0, strides = input_511_strides_0, weight = module_layers_9_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_509_cast_fp16)[name = string("input_511_cast_fp16")];
+            int32 x_217_split_num_splits_0 = const()[name = string("x_217_split_num_splits_0"), val = int32(2)];
+            int32 x_217_split_axis_0 = const()[name = string("x_217_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 64]> x_217_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_217_split_cast_fp16_1 = split(axis = x_217_split_axis_0, num_splits = x_217_split_num_splits_0, x = input_511_cast_fp16)[name = string("x_217_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_217_split_1_sigmoid_cast_fp16 = sigmoid(x = x_217_split_cast_fp16_1)[name = string("x_217_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_217_cast_fp16 = mul(x = x_217_split_cast_fp16_0, y = x_217_split_1_sigmoid_cast_fp16)[name = string("x_217_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_513_cast_fp16 = select(a = var_11_to_fp16, b = x_217_cast_fp16, cond = var_328)[name = string("input_513_cast_fp16")];
+            tensor<int32, [6]> input_515_pad_0 = const()[name = string("input_515_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_515_mode_0 = const()[name = string("input_515_mode_0"), val = string("constant")];
+            fp16 const_107_to_fp16 = const()[name = string("const_107_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 72]> input_515_cast_fp16 = pad(constant_val = const_107_to_fp16, mode = input_515_mode_0, pad = input_515_pad_0, x = input_513_cast_fp16)[name = string("input_515_cast_fp16")];
+            string input_517_pad_type_0 = const()[name = string("input_517_pad_type_0"), val = string("valid")];
+            int32 input_517_groups_0 = const()[name = string("input_517_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_517_strides_0 = const()[name = string("input_517_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_517_pad_0 = const()[name = string("input_517_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_517_dilations_0 = const()[name = string("input_517_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_266_to_fp16 = const()[name = string("const_266_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128011520)))];
+            tensor<fp16, [1024]> const_267_to_fp16 = const()[name = string("const_267_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128030016)))];
+            tensor<fp16, [1, 1024, 64]> input_519_cast_fp16 = conv(bias = const_267_to_fp16, dilations = input_517_dilations_0, groups = input_517_groups_0, pad = input_517_pad_0, pad_type = input_517_pad_type_0, strides = input_517_strides_0, weight = const_266_to_fp16, x = input_515_cast_fp16)[name = string("input_519_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_521_cast_fp16 = silu(x = input_519_cast_fp16)[name = string("input_521_cast_fp16")];
+            string x_219_pad_type_0 = const()[name = string("x_219_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_219_strides_0 = const()[name = string("x_219_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_219_pad_0 = const()[name = string("x_219_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_219_dilations_0 = const()[name = string("x_219_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_219_groups_0 = const()[name = string("x_219_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519248]> module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128032128))), indices_nonzero_data = tensor<uint6, [519248]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128163264))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128552768))))[name = string("module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> x_219_cast_fp16 = conv(dilations = x_219_dilations_0, groups = x_219_groups_0, pad = x_219_pad_0, pad_type = x_219_pad_type_0, strides = x_219_strides_0, weight = module_layers_9_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_521_cast_fp16)[name = string("x_219_cast_fp16")];
+            tensor<int32, [3]> input_523_perm_0 = const()[name = string("input_523_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 64, 1024]> input_523_cast_fp16 = transpose(perm = input_523_perm_0, x = x_219_cast_fp16)[name = string("transpose_243")];
+            tensor<fp16, [1, 64, 1024]> input_525_cast_fp16 = add(x = input_507_cast_fp16, y = input_523_cast_fp16)[name = string("input_525_cast_fp16")];
+            tensor<int32, [1]> input_527_axes_0 = const()[name = string("input_527_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_9_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_9_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128569216)))];
+            tensor<fp16, [1024]> module_layers_9_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_9_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128571328)))];
+            tensor<fp16, [1, 64, 1024]> input_527_cast_fp16 = layer_norm(axes = input_527_axes_0, beta = module_layers_9_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_feed_forward2_weight_to_fp16, x = input_525_cast_fp16)[name = string("input_527_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2059311]> module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128573440))), indices_nonzero_data = tensor<uint6, [2059311]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129097792))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130642368))))[name = string("module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_89_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_9_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_527_cast_fp16)[name = string("linear_89_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_531_cast_fp16 = silu(x = linear_89_cast_fp16)[name = string("input_531_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2055301]> module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130707968))), indices_nonzero_data = tensor<uint6, [2055301]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131232320))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132773888))))[name = string("module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_90_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_9_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_531_cast_fp16)[name = string("linear_90_cast_fp16")];
+            fp16 var_1833_to_fp16 = const()[name = string("op_1833_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_1834_cast_fp16 = mul(x = linear_90_cast_fp16, y = var_1833_to_fp16)[name = string("op_1834_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_537_cast_fp16 = add(x = input_525_cast_fp16, y = var_1834_cast_fp16)[name = string("input_537_cast_fp16")];
+            tensor<int32, [1]> input_539_axes_0 = const()[name = string("input_539_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_9_norm_out_weight_to_fp16 = const()[name = string("module_layers_9_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132790336)))];
+            tensor<fp16, [1024]> module_layers_9_norm_out_bias_to_fp16 = const()[name = string("module_layers_9_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132792448)))];
+            tensor<fp16, [1, 64, 1024]> input_539_cast_fp16 = layer_norm(axes = input_539_axes_0, beta = module_layers_9_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_out_weight_to_fp16, x = input_537_cast_fp16)[name = string("input_539_cast_fp16")];
+            tensor<int32, [1]> input_541_axes_0 = const()[name = string("input_541_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_10_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_10_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132794560)))];
+            tensor<fp16, [1024]> module_layers_10_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_10_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132796672)))];
+            tensor<fp16, [1, 64, 1024]> input_541_cast_fp16 = layer_norm(axes = input_541_axes_0, beta = module_layers_10_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_feed_forward1_weight_to_fp16, x = input_539_cast_fp16)[name = string("input_541_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2042962]> module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132798784))), indices_nonzero_data = tensor<uint6, [2042962]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133323136))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134855424))))[name = string("module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_91_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_10_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_541_cast_fp16)[name = string("linear_91_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_545_cast_fp16 = silu(x = linear_91_cast_fp16)[name = string("input_545_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1990388]> module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134921024))), indices_nonzero_data = tensor<uint6, [1990388]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135445376))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136938240))))[name = string("module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_92_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_545_cast_fp16)[name = string("linear_92_cast_fp16")];
+            fp16 var_1862_to_fp16 = const()[name = string("op_1862_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_1863_cast_fp16 = mul(x = linear_92_cast_fp16, y = var_1862_to_fp16)[name = string("op_1863_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_551_cast_fp16 = add(x = input_539_cast_fp16, y = var_1863_cast_fp16)[name = string("input_551_cast_fp16")];
+            tensor<int32, [1]> query_21_axes_0 = const()[name = string("query_21_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_10_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_10_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136954688)))];
+            tensor<fp16, [1024]> module_layers_10_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_10_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136956800)))];
+            tensor<fp16, [1, 64, 1024]> query_21_cast_fp16 = layer_norm(axes = query_21_axes_0, beta = module_layers_10_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_self_att_weight_to_fp16, x = input_551_cast_fp16)[name = string("query_21_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525519]> module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136958912))), indices_nonzero_data = tensor<uint6, [525519]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137090048))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137484288))))[name = string("module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_93_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_21_cast_fp16)[name = string("linear_93_cast_fp16")];
+            tensor<int32, [4]> var_1879 = const()[name = string("op_1879"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> q_61_cast_fp16 = reshape(shape = var_1879, x = linear_93_cast_fp16)[name = string("q_61_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523272]> module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137500736))), indices_nonzero_data = tensor<uint6, [523272]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137631872))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138024448))))[name = string("module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_94_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_21_cast_fp16)[name = string("linear_94_cast_fp16")];
+            tensor<int32, [4]> var_1883 = const()[name = string("op_1883"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> k_41_cast_fp16 = reshape(shape = var_1883, x = linear_94_cast_fp16)[name = string("k_41_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523359]> module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138040896))), indices_nonzero_data = tensor<uint6, [523359]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138172032))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138564672))))[name = string("module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_95_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_21_cast_fp16)[name = string("linear_95_cast_fp16")];
+            tensor<int32, [4]> var_1887 = const()[name = string("op_1887"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> v_21_cast_fp16 = reshape(shape = var_1887, x = linear_95_cast_fp16)[name = string("v_21_cast_fp16")];
+            tensor<int32, [4]> value_23_perm_0 = const()[name = string("value_23_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_10_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_10_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138581120)))];
+            tensor<fp16, [1, 64, 8, 128]> var_1899_cast_fp16 = add(x = q_61_cast_fp16, y = module_layers_10_self_attn_pos_bias_u_to_fp16)[name = string("op_1899_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_10_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_10_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138583232)))];
+            tensor<fp16, [1, 64, 8, 128]> var_1901_cast_fp16 = add(x = q_61_cast_fp16, y = module_layers_10_self_attn_pos_bias_v_to_fp16)[name = string("op_1901_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_21_perm_0 = const()[name = string("q_with_bias_v_21_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_227_transpose_x_0 = const()[name = string("x_227_transpose_x_0"), val = bool(false)];
+            bool x_227_transpose_y_0 = const()[name = string("x_227_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 127]> var_1903_to_fp16 = const()[name = string("op_1903_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138585344)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_21_cast_fp16 = transpose(perm = q_with_bias_v_21_perm_0, x = var_1901_cast_fp16)[name = string("transpose_242")];
+            tensor<fp16, [1, 8, 64, 127]> x_227_cast_fp16 = matmul(transpose_x = x_227_transpose_x_0, transpose_y = x_227_transpose_y_0, x = q_with_bias_v_21_cast_fp16, y = var_1903_to_fp16)[name = string("x_227_cast_fp16")];
+            tensor<int32, [8]> x_229_pad_0 = const()[name = string("x_229_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_229_mode_0 = const()[name = string("x_229_mode_0"), val = string("constant")];
+            fp16 const_114_to_fp16 = const()[name = string("const_114_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 64, 128]> x_229_cast_fp16 = pad(constant_val = const_114_to_fp16, mode = x_229_mode_0, pad = x_229_pad_0, x = x_227_cast_fp16)[name = string("x_229_cast_fp16")];
+            tensor<int32, [4]> var_1911 = const()[name = string("op_1911"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_231_cast_fp16 = reshape(shape = var_1911, x = x_229_cast_fp16)[name = string("x_231_cast_fp16")];
+            tensor<int32, [4]> var_1915_begin_0 = const()[name = string("op_1915_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1915_end_0 = const()[name = string("op_1915_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
+            tensor<bool, [4]> var_1915_end_mask_0 = const()[name = string("op_1915_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 127, 64]> var_1915_cast_fp16 = slice_by_index(begin = var_1915_begin_0, end = var_1915_end_0, end_mask = var_1915_end_mask_0, x = x_231_cast_fp16)[name = string("op_1915_cast_fp16")];
+            tensor<int32, [4]> var_1916 = const()[name = string("op_1916"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_41_cast_fp16 = reshape(shape = var_1916, x = var_1915_cast_fp16)[name = string("matrix_bd_41_cast_fp16")];
+            bool matrix_ac_21_transpose_x_0 = const()[name = string("matrix_ac_21_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_21_transpose_y_0 = const()[name = string("matrix_ac_21_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_116_perm_0 = const()[name = string("transpose_116_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_117_perm_0 = const()[name = string("transpose_117_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 64]> transpose_117 = transpose(perm = transpose_117_perm_0, x = k_41_cast_fp16)[name = string("transpose_240")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_116 = transpose(perm = transpose_116_perm_0, x = var_1899_cast_fp16)[name = string("transpose_241")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_21_cast_fp16 = matmul(transpose_x = matrix_ac_21_transpose_x_0, transpose_y = matrix_ac_21_transpose_y_0, x = transpose_116, y = transpose_117)[name = string("matrix_ac_21_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_43_begin_0 = const()[name = string("matrix_bd_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_43_end_0 = const()[name = string("matrix_bd_43_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
+            tensor<bool, [4]> matrix_bd_43_end_mask_0 = const()[name = string("matrix_bd_43_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_43_cast_fp16 = slice_by_index(begin = matrix_bd_43_begin_0, end = matrix_bd_43_end_0, end_mask = matrix_bd_43_end_mask_0, x = matrix_bd_41_cast_fp16)[name = string("matrix_bd_43_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_1925_cast_fp16 = add(x = matrix_ac_21_cast_fp16, y = matrix_bd_43_cast_fp16)[name = string("op_1925_cast_fp16")];
+            fp16 _inversed_scores_41_y_0_to_fp16 = const()[name = string("_inversed_scores_41_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_41_cast_fp16 = mul(x = var_1925_cast_fp16, y = _inversed_scores_41_y_0_to_fp16)[name = string("_inversed_scores_41_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_43_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_41_cast_fp16, cond = mask_3)[name = string("scores_43_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_1931_cast_fp16 = softmax(axis = var_30, x = scores_43_cast_fp16)[name = string("op_1931_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_553_cast_fp16 = select(a = var_11_to_fp16, b = var_1931_cast_fp16, cond = mask_3)[name = string("input_553_cast_fp16")];
+            bool x_233_transpose_x_0 = const()[name = string("x_233_transpose_x_0"), val = bool(false)];
+            bool x_233_transpose_y_0 = const()[name = string("x_233_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 64, 128]> value_23_cast_fp16 = transpose(perm = value_23_perm_0, x = v_21_cast_fp16)[name = string("transpose_239")];
+            tensor<fp16, [1, 8, 64, 128]> x_233_cast_fp16 = matmul(transpose_x = x_233_transpose_x_0, transpose_y = x_233_transpose_y_0, x = input_553_cast_fp16, y = value_23_cast_fp16)[name = string("x_233_cast_fp16")];
+            tensor<int32, [4]> var_1935_perm_0 = const()[name = string("op_1935_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1936 = const()[name = string("op_1936"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 64, 8, 128]> var_1935_cast_fp16 = transpose(perm = var_1935_perm_0, x = x_233_cast_fp16)[name = string("transpose_238")];
+            tensor<fp16, [1, 64, 1024]> input_555_cast_fp16 = reshape(shape = var_1936, x = var_1935_cast_fp16)[name = string("input_555_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524561]> module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138845504))), indices_nonzero_data = tensor<uint6, [524561]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138976640))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139370176))))[name = string("module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_97_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_555_cast_fp16)[name = string("linear_97_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_559_cast_fp16 = add(x = input_551_cast_fp16, y = linear_97_cast_fp16)[name = string("input_559_cast_fp16")];
+            tensor<int32, [1]> x_237_axes_0 = const()[name = string("x_237_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_10_norm_conv_weight_to_fp16 = const()[name = string("module_layers_10_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139386624)))];
+            tensor<fp16, [1024]> module_layers_10_norm_conv_bias_to_fp16 = const()[name = string("module_layers_10_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139388736)))];
+            tensor<fp16, [1, 64, 1024]> x_237_cast_fp16 = layer_norm(axes = x_237_axes_0, beta = module_layers_10_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_conv_weight_to_fp16, x = input_559_cast_fp16)[name = string("x_237_cast_fp16")];
+            tensor<int32, [3]> input_561_perm_0 = const()[name = string("input_561_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_563_pad_type_0 = const()[name = string("input_563_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_563_strides_0 = const()[name = string("input_563_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_563_pad_0 = const()[name = string("input_563_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_563_dilations_0 = const()[name = string("input_563_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_563_groups_0 = const()[name = string("input_563_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1050237]> module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139390848))), indices_nonzero_data = tensor<uint6, [1050237]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139653056))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140440832))))[name = string("module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> input_561_cast_fp16 = transpose(perm = input_561_perm_0, x = x_237_cast_fp16)[name = string("transpose_237")];
+            tensor<fp16, [1, 2048, 64]> input_563_cast_fp16 = conv(dilations = input_563_dilations_0, groups = input_563_groups_0, pad = input_563_pad_0, pad_type = input_563_pad_type_0, strides = input_563_strides_0, weight = module_layers_10_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_561_cast_fp16)[name = string("input_563_cast_fp16")];
+            int32 x_239_split_num_splits_0 = const()[name = string("x_239_split_num_splits_0"), val = int32(2)];
+            int32 x_239_split_axis_0 = const()[name = string("x_239_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 64]> x_239_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_239_split_cast_fp16_1 = split(axis = x_239_split_axis_0, num_splits = x_239_split_num_splits_0, x = input_563_cast_fp16)[name = string("x_239_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_239_split_1_sigmoid_cast_fp16 = sigmoid(x = x_239_split_cast_fp16_1)[name = string("x_239_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_239_cast_fp16 = mul(x = x_239_split_cast_fp16_0, y = x_239_split_1_sigmoid_cast_fp16)[name = string("x_239_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_565_cast_fp16 = select(a = var_11_to_fp16, b = x_239_cast_fp16, cond = var_328)[name = string("input_565_cast_fp16")];
+            tensor<int32, [6]> input_567_pad_0 = const()[name = string("input_567_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_567_mode_0 = const()[name = string("input_567_mode_0"), val = string("constant")];
+            fp16 const_117_to_fp16 = const()[name = string("const_117_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 72]> input_567_cast_fp16 = pad(constant_val = const_117_to_fp16, mode = input_567_mode_0, pad = input_567_pad_0, x = input_565_cast_fp16)[name = string("input_567_cast_fp16")];
+            string input_569_pad_type_0 = const()[name = string("input_569_pad_type_0"), val = string("valid")];
+            int32 input_569_groups_0 = const()[name = string("input_569_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_569_strides_0 = const()[name = string("input_569_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_569_pad_0 = const()[name = string("input_569_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_569_dilations_0 = const()[name = string("input_569_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_268_to_fp16 = const()[name = string("const_268_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140473664)))];
+            tensor<fp16, [1024]> const_269_to_fp16 = const()[name = string("const_269_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140492160)))];
+            tensor<fp16, [1, 1024, 64]> input_571_cast_fp16 = conv(bias = const_269_to_fp16, dilations = input_569_dilations_0, groups = input_569_groups_0, pad = input_569_pad_0, pad_type = input_569_pad_type_0, strides = input_569_strides_0, weight = const_268_to_fp16, x = input_567_cast_fp16)[name = string("input_571_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_573_cast_fp16 = silu(x = input_571_cast_fp16)[name = string("input_573_cast_fp16")];
+            string x_241_pad_type_0 = const()[name = string("x_241_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_241_strides_0 = const()[name = string("x_241_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_241_pad_0 = const()[name = string("x_241_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_241_dilations_0 = const()[name = string("x_241_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_241_groups_0 = const()[name = string("x_241_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520978]> module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140494272))), indices_nonzero_data = tensor<uint6, [520978]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140625408))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141016256))))[name = string("module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> x_241_cast_fp16 = conv(dilations = x_241_dilations_0, groups = x_241_groups_0, pad = x_241_pad_0, pad_type = x_241_pad_type_0, strides = x_241_strides_0, weight = module_layers_10_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_573_cast_fp16)[name = string("x_241_cast_fp16")];
+            tensor<int32, [3]> input_575_perm_0 = const()[name = string("input_575_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 64, 1024]> input_575_cast_fp16 = transpose(perm = input_575_perm_0, x = x_241_cast_fp16)[name = string("transpose_236")];
+            tensor<fp16, [1, 64, 1024]> input_577_cast_fp16 = add(x = input_559_cast_fp16, y = input_575_cast_fp16)[name = string("input_577_cast_fp16")];
+            tensor<int32, [1]> input_579_axes_0 = const()[name = string("input_579_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_10_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_10_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141032704)))];
+            tensor<fp16, [1024]> module_layers_10_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_10_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141034816)))];
+            tensor<fp16, [1, 64, 1024]> input_579_cast_fp16 = layer_norm(axes = input_579_axes_0, beta = module_layers_10_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_feed_forward2_weight_to_fp16, x = input_577_cast_fp16)[name = string("input_579_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2057233]> module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141036928))), indices_nonzero_data = tensor<uint6, [2057233]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141561280))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143104320))))[name = string("module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_98_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_10_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_579_cast_fp16)[name = string("linear_98_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_583_cast_fp16 = silu(x = linear_98_cast_fp16)[name = string("input_583_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2031712]> module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143169920))), indices_nonzero_data = tensor<uint6, [2031712]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143694272))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145218176))))[name = string("module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_99_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_10_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_583_cast_fp16)[name = string("linear_99_cast_fp16")];
+            fp16 var_1996_to_fp16 = const()[name = string("op_1996_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_1997_cast_fp16 = mul(x = linear_99_cast_fp16, y = var_1996_to_fp16)[name = string("op_1997_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_589_cast_fp16 = add(x = input_577_cast_fp16, y = var_1997_cast_fp16)[name = string("input_589_cast_fp16")];
+            tensor<int32, [1]> input_591_axes_0 = const()[name = string("input_591_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_10_norm_out_weight_to_fp16 = const()[name = string("module_layers_10_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145234624)))];
+            tensor<fp16, [1024]> module_layers_10_norm_out_bias_to_fp16 = const()[name = string("module_layers_10_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145236736)))];
+            tensor<fp16, [1, 64, 1024]> input_591_cast_fp16 = layer_norm(axes = input_591_axes_0, beta = module_layers_10_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_out_weight_to_fp16, x = input_589_cast_fp16)[name = string("input_591_cast_fp16")];
+            tensor<int32, [1]> input_593_axes_0 = const()[name = string("input_593_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_11_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_11_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145238848)))];
+            tensor<fp16, [1024]> module_layers_11_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_11_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145240960)))];
+            tensor<fp16, [1, 64, 1024]> input_593_cast_fp16 = layer_norm(axes = input_593_axes_0, beta = module_layers_11_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_feed_forward1_weight_to_fp16, x = input_591_cast_fp16)[name = string("input_593_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2039284]> module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145243072))), indices_nonzero_data = tensor<uint6, [2039284]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145767424))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147296960))))[name = string("module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_100_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_11_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_593_cast_fp16)[name = string("linear_100_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_597_cast_fp16 = silu(x = linear_100_cast_fp16)[name = string("input_597_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2009766]> module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147362560))), indices_nonzero_data = tensor<uint6, [2009766]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147886912))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149394304))))[name = string("module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_101_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_597_cast_fp16)[name = string("linear_101_cast_fp16")];
+            fp16 var_2025_to_fp16 = const()[name = string("op_2025_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_2026_cast_fp16 = mul(x = linear_101_cast_fp16, y = var_2025_to_fp16)[name = string("op_2026_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_603_cast_fp16 = add(x = input_591_cast_fp16, y = var_2026_cast_fp16)[name = string("input_603_cast_fp16")];
+            tensor<int32, [1]> query_23_axes_0 = const()[name = string("query_23_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_11_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_11_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149410752)))];
+            tensor<fp16, [1024]> module_layers_11_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_11_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149412864)))];
+            tensor<fp16, [1, 64, 1024]> query_23_cast_fp16 = layer_norm(axes = query_23_axes_0, beta = module_layers_11_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_self_att_weight_to_fp16, x = input_603_cast_fp16)[name = string("query_23_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525654]> module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149414976))), indices_nonzero_data = tensor<uint6, [525654]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149546112))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149940480))))[name = string("module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_102_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_23_cast_fp16)[name = string("linear_102_cast_fp16")];
+            tensor<int32, [4]> var_2042 = const()[name = string("op_2042"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> q_67_cast_fp16 = reshape(shape = var_2042, x = linear_102_cast_fp16)[name = string("q_67_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522850]> module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149956928))), indices_nonzero_data = tensor<uint6, [522850]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150088064))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150480320))))[name = string("module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_103_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_23_cast_fp16)[name = string("linear_103_cast_fp16")];
+            tensor<int32, [4]> var_2046 = const()[name = string("op_2046"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> k_45_cast_fp16 = reshape(shape = var_2046, x = linear_103_cast_fp16)[name = string("k_45_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522138]> module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150496768))), indices_nonzero_data = tensor<uint6, [522138]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150627904))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151019584))))[name = string("module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_104_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_23_cast_fp16)[name = string("linear_104_cast_fp16")];
+            tensor<int32, [4]> var_2050 = const()[name = string("op_2050"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> v_23_cast_fp16 = reshape(shape = var_2050, x = linear_104_cast_fp16)[name = string("v_23_cast_fp16")];
+            tensor<int32, [4]> value_25_perm_0 = const()[name = string("value_25_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_11_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_11_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151036032)))];
+            tensor<fp16, [1, 64, 8, 128]> var_2062_cast_fp16 = add(x = q_67_cast_fp16, y = module_layers_11_self_attn_pos_bias_u_to_fp16)[name = string("op_2062_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_11_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_11_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151038144)))];
+            tensor<fp16, [1, 64, 8, 128]> var_2064_cast_fp16 = add(x = q_67_cast_fp16, y = module_layers_11_self_attn_pos_bias_v_to_fp16)[name = string("op_2064_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_23_perm_0 = const()[name = string("q_with_bias_v_23_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_249_transpose_x_0 = const()[name = string("x_249_transpose_x_0"), val = bool(false)];
+            bool x_249_transpose_y_0 = const()[name = string("x_249_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 127]> var_2066_to_fp16 = const()[name = string("op_2066_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151040256)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_23_cast_fp16 = transpose(perm = q_with_bias_v_23_perm_0, x = var_2064_cast_fp16)[name = string("transpose_235")];
+            tensor<fp16, [1, 8, 64, 127]> x_249_cast_fp16 = matmul(transpose_x = x_249_transpose_x_0, transpose_y = x_249_transpose_y_0, x = q_with_bias_v_23_cast_fp16, y = var_2066_to_fp16)[name = string("x_249_cast_fp16")];
+            tensor<int32, [8]> x_251_pad_0 = const()[name = string("x_251_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_251_mode_0 = const()[name = string("x_251_mode_0"), val = string("constant")];
+            fp16 const_124_to_fp16 = const()[name = string("const_124_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 64, 128]> x_251_cast_fp16 = pad(constant_val = const_124_to_fp16, mode = x_251_mode_0, pad = x_251_pad_0, x = x_249_cast_fp16)[name = string("x_251_cast_fp16")];
+            tensor<int32, [4]> var_2074 = const()[name = string("op_2074"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_253_cast_fp16 = reshape(shape = var_2074, x = x_251_cast_fp16)[name = string("x_253_cast_fp16")];
+            tensor<int32, [4]> var_2078_begin_0 = const()[name = string("op_2078_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2078_end_0 = const()[name = string("op_2078_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
+            tensor<bool, [4]> var_2078_end_mask_0 = const()[name = string("op_2078_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 127, 64]> var_2078_cast_fp16 = slice_by_index(begin = var_2078_begin_0, end = var_2078_end_0, end_mask = var_2078_end_mask_0, x = x_253_cast_fp16)[name = string("op_2078_cast_fp16")];
+            tensor<int32, [4]> var_2079 = const()[name = string("op_2079"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_45_cast_fp16 = reshape(shape = var_2079, x = var_2078_cast_fp16)[name = string("matrix_bd_45_cast_fp16")];
+            bool matrix_ac_23_transpose_x_0 = const()[name = string("matrix_ac_23_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_23_transpose_y_0 = const()[name = string("matrix_ac_23_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_118_perm_0 = const()[name = string("transpose_118_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_119_perm_0 = const()[name = string("transpose_119_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 64]> transpose_119 = transpose(perm = transpose_119_perm_0, x = k_45_cast_fp16)[name = string("transpose_233")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_118 = transpose(perm = transpose_118_perm_0, x = var_2062_cast_fp16)[name = string("transpose_234")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_23_cast_fp16 = matmul(transpose_x = matrix_ac_23_transpose_x_0, transpose_y = matrix_ac_23_transpose_y_0, x = transpose_118, y = transpose_119)[name = string("matrix_ac_23_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_47_begin_0 = const()[name = string("matrix_bd_47_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_47_end_0 = const()[name = string("matrix_bd_47_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
+            tensor<bool, [4]> matrix_bd_47_end_mask_0 = const()[name = string("matrix_bd_47_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_47_cast_fp16 = slice_by_index(begin = matrix_bd_47_begin_0, end = matrix_bd_47_end_0, end_mask = matrix_bd_47_end_mask_0, x = matrix_bd_45_cast_fp16)[name = string("matrix_bd_47_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_2088_cast_fp16 = add(x = matrix_ac_23_cast_fp16, y = matrix_bd_47_cast_fp16)[name = string("op_2088_cast_fp16")];
+            fp16 _inversed_scores_45_y_0_to_fp16 = const()[name = string("_inversed_scores_45_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_45_cast_fp16 = mul(x = var_2088_cast_fp16, y = _inversed_scores_45_y_0_to_fp16)[name = string("_inversed_scores_45_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_47_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_45_cast_fp16, cond = mask_3)[name = string("scores_47_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_2094_cast_fp16 = softmax(axis = var_30, x = scores_47_cast_fp16)[name = string("op_2094_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_605_cast_fp16 = select(a = var_11_to_fp16, b = var_2094_cast_fp16, cond = mask_3)[name = string("input_605_cast_fp16")];
+            bool x_255_transpose_x_0 = const()[name = string("x_255_transpose_x_0"), val = bool(false)];
+            bool x_255_transpose_y_0 = const()[name = string("x_255_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 64, 128]> value_25_cast_fp16 = transpose(perm = value_25_perm_0, x = v_23_cast_fp16)[name = string("transpose_232")];
+            tensor<fp16, [1, 8, 64, 128]> x_255_cast_fp16 = matmul(transpose_x = x_255_transpose_x_0, transpose_y = x_255_transpose_y_0, x = input_605_cast_fp16, y = value_25_cast_fp16)[name = string("x_255_cast_fp16")];
+            tensor<int32, [4]> var_2098_perm_0 = const()[name = string("op_2098_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2099 = const()[name = string("op_2099"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 64, 8, 128]> var_2098_cast_fp16 = transpose(perm = var_2098_perm_0, x = x_255_cast_fp16)[name = string("transpose_231")];
+            tensor<fp16, [1, 64, 1024]> input_607_cast_fp16 = reshape(shape = var_2099, x = var_2098_cast_fp16)[name = string("input_607_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524206]> module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151300416))), indices_nonzero_data = tensor<uint6, [524206]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151431552))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151824832))))[name = string("module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_106_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_607_cast_fp16)[name = string("linear_106_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_611_cast_fp16 = add(x = input_603_cast_fp16, y = linear_106_cast_fp16)[name = string("input_611_cast_fp16")];
+            tensor<int32, [1]> x_259_axes_0 = const()[name = string("x_259_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_11_norm_conv_weight_to_fp16 = const()[name = string("module_layers_11_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151841280)))];
+            tensor<fp16, [1024]> module_layers_11_norm_conv_bias_to_fp16 = const()[name = string("module_layers_11_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151843392)))];
+            tensor<fp16, [1, 64, 1024]> x_259_cast_fp16 = layer_norm(axes = x_259_axes_0, beta = module_layers_11_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_conv_weight_to_fp16, x = input_611_cast_fp16)[name = string("x_259_cast_fp16")];
+            tensor<int32, [3]> input_613_perm_0 = const()[name = string("input_613_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_615_pad_type_0 = const()[name = string("input_615_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_615_strides_0 = const()[name = string("input_615_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_615_pad_0 = const()[name = string("input_615_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_615_dilations_0 = const()[name = string("input_615_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_615_groups_0 = const()[name = string("input_615_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1049241]> module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151845504))), indices_nonzero_data = tensor<uint6, [1049241]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152107712))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152894720))))[name = string("module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> input_613_cast_fp16 = transpose(perm = input_613_perm_0, x = x_259_cast_fp16)[name = string("transpose_230")];
+            tensor<fp16, [1, 2048, 64]> input_615_cast_fp16 = conv(dilations = input_615_dilations_0, groups = input_615_groups_0, pad = input_615_pad_0, pad_type = input_615_pad_type_0, strides = input_615_strides_0, weight = module_layers_11_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_613_cast_fp16)[name = string("input_615_cast_fp16")];
+            int32 x_261_split_num_splits_0 = const()[name = string("x_261_split_num_splits_0"), val = int32(2)];
+            int32 x_261_split_axis_0 = const()[name = string("x_261_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 64]> x_261_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_261_split_cast_fp16_1 = split(axis = x_261_split_axis_0, num_splits = x_261_split_num_splits_0, x = input_615_cast_fp16)[name = string("x_261_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_261_split_1_sigmoid_cast_fp16 = sigmoid(x = x_261_split_cast_fp16_1)[name = string("x_261_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_261_cast_fp16 = mul(x = x_261_split_cast_fp16_0, y = x_261_split_1_sigmoid_cast_fp16)[name = string("x_261_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_617_cast_fp16 = select(a = var_11_to_fp16, b = x_261_cast_fp16, cond = var_328)[name = string("input_617_cast_fp16")];
+            tensor<int32, [6]> input_619_pad_0 = const()[name = string("input_619_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_619_mode_0 = const()[name = string("input_619_mode_0"), val = string("constant")];
+            fp16 const_127_to_fp16 = const()[name = string("const_127_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 72]> input_619_cast_fp16 = pad(constant_val = const_127_to_fp16, mode = input_619_mode_0, pad = input_619_pad_0, x = input_617_cast_fp16)[name = string("input_619_cast_fp16")];
+            string input_621_pad_type_0 = const()[name = string("input_621_pad_type_0"), val = string("valid")];
+            int32 input_621_groups_0 = const()[name = string("input_621_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_621_strides_0 = const()[name = string("input_621_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_621_pad_0 = const()[name = string("input_621_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_621_dilations_0 = const()[name = string("input_621_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_270_to_fp16 = const()[name = string("const_270_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152927552)))];
+            tensor<fp16, [1024]> const_271_to_fp16 = const()[name = string("const_271_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152946048)))];
+            tensor<fp16, [1, 1024, 64]> input_623_cast_fp16 = conv(bias = const_271_to_fp16, dilations = input_621_dilations_0, groups = input_621_groups_0, pad = input_621_pad_0, pad_type = input_621_pad_type_0, strides = input_621_strides_0, weight = const_270_to_fp16, x = input_619_cast_fp16)[name = string("input_623_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_625_cast_fp16 = silu(x = input_623_cast_fp16)[name = string("input_625_cast_fp16")];
+            string x_263_pad_type_0 = const()[name = string("x_263_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_263_strides_0 = const()[name = string("x_263_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_263_pad_0 = const()[name = string("x_263_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_263_dilations_0 = const()[name = string("x_263_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_263_groups_0 = const()[name = string("x_263_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522671]> module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152948160))), indices_nonzero_data = tensor<uint6, [522671]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153079296))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153471424))))[name = string("module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> x_263_cast_fp16 = conv(dilations = x_263_dilations_0, groups = x_263_groups_0, pad = x_263_pad_0, pad_type = x_263_pad_type_0, strides = x_263_strides_0, weight = module_layers_11_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_625_cast_fp16)[name = string("x_263_cast_fp16")];
+            tensor<int32, [3]> input_627_perm_0 = const()[name = string("input_627_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 64, 1024]> input_627_cast_fp16 = transpose(perm = input_627_perm_0, x = x_263_cast_fp16)[name = string("transpose_229")];
+            tensor<fp16, [1, 64, 1024]> input_629_cast_fp16 = add(x = input_611_cast_fp16, y = input_627_cast_fp16)[name = string("input_629_cast_fp16")];
+            tensor<int32, [1]> input_631_axes_0 = const()[name = string("input_631_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_11_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_11_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153487872)))];
+            tensor<fp16, [1024]> module_layers_11_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_11_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153489984)))];
+            tensor<fp16, [1, 64, 1024]> input_631_cast_fp16 = layer_norm(axes = input_631_axes_0, beta = module_layers_11_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_feed_forward2_weight_to_fp16, x = input_629_cast_fp16)[name = string("input_631_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2065473]> module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153492096))), indices_nonzero_data = tensor<uint6, [2065473]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154016448))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155565632))))[name = string("module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_107_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_11_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_631_cast_fp16)[name = string("linear_107_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_635_cast_fp16 = silu(x = linear_107_cast_fp16)[name = string("input_635_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1993236]> module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155631232))), indices_nonzero_data = tensor<uint6, [1993236]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156155584))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157650624))))[name = string("module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_108_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_11_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_635_cast_fp16)[name = string("linear_108_cast_fp16")];
+            fp16 var_2159_to_fp16 = const()[name = string("op_2159_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_2160_cast_fp16 = mul(x = linear_108_cast_fp16, y = var_2159_to_fp16)[name = string("op_2160_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_641_cast_fp16 = add(x = input_629_cast_fp16, y = var_2160_cast_fp16)[name = string("input_641_cast_fp16")];
+            tensor<int32, [1]> input_643_axes_0 = const()[name = string("input_643_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_11_norm_out_weight_to_fp16 = const()[name = string("module_layers_11_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157667072)))];
+            tensor<fp16, [1024]> module_layers_11_norm_out_bias_to_fp16 = const()[name = string("module_layers_11_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157669184)))];
+            tensor<fp16, [1, 64, 1024]> input_643_cast_fp16 = layer_norm(axes = input_643_axes_0, beta = module_layers_11_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_out_weight_to_fp16, x = input_641_cast_fp16)[name = string("input_643_cast_fp16")];
+            tensor<int32, [1]> input_645_axes_0 = const()[name = string("input_645_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_12_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_12_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157671296)))];
+            tensor<fp16, [1024]> module_layers_12_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_12_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157673408)))];
+            tensor<fp16, [1, 64, 1024]> input_645_cast_fp16 = layer_norm(axes = input_645_axes_0, beta = module_layers_12_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_feed_forward1_weight_to_fp16, x = input_643_cast_fp16)[name = string("input_645_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2047312]> module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157675520))), indices_nonzero_data = tensor<uint6, [2047312]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158199872))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159735424))))[name = string("module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_109_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_12_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_645_cast_fp16)[name = string("linear_109_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_649_cast_fp16 = silu(x = linear_109_cast_fp16)[name = string("input_649_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1984634]> module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159801024))), indices_nonzero_data = tensor<uint6, [1984634]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160325376))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161813952))))[name = string("module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_110_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_649_cast_fp16)[name = string("linear_110_cast_fp16")];
+            fp16 var_2188_to_fp16 = const()[name = string("op_2188_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_2189_cast_fp16 = mul(x = linear_110_cast_fp16, y = var_2188_to_fp16)[name = string("op_2189_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_655_cast_fp16 = add(x = input_643_cast_fp16, y = var_2189_cast_fp16)[name = string("input_655_cast_fp16")];
+            tensor<int32, [1]> query_25_axes_0 = const()[name = string("query_25_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_12_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_12_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161830400)))];
+            tensor<fp16, [1024]> module_layers_12_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_12_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161832512)))];
+            tensor<fp16, [1, 64, 1024]> query_25_cast_fp16 = layer_norm(axes = query_25_axes_0, beta = module_layers_12_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_self_att_weight_to_fp16, x = input_655_cast_fp16)[name = string("query_25_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525969]> module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161834624))), indices_nonzero_data = tensor<uint6, [525969]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161965760))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162360320))))[name = string("module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_111_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_25_cast_fp16)[name = string("linear_111_cast_fp16")];
+            tensor<int32, [4]> var_2205 = const()[name = string("op_2205"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> q_73_cast_fp16 = reshape(shape = var_2205, x = linear_111_cast_fp16)[name = string("q_73_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522730]> module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162376768))), indices_nonzero_data = tensor<uint6, [522730]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162507904))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162900032))))[name = string("module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_112_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_25_cast_fp16)[name = string("linear_112_cast_fp16")];
+            tensor<int32, [4]> var_2209 = const()[name = string("op_2209"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> k_49_cast_fp16 = reshape(shape = var_2209, x = linear_112_cast_fp16)[name = string("k_49_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521113]> module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162916480))), indices_nonzero_data = tensor<uint6, [521113]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163047616))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163438528))))[name = string("module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_113_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_25_cast_fp16)[name = string("linear_113_cast_fp16")];
+            tensor<int32, [4]> var_2213 = const()[name = string("op_2213"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> v_25_cast_fp16 = reshape(shape = var_2213, x = linear_113_cast_fp16)[name = string("v_25_cast_fp16")];
+            tensor<int32, [4]> value_27_perm_0 = const()[name = string("value_27_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_12_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_12_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163454976)))];
+            tensor<fp16, [1, 64, 8, 128]> var_2225_cast_fp16 = add(x = q_73_cast_fp16, y = module_layers_12_self_attn_pos_bias_u_to_fp16)[name = string("op_2225_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_12_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_12_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163457088)))];
+            tensor<fp16, [1, 64, 8, 128]> var_2227_cast_fp16 = add(x = q_73_cast_fp16, y = module_layers_12_self_attn_pos_bias_v_to_fp16)[name = string("op_2227_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_25_perm_0 = const()[name = string("q_with_bias_v_25_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_271_transpose_x_0 = const()[name = string("x_271_transpose_x_0"), val = bool(false)];
+            bool x_271_transpose_y_0 = const()[name = string("x_271_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 127]> var_2229_to_fp16 = const()[name = string("op_2229_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163459200)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_25_cast_fp16 = transpose(perm = q_with_bias_v_25_perm_0, x = var_2227_cast_fp16)[name = string("transpose_228")];
+            tensor<fp16, [1, 8, 64, 127]> x_271_cast_fp16 = matmul(transpose_x = x_271_transpose_x_0, transpose_y = x_271_transpose_y_0, x = q_with_bias_v_25_cast_fp16, y = var_2229_to_fp16)[name = string("x_271_cast_fp16")];
+            tensor<int32, [8]> x_273_pad_0 = const()[name = string("x_273_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_273_mode_0 = const()[name = string("x_273_mode_0"), val = string("constant")];
+            fp16 const_134_to_fp16 = const()[name = string("const_134_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 64, 128]> x_273_cast_fp16 = pad(constant_val = const_134_to_fp16, mode = x_273_mode_0, pad = x_273_pad_0, x = x_271_cast_fp16)[name = string("x_273_cast_fp16")];
+            tensor<int32, [4]> var_2237 = const()[name = string("op_2237"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_275_cast_fp16 = reshape(shape = var_2237, x = x_273_cast_fp16)[name = string("x_275_cast_fp16")];
+            tensor<int32, [4]> var_2241_begin_0 = const()[name = string("op_2241_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2241_end_0 = const()[name = string("op_2241_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
+            tensor<bool, [4]> var_2241_end_mask_0 = const()[name = string("op_2241_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 127, 64]> var_2241_cast_fp16 = slice_by_index(begin = var_2241_begin_0, end = var_2241_end_0, end_mask = var_2241_end_mask_0, x = x_275_cast_fp16)[name = string("op_2241_cast_fp16")];
+            tensor<int32, [4]> var_2242 = const()[name = string("op_2242"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_49_cast_fp16 = reshape(shape = var_2242, x = var_2241_cast_fp16)[name = string("matrix_bd_49_cast_fp16")];
+            bool matrix_ac_25_transpose_x_0 = const()[name = string("matrix_ac_25_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_25_transpose_y_0 = const()[name = string("matrix_ac_25_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_120_perm_0 = const()[name = string("transpose_120_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_121_perm_0 = const()[name = string("transpose_121_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 64]> transpose_121 = transpose(perm = transpose_121_perm_0, x = k_49_cast_fp16)[name = string("transpose_226")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_120 = transpose(perm = transpose_120_perm_0, x = var_2225_cast_fp16)[name = string("transpose_227")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_25_cast_fp16 = matmul(transpose_x = matrix_ac_25_transpose_x_0, transpose_y = matrix_ac_25_transpose_y_0, x = transpose_120, y = transpose_121)[name = string("matrix_ac_25_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_51_begin_0 = const()[name = string("matrix_bd_51_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_51_end_0 = const()[name = string("matrix_bd_51_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
+            tensor<bool, [4]> matrix_bd_51_end_mask_0 = const()[name = string("matrix_bd_51_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_51_cast_fp16 = slice_by_index(begin = matrix_bd_51_begin_0, end = matrix_bd_51_end_0, end_mask = matrix_bd_51_end_mask_0, x = matrix_bd_49_cast_fp16)[name = string("matrix_bd_51_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_2251_cast_fp16 = add(x = matrix_ac_25_cast_fp16, y = matrix_bd_51_cast_fp16)[name = string("op_2251_cast_fp16")];
+            fp16 _inversed_scores_49_y_0_to_fp16 = const()[name = string("_inversed_scores_49_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_49_cast_fp16 = mul(x = var_2251_cast_fp16, y = _inversed_scores_49_y_0_to_fp16)[name = string("_inversed_scores_49_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_51_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_49_cast_fp16, cond = mask_3)[name = string("scores_51_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_2257_cast_fp16 = softmax(axis = var_30, x = scores_51_cast_fp16)[name = string("op_2257_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_657_cast_fp16 = select(a = var_11_to_fp16, b = var_2257_cast_fp16, cond = mask_3)[name = string("input_657_cast_fp16")];
+            bool x_277_transpose_x_0 = const()[name = string("x_277_transpose_x_0"), val = bool(false)];
+            bool x_277_transpose_y_0 = const()[name = string("x_277_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 64, 128]> value_27_cast_fp16 = transpose(perm = value_27_perm_0, x = v_25_cast_fp16)[name = string("transpose_225")];
+            tensor<fp16, [1, 8, 64, 128]> x_277_cast_fp16 = matmul(transpose_x = x_277_transpose_x_0, transpose_y = x_277_transpose_y_0, x = input_657_cast_fp16, y = value_27_cast_fp16)[name = string("x_277_cast_fp16")];
+            tensor<int32, [4]> var_2261_perm_0 = const()[name = string("op_2261_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2262 = const()[name = string("op_2262"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 64, 8, 128]> var_2261_cast_fp16 = transpose(perm = var_2261_perm_0, x = x_277_cast_fp16)[name = string("transpose_224")];
+            tensor<fp16, [1, 64, 1024]> input_659_cast_fp16 = reshape(shape = var_2262, x = var_2261_cast_fp16)[name = string("input_659_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [527263]> module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163719360))), indices_nonzero_data = tensor<uint6, [527263]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163850496))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164246016))))[name = string("module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_115_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_659_cast_fp16)[name = string("linear_115_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_663_cast_fp16 = add(x = input_655_cast_fp16, y = linear_115_cast_fp16)[name = string("input_663_cast_fp16")];
+            tensor<int32, [1]> x_281_axes_0 = const()[name = string("x_281_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_12_norm_conv_weight_to_fp16 = const()[name = string("module_layers_12_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164262464)))];
+            tensor<fp16, [1024]> module_layers_12_norm_conv_bias_to_fp16 = const()[name = string("module_layers_12_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164264576)))];
+            tensor<fp16, [1, 64, 1024]> x_281_cast_fp16 = layer_norm(axes = x_281_axes_0, beta = module_layers_12_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_conv_weight_to_fp16, x = input_663_cast_fp16)[name = string("x_281_cast_fp16")];
+            tensor<int32, [3]> input_665_perm_0 = const()[name = string("input_665_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_667_pad_type_0 = const()[name = string("input_667_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_667_strides_0 = const()[name = string("input_667_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_667_pad_0 = const()[name = string("input_667_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_667_dilations_0 = const()[name = string("input_667_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_667_groups_0 = const()[name = string("input_667_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1044574]> module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164266688))), indices_nonzero_data = tensor<uint6, [1044574]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164528896))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165312448))))[name = string("module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> input_665_cast_fp16 = transpose(perm = input_665_perm_0, x = x_281_cast_fp16)[name = string("transpose_223")];
+            tensor<fp16, [1, 2048, 64]> input_667_cast_fp16 = conv(dilations = input_667_dilations_0, groups = input_667_groups_0, pad = input_667_pad_0, pad_type = input_667_pad_type_0, strides = input_667_strides_0, weight = module_layers_12_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_665_cast_fp16)[name = string("input_667_cast_fp16")];
+            int32 x_283_split_num_splits_0 = const()[name = string("x_283_split_num_splits_0"), val = int32(2)];
+            int32 x_283_split_axis_0 = const()[name = string("x_283_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 64]> x_283_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_283_split_cast_fp16_1 = split(axis = x_283_split_axis_0, num_splits = x_283_split_num_splits_0, x = input_667_cast_fp16)[name = string("x_283_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_283_split_1_sigmoid_cast_fp16 = sigmoid(x = x_283_split_cast_fp16_1)[name = string("x_283_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_283_cast_fp16 = mul(x = x_283_split_cast_fp16_0, y = x_283_split_1_sigmoid_cast_fp16)[name = string("x_283_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_669_cast_fp16 = select(a = var_11_to_fp16, b = x_283_cast_fp16, cond = var_328)[name = string("input_669_cast_fp16")];
+            tensor<int32, [6]> input_671_pad_0 = const()[name = string("input_671_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_671_mode_0 = const()[name = string("input_671_mode_0"), val = string("constant")];
+            fp16 const_137_to_fp16 = const()[name = string("const_137_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 72]> input_671_cast_fp16 = pad(constant_val = const_137_to_fp16, mode = input_671_mode_0, pad = input_671_pad_0, x = input_669_cast_fp16)[name = string("input_671_cast_fp16")];
+            string input_673_pad_type_0 = const()[name = string("input_673_pad_type_0"), val = string("valid")];
+            int32 input_673_groups_0 = const()[name = string("input_673_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_673_strides_0 = const()[name = string("input_673_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_673_pad_0 = const()[name = string("input_673_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_673_dilations_0 = const()[name = string("input_673_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_272_to_fp16 = const()[name = string("const_272_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165345280)))];
+            tensor<fp16, [1024]> const_273_to_fp16 = const()[name = string("const_273_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165363776)))];
+            tensor<fp16, [1, 1024, 64]> input_675_cast_fp16 = conv(bias = const_273_to_fp16, dilations = input_673_dilations_0, groups = input_673_groups_0, pad = input_673_pad_0, pad_type = input_673_pad_type_0, strides = input_673_strides_0, weight = const_272_to_fp16, x = input_671_cast_fp16)[name = string("input_675_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_677_cast_fp16 = silu(x = input_675_cast_fp16)[name = string("input_677_cast_fp16")];
+            string x_285_pad_type_0 = const()[name = string("x_285_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_285_strides_0 = const()[name = string("x_285_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_285_pad_0 = const()[name = string("x_285_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_285_dilations_0 = const()[name = string("x_285_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_285_groups_0 = const()[name = string("x_285_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522908]> module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165365888))), indices_nonzero_data = tensor<uint6, [522908]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165497024))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165889280))))[name = string("module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> x_285_cast_fp16 = conv(dilations = x_285_dilations_0, groups = x_285_groups_0, pad = x_285_pad_0, pad_type = x_285_pad_type_0, strides = x_285_strides_0, weight = module_layers_12_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_677_cast_fp16)[name = string("x_285_cast_fp16")];
+            tensor<int32, [3]> input_679_perm_0 = const()[name = string("input_679_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 64, 1024]> input_679_cast_fp16 = transpose(perm = input_679_perm_0, x = x_285_cast_fp16)[name = string("transpose_222")];
+            tensor<fp16, [1, 64, 1024]> input_681_cast_fp16 = add(x = input_663_cast_fp16, y = input_679_cast_fp16)[name = string("input_681_cast_fp16")];
+            tensor<int32, [1]> input_683_axes_0 = const()[name = string("input_683_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_12_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_12_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165905728)))];
+            tensor<fp16, [1024]> module_layers_12_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_12_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165907840)))];
+            tensor<fp16, [1, 64, 1024]> input_683_cast_fp16 = layer_norm(axes = input_683_axes_0, beta = module_layers_12_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_feed_forward2_weight_to_fp16, x = input_681_cast_fp16)[name = string("input_683_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2072965]> module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165909952))), indices_nonzero_data = tensor<uint6, [2072965]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166434304))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167989120))))[name = string("module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_116_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_12_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_683_cast_fp16)[name = string("linear_116_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_687_cast_fp16 = silu(x = linear_116_cast_fp16)[name = string("input_687_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1985707]> module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168054720))), indices_nonzero_data = tensor<uint6, [1985707]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168579072))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170068480))))[name = string("module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_117_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_12_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_687_cast_fp16)[name = string("linear_117_cast_fp16")];
+            fp16 var_2322_to_fp16 = const()[name = string("op_2322_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_2323_cast_fp16 = mul(x = linear_117_cast_fp16, y = var_2322_to_fp16)[name = string("op_2323_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_693_cast_fp16 = add(x = input_681_cast_fp16, y = var_2323_cast_fp16)[name = string("input_693_cast_fp16")];
+            tensor<int32, [1]> input_695_axes_0 = const()[name = string("input_695_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_12_norm_out_weight_to_fp16 = const()[name = string("module_layers_12_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170084928)))];
+            tensor<fp16, [1024]> module_layers_12_norm_out_bias_to_fp16 = const()[name = string("module_layers_12_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170087040)))];
+            tensor<fp16, [1, 64, 1024]> input_695_cast_fp16 = layer_norm(axes = input_695_axes_0, beta = module_layers_12_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_out_weight_to_fp16, x = input_693_cast_fp16)[name = string("input_695_cast_fp16")];
+            tensor<int32, [1]> input_697_axes_0 = const()[name = string("input_697_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_13_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_13_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170089152)))];
+            tensor<fp16, [1024]> module_layers_13_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_13_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170091264)))];
+            tensor<fp16, [1, 64, 1024]> input_697_cast_fp16 = layer_norm(axes = input_697_axes_0, beta = module_layers_13_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_feed_forward1_weight_to_fp16, x = input_695_cast_fp16)[name = string("input_697_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2052653]> module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170093376))), indices_nonzero_data = tensor<uint6, [2052653]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170617728))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172157312))))[name = string("module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_118_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_13_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_697_cast_fp16)[name = string("linear_118_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_701_cast_fp16 = silu(x = linear_118_cast_fp16)[name = string("input_701_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1946459]> module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172222912))), indices_nonzero_data = tensor<uint6, [1946459]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172747264))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174207232))))[name = string("module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_119_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_701_cast_fp16)[name = string("linear_119_cast_fp16")];
+            fp16 var_2351_to_fp16 = const()[name = string("op_2351_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_2352_cast_fp16 = mul(x = linear_119_cast_fp16, y = var_2351_to_fp16)[name = string("op_2352_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_707_cast_fp16 = add(x = input_695_cast_fp16, y = var_2352_cast_fp16)[name = string("input_707_cast_fp16")];
+            tensor<int32, [1]> query_27_axes_0 = const()[name = string("query_27_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_13_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_13_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174223680)))];
+            tensor<fp16, [1024]> module_layers_13_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_13_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174225792)))];
+            tensor<fp16, [1, 64, 1024]> query_27_cast_fp16 = layer_norm(axes = query_27_axes_0, beta = module_layers_13_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_self_att_weight_to_fp16, x = input_707_cast_fp16)[name = string("query_27_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524122]> module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174227904))), indices_nonzero_data = tensor<uint6, [524122]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174359040))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174752256))))[name = string("module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_120_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_27_cast_fp16)[name = string("linear_120_cast_fp16")];
+            tensor<int32, [4]> var_2368 = const()[name = string("op_2368"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> q_79_cast_fp16 = reshape(shape = var_2368, x = linear_120_cast_fp16)[name = string("q_79_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521808]> module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174768704))), indices_nonzero_data = tensor<uint6, [521808]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174899840))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175291264))))[name = string("module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_121_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_27_cast_fp16)[name = string("linear_121_cast_fp16")];
+            tensor<int32, [4]> var_2372 = const()[name = string("op_2372"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> k_53_cast_fp16 = reshape(shape = var_2372, x = linear_121_cast_fp16)[name = string("k_53_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521835]> module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175307712))), indices_nonzero_data = tensor<uint6, [521835]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175438848))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175830336))))[name = string("module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_122_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_27_cast_fp16)[name = string("linear_122_cast_fp16")];
+            tensor<int32, [4]> var_2376 = const()[name = string("op_2376"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> v_27_cast_fp16 = reshape(shape = var_2376, x = linear_122_cast_fp16)[name = string("v_27_cast_fp16")];
+            tensor<int32, [4]> value_29_perm_0 = const()[name = string("value_29_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_13_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_13_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175846784)))];
+            tensor<fp16, [1, 64, 8, 128]> var_2388_cast_fp16 = add(x = q_79_cast_fp16, y = module_layers_13_self_attn_pos_bias_u_to_fp16)[name = string("op_2388_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_13_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_13_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175848896)))];
+            tensor<fp16, [1, 64, 8, 128]> var_2390_cast_fp16 = add(x = q_79_cast_fp16, y = module_layers_13_self_attn_pos_bias_v_to_fp16)[name = string("op_2390_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_27_perm_0 = const()[name = string("q_with_bias_v_27_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_293_transpose_x_0 = const()[name = string("x_293_transpose_x_0"), val = bool(false)];
+            bool x_293_transpose_y_0 = const()[name = string("x_293_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 127]> var_2392_to_fp16 = const()[name = string("op_2392_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175851008)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_27_cast_fp16 = transpose(perm = q_with_bias_v_27_perm_0, x = var_2390_cast_fp16)[name = string("transpose_221")];
+            tensor<fp16, [1, 8, 64, 127]> x_293_cast_fp16 = matmul(transpose_x = x_293_transpose_x_0, transpose_y = x_293_transpose_y_0, x = q_with_bias_v_27_cast_fp16, y = var_2392_to_fp16)[name = string("x_293_cast_fp16")];
+            tensor<int32, [8]> x_295_pad_0 = const()[name = string("x_295_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_295_mode_0 = const()[name = string("x_295_mode_0"), val = string("constant")];
+            fp16 const_144_to_fp16 = const()[name = string("const_144_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 64, 128]> x_295_cast_fp16 = pad(constant_val = const_144_to_fp16, mode = x_295_mode_0, pad = x_295_pad_0, x = x_293_cast_fp16)[name = string("x_295_cast_fp16")];
+            tensor<int32, [4]> var_2400 = const()[name = string("op_2400"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_297_cast_fp16 = reshape(shape = var_2400, x = x_295_cast_fp16)[name = string("x_297_cast_fp16")];
+            tensor<int32, [4]> var_2404_begin_0 = const()[name = string("op_2404_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2404_end_0 = const()[name = string("op_2404_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
+            tensor<bool, [4]> var_2404_end_mask_0 = const()[name = string("op_2404_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 127, 64]> var_2404_cast_fp16 = slice_by_index(begin = var_2404_begin_0, end = var_2404_end_0, end_mask = var_2404_end_mask_0, x = x_297_cast_fp16)[name = string("op_2404_cast_fp16")];
+            tensor<int32, [4]> var_2405 = const()[name = string("op_2405"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_53_cast_fp16 = reshape(shape = var_2405, x = var_2404_cast_fp16)[name = string("matrix_bd_53_cast_fp16")];
+            bool matrix_ac_27_transpose_x_0 = const()[name = string("matrix_ac_27_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_27_transpose_y_0 = const()[name = string("matrix_ac_27_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_122_perm_0 = const()[name = string("transpose_122_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_123_perm_0 = const()[name = string("transpose_123_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 64]> transpose_123 = transpose(perm = transpose_123_perm_0, x = k_53_cast_fp16)[name = string("transpose_219")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_122 = transpose(perm = transpose_122_perm_0, x = var_2388_cast_fp16)[name = string("transpose_220")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_27_cast_fp16 = matmul(transpose_x = matrix_ac_27_transpose_x_0, transpose_y = matrix_ac_27_transpose_y_0, x = transpose_122, y = transpose_123)[name = string("matrix_ac_27_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_55_begin_0 = const()[name = string("matrix_bd_55_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_55_end_0 = const()[name = string("matrix_bd_55_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
+            tensor<bool, [4]> matrix_bd_55_end_mask_0 = const()[name = string("matrix_bd_55_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_55_cast_fp16 = slice_by_index(begin = matrix_bd_55_begin_0, end = matrix_bd_55_end_0, end_mask = matrix_bd_55_end_mask_0, x = matrix_bd_53_cast_fp16)[name = string("matrix_bd_55_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_2414_cast_fp16 = add(x = matrix_ac_27_cast_fp16, y = matrix_bd_55_cast_fp16)[name = string("op_2414_cast_fp16")];
+            fp16 _inversed_scores_53_y_0_to_fp16 = const()[name = string("_inversed_scores_53_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_53_cast_fp16 = mul(x = var_2414_cast_fp16, y = _inversed_scores_53_y_0_to_fp16)[name = string("_inversed_scores_53_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_55_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_53_cast_fp16, cond = mask_3)[name = string("scores_55_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_2420_cast_fp16 = softmax(axis = var_30, x = scores_55_cast_fp16)[name = string("op_2420_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_709_cast_fp16 = select(a = var_11_to_fp16, b = var_2420_cast_fp16, cond = mask_3)[name = string("input_709_cast_fp16")];
+            bool x_299_transpose_x_0 = const()[name = string("x_299_transpose_x_0"), val = bool(false)];
+            bool x_299_transpose_y_0 = const()[name = string("x_299_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 64, 128]> value_29_cast_fp16 = transpose(perm = value_29_perm_0, x = v_27_cast_fp16)[name = string("transpose_218")];
+            tensor<fp16, [1, 8, 64, 128]> x_299_cast_fp16 = matmul(transpose_x = x_299_transpose_x_0, transpose_y = x_299_transpose_y_0, x = input_709_cast_fp16, y = value_29_cast_fp16)[name = string("x_299_cast_fp16")];
+            tensor<int32, [4]> var_2424_perm_0 = const()[name = string("op_2424_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2425 = const()[name = string("op_2425"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 64, 8, 128]> var_2424_cast_fp16 = transpose(perm = var_2424_perm_0, x = x_299_cast_fp16)[name = string("transpose_217")];
+            tensor<fp16, [1, 64, 1024]> input_711_cast_fp16 = reshape(shape = var_2425, x = var_2424_cast_fp16)[name = string("input_711_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [528285]> module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176111168))), indices_nonzero_data = tensor<uint6, [528285]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176242304))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176638592))))[name = string("module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_124_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_711_cast_fp16)[name = string("linear_124_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_715_cast_fp16 = add(x = input_707_cast_fp16, y = linear_124_cast_fp16)[name = string("input_715_cast_fp16")];
+            tensor<int32, [1]> x_303_axes_0 = const()[name = string("x_303_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_13_norm_conv_weight_to_fp16 = const()[name = string("module_layers_13_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176655040)))];
+            tensor<fp16, [1024]> module_layers_13_norm_conv_bias_to_fp16 = const()[name = string("module_layers_13_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176657152)))];
+            tensor<fp16, [1, 64, 1024]> x_303_cast_fp16 = layer_norm(axes = x_303_axes_0, beta = module_layers_13_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_conv_weight_to_fp16, x = input_715_cast_fp16)[name = string("x_303_cast_fp16")];
+            tensor<int32, [3]> input_717_perm_0 = const()[name = string("input_717_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_719_pad_type_0 = const()[name = string("input_719_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_719_strides_0 = const()[name = string("input_719_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_719_pad_0 = const()[name = string("input_719_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_719_dilations_0 = const()[name = string("input_719_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_719_groups_0 = const()[name = string("input_719_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1045290]> module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176659264))), indices_nonzero_data = tensor<uint6, [1045290]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176921472))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177705536))))[name = string("module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> input_717_cast_fp16 = transpose(perm = input_717_perm_0, x = x_303_cast_fp16)[name = string("transpose_216")];
+            tensor<fp16, [1, 2048, 64]> input_719_cast_fp16 = conv(dilations = input_719_dilations_0, groups = input_719_groups_0, pad = input_719_pad_0, pad_type = input_719_pad_type_0, strides = input_719_strides_0, weight = module_layers_13_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_717_cast_fp16)[name = string("input_719_cast_fp16")];
+            int32 x_305_split_num_splits_0 = const()[name = string("x_305_split_num_splits_0"), val = int32(2)];
+            int32 x_305_split_axis_0 = const()[name = string("x_305_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 64]> x_305_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_305_split_cast_fp16_1 = split(axis = x_305_split_axis_0, num_splits = x_305_split_num_splits_0, x = input_719_cast_fp16)[name = string("x_305_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_305_split_1_sigmoid_cast_fp16 = sigmoid(x = x_305_split_cast_fp16_1)[name = string("x_305_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_305_cast_fp16 = mul(x = x_305_split_cast_fp16_0, y = x_305_split_1_sigmoid_cast_fp16)[name = string("x_305_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_721_cast_fp16 = select(a = var_11_to_fp16, b = x_305_cast_fp16, cond = var_328)[name = string("input_721_cast_fp16")];
+            tensor<int32, [6]> input_723_pad_0 = const()[name = string("input_723_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_723_mode_0 = const()[name = string("input_723_mode_0"), val = string("constant")];
+            fp16 const_147_to_fp16 = const()[name = string("const_147_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 72]> input_723_cast_fp16 = pad(constant_val = const_147_to_fp16, mode = input_723_mode_0, pad = input_723_pad_0, x = input_721_cast_fp16)[name = string("input_723_cast_fp16")];
+            string input_725_pad_type_0 = const()[name = string("input_725_pad_type_0"), val = string("valid")];
+            int32 input_725_groups_0 = const()[name = string("input_725_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_725_strides_0 = const()[name = string("input_725_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_725_pad_0 = const()[name = string("input_725_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_725_dilations_0 = const()[name = string("input_725_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_274_to_fp16 = const()[name = string("const_274_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177738368)))];
+            tensor<fp16, [1024]> const_275_to_fp16 = const()[name = string("const_275_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177756864)))];
+            tensor<fp16, [1, 1024, 64]> input_727_cast_fp16 = conv(bias = const_275_to_fp16, dilations = input_725_dilations_0, groups = input_725_groups_0, pad = input_725_pad_0, pad_type = input_725_pad_type_0, strides = input_725_strides_0, weight = const_274_to_fp16, x = input_723_cast_fp16)[name = string("input_727_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_729_cast_fp16 = silu(x = input_727_cast_fp16)[name = string("input_729_cast_fp16")];
+            string x_307_pad_type_0 = const()[name = string("x_307_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_307_strides_0 = const()[name = string("x_307_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_307_pad_0 = const()[name = string("x_307_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_307_dilations_0 = const()[name = string("x_307_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_307_groups_0 = const()[name = string("x_307_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524845]> module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177758976))), indices_nonzero_data = tensor<uint6, [524845]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177890112))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178283840))))[name = string("module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> x_307_cast_fp16 = conv(dilations = x_307_dilations_0, groups = x_307_groups_0, pad = x_307_pad_0, pad_type = x_307_pad_type_0, strides = x_307_strides_0, weight = module_layers_13_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_729_cast_fp16)[name = string("x_307_cast_fp16")];
+            tensor<int32, [3]> input_731_perm_0 = const()[name = string("input_731_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 64, 1024]> input_731_cast_fp16 = transpose(perm = input_731_perm_0, x = x_307_cast_fp16)[name = string("transpose_215")];
+            tensor<fp16, [1, 64, 1024]> input_733_cast_fp16 = add(x = input_715_cast_fp16, y = input_731_cast_fp16)[name = string("input_733_cast_fp16")];
+            tensor<int32, [1]> input_735_axes_0 = const()[name = string("input_735_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_13_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_13_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178300288)))];
+            tensor<fp16, [1024]> module_layers_13_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_13_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178302400)))];
+            tensor<fp16, [1, 64, 1024]> input_735_cast_fp16 = layer_norm(axes = input_735_axes_0, beta = module_layers_13_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_feed_forward2_weight_to_fp16, x = input_733_cast_fp16)[name = string("input_735_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2074364]> module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178304512))), indices_nonzero_data = tensor<uint6, [2074364]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178828864))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180384704))))[name = string("module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_125_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_13_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_735_cast_fp16)[name = string("linear_125_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_739_cast_fp16 = silu(x = linear_125_cast_fp16)[name = string("input_739_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1986819]> module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180450304))), indices_nonzero_data = tensor<uint6, [1986819]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180974656))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182464896))))[name = string("module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_126_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_13_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_739_cast_fp16)[name = string("linear_126_cast_fp16")];
+            fp16 var_2485_to_fp16 = const()[name = string("op_2485_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_2486_cast_fp16 = mul(x = linear_126_cast_fp16, y = var_2485_to_fp16)[name = string("op_2486_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_745_cast_fp16 = add(x = input_733_cast_fp16, y = var_2486_cast_fp16)[name = string("input_745_cast_fp16")];
+            tensor<int32, [1]> input_747_axes_0 = const()[name = string("input_747_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_13_norm_out_weight_to_fp16 = const()[name = string("module_layers_13_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182481344)))];
+            tensor<fp16, [1024]> module_layers_13_norm_out_bias_to_fp16 = const()[name = string("module_layers_13_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182483456)))];
+            tensor<fp16, [1, 64, 1024]> input_747_cast_fp16 = layer_norm(axes = input_747_axes_0, beta = module_layers_13_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_out_weight_to_fp16, x = input_745_cast_fp16)[name = string("input_747_cast_fp16")];
+            tensor<int32, [1]> input_749_axes_0 = const()[name = string("input_749_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_14_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_14_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182485568)))];
+            tensor<fp16, [1024]> module_layers_14_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_14_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182487680)))];
+            tensor<fp16, [1, 64, 1024]> input_749_cast_fp16 = layer_norm(axes = input_749_axes_0, beta = module_layers_14_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_feed_forward1_weight_to_fp16, x = input_747_cast_fp16)[name = string("input_749_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2059717]> module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182489792))), indices_nonzero_data = tensor<uint6, [2059717]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183014144))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184559040))))[name = string("module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_127_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_14_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_749_cast_fp16)[name = string("linear_127_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_753_cast_fp16 = silu(x = linear_127_cast_fp16)[name = string("input_753_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1924564]> module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184624640))), indices_nonzero_data = tensor<uint6, [1924564]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185148992))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186592512))))[name = string("module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_128_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_753_cast_fp16)[name = string("linear_128_cast_fp16")];
+            fp16 var_2514_to_fp16 = const()[name = string("op_2514_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_2515_cast_fp16 = mul(x = linear_128_cast_fp16, y = var_2514_to_fp16)[name = string("op_2515_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_759_cast_fp16 = add(x = input_747_cast_fp16, y = var_2515_cast_fp16)[name = string("input_759_cast_fp16")];
+            tensor<int32, [1]> query_29_axes_0 = const()[name = string("query_29_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_14_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_14_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186608960)))];
+            tensor<fp16, [1024]> module_layers_14_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_14_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186611072)))];
+            tensor<fp16, [1, 64, 1024]> query_29_cast_fp16 = layer_norm(axes = query_29_axes_0, beta = module_layers_14_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_self_att_weight_to_fp16, x = input_759_cast_fp16)[name = string("query_29_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525274]> module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186613184))), indices_nonzero_data = tensor<uint6, [525274]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186744320))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187138368))))[name = string("module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_129_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_29_cast_fp16)[name = string("linear_129_cast_fp16")];
+            tensor<int32, [4]> var_2531 = const()[name = string("op_2531"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> q_85_cast_fp16 = reshape(shape = var_2531, x = linear_129_cast_fp16)[name = string("q_85_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522998]> module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187154816))), indices_nonzero_data = tensor<uint6, [522998]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187285952))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187678272))))[name = string("module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_130_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_29_cast_fp16)[name = string("linear_130_cast_fp16")];
+            tensor<int32, [4]> var_2535 = const()[name = string("op_2535"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> k_57_cast_fp16 = reshape(shape = var_2535, x = linear_130_cast_fp16)[name = string("k_57_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [522224]> module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187694720))), indices_nonzero_data = tensor<uint6, [522224]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187825856))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188217600))))[name = string("module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_131_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_29_cast_fp16)[name = string("linear_131_cast_fp16")];
+            tensor<int32, [4]> var_2539 = const()[name = string("op_2539"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> v_29_cast_fp16 = reshape(shape = var_2539, x = linear_131_cast_fp16)[name = string("v_29_cast_fp16")];
+            tensor<int32, [4]> value_31_perm_0 = const()[name = string("value_31_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_14_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_14_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188234048)))];
+            tensor<fp16, [1, 64, 8, 128]> var_2551_cast_fp16 = add(x = q_85_cast_fp16, y = module_layers_14_self_attn_pos_bias_u_to_fp16)[name = string("op_2551_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_14_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_14_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188236160)))];
+            tensor<fp16, [1, 64, 8, 128]> var_2553_cast_fp16 = add(x = q_85_cast_fp16, y = module_layers_14_self_attn_pos_bias_v_to_fp16)[name = string("op_2553_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_29_perm_0 = const()[name = string("q_with_bias_v_29_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_315_transpose_x_0 = const()[name = string("x_315_transpose_x_0"), val = bool(false)];
+            bool x_315_transpose_y_0 = const()[name = string("x_315_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 127]> var_2555_to_fp16 = const()[name = string("op_2555_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188238272)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_29_cast_fp16 = transpose(perm = q_with_bias_v_29_perm_0, x = var_2553_cast_fp16)[name = string("transpose_214")];
+            tensor<fp16, [1, 8, 64, 127]> x_315_cast_fp16 = matmul(transpose_x = x_315_transpose_x_0, transpose_y = x_315_transpose_y_0, x = q_with_bias_v_29_cast_fp16, y = var_2555_to_fp16)[name = string("x_315_cast_fp16")];
+            tensor<int32, [8]> x_317_pad_0 = const()[name = string("x_317_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_317_mode_0 = const()[name = string("x_317_mode_0"), val = string("constant")];
+            fp16 const_154_to_fp16 = const()[name = string("const_154_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 64, 128]> x_317_cast_fp16 = pad(constant_val = const_154_to_fp16, mode = x_317_mode_0, pad = x_317_pad_0, x = x_315_cast_fp16)[name = string("x_317_cast_fp16")];
+            tensor<int32, [4]> var_2563 = const()[name = string("op_2563"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_319_cast_fp16 = reshape(shape = var_2563, x = x_317_cast_fp16)[name = string("x_319_cast_fp16")];
+            tensor<int32, [4]> var_2567_begin_0 = const()[name = string("op_2567_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2567_end_0 = const()[name = string("op_2567_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
+            tensor<bool, [4]> var_2567_end_mask_0 = const()[name = string("op_2567_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 127, 64]> var_2567_cast_fp16 = slice_by_index(begin = var_2567_begin_0, end = var_2567_end_0, end_mask = var_2567_end_mask_0, x = x_319_cast_fp16)[name = string("op_2567_cast_fp16")];
+            tensor<int32, [4]> var_2568 = const()[name = string("op_2568"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_57_cast_fp16 = reshape(shape = var_2568, x = var_2567_cast_fp16)[name = string("matrix_bd_57_cast_fp16")];
+            bool matrix_ac_29_transpose_x_0 = const()[name = string("matrix_ac_29_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_29_transpose_y_0 = const()[name = string("matrix_ac_29_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_124_perm_0 = const()[name = string("transpose_124_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_125_perm_0 = const()[name = string("transpose_125_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 64]> transpose_125 = transpose(perm = transpose_125_perm_0, x = k_57_cast_fp16)[name = string("transpose_212")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_124 = transpose(perm = transpose_124_perm_0, x = var_2551_cast_fp16)[name = string("transpose_213")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_29_cast_fp16 = matmul(transpose_x = matrix_ac_29_transpose_x_0, transpose_y = matrix_ac_29_transpose_y_0, x = transpose_124, y = transpose_125)[name = string("matrix_ac_29_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_59_begin_0 = const()[name = string("matrix_bd_59_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_59_end_0 = const()[name = string("matrix_bd_59_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
+            tensor<bool, [4]> matrix_bd_59_end_mask_0 = const()[name = string("matrix_bd_59_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_59_cast_fp16 = slice_by_index(begin = matrix_bd_59_begin_0, end = matrix_bd_59_end_0, end_mask = matrix_bd_59_end_mask_0, x = matrix_bd_57_cast_fp16)[name = string("matrix_bd_59_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_2577_cast_fp16 = add(x = matrix_ac_29_cast_fp16, y = matrix_bd_59_cast_fp16)[name = string("op_2577_cast_fp16")];
+            fp16 _inversed_scores_57_y_0_to_fp16 = const()[name = string("_inversed_scores_57_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_57_cast_fp16 = mul(x = var_2577_cast_fp16, y = _inversed_scores_57_y_0_to_fp16)[name = string("_inversed_scores_57_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_59_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_57_cast_fp16, cond = mask_3)[name = string("scores_59_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_2583_cast_fp16 = softmax(axis = var_30, x = scores_59_cast_fp16)[name = string("op_2583_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_761_cast_fp16 = select(a = var_11_to_fp16, b = var_2583_cast_fp16, cond = mask_3)[name = string("input_761_cast_fp16")];
+            bool x_321_transpose_x_0 = const()[name = string("x_321_transpose_x_0"), val = bool(false)];
+            bool x_321_transpose_y_0 = const()[name = string("x_321_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 64, 128]> value_31_cast_fp16 = transpose(perm = value_31_perm_0, x = v_29_cast_fp16)[name = string("transpose_211")];
+            tensor<fp16, [1, 8, 64, 128]> x_321_cast_fp16 = matmul(transpose_x = x_321_transpose_x_0, transpose_y = x_321_transpose_y_0, x = input_761_cast_fp16, y = value_31_cast_fp16)[name = string("x_321_cast_fp16")];
+            tensor<int32, [4]> var_2587_perm_0 = const()[name = string("op_2587_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2588 = const()[name = string("op_2588"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 64, 8, 128]> var_2587_cast_fp16 = transpose(perm = var_2587_perm_0, x = x_321_cast_fp16)[name = string("transpose_210")];
+            tensor<fp16, [1, 64, 1024]> input_763_cast_fp16 = reshape(shape = var_2588, x = var_2587_cast_fp16)[name = string("input_763_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524764]> module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188498432))), indices_nonzero_data = tensor<uint6, [524764]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188629568))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189023232))))[name = string("module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_133_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_763_cast_fp16)[name = string("linear_133_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_767_cast_fp16 = add(x = input_759_cast_fp16, y = linear_133_cast_fp16)[name = string("input_767_cast_fp16")];
+            tensor<int32, [1]> x_325_axes_0 = const()[name = string("x_325_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_14_norm_conv_weight_to_fp16 = const()[name = string("module_layers_14_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189039680)))];
+            tensor<fp16, [1024]> module_layers_14_norm_conv_bias_to_fp16 = const()[name = string("module_layers_14_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189041792)))];
+            tensor<fp16, [1, 64, 1024]> x_325_cast_fp16 = layer_norm(axes = x_325_axes_0, beta = module_layers_14_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_conv_weight_to_fp16, x = input_767_cast_fp16)[name = string("x_325_cast_fp16")];
+            tensor<int32, [3]> input_769_perm_0 = const()[name = string("input_769_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_771_pad_type_0 = const()[name = string("input_771_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_771_strides_0 = const()[name = string("input_771_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_771_pad_0 = const()[name = string("input_771_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_771_dilations_0 = const()[name = string("input_771_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_771_groups_0 = const()[name = string("input_771_groups_0"), val = int32(1)];
+            tensor<uint1, [2048, 1024, 1]> module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1043300]> module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189043904))), indices_nonzero_data = tensor<uint6, [1043300]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189306112))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190088704))))[name = string("module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [2048, 1024, 1]> module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> input_769_cast_fp16 = transpose(perm = input_769_perm_0, x = x_325_cast_fp16)[name = string("transpose_209")];
+            tensor<fp16, [1, 2048, 64]> input_771_cast_fp16 = conv(dilations = input_771_dilations_0, groups = input_771_groups_0, pad = input_771_pad_0, pad_type = input_771_pad_type_0, strides = input_771_strides_0, weight = module_layers_14_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_769_cast_fp16)[name = string("input_771_cast_fp16")];
+            int32 x_327_split_num_splits_0 = const()[name = string("x_327_split_num_splits_0"), val = int32(2)];
+            int32 x_327_split_axis_0 = const()[name = string("x_327_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 64]> x_327_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_327_split_cast_fp16_1 = split(axis = x_327_split_axis_0, num_splits = x_327_split_num_splits_0, x = input_771_cast_fp16)[name = string("x_327_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_327_split_1_sigmoid_cast_fp16 = sigmoid(x = x_327_split_cast_fp16_1)[name = string("x_327_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_327_cast_fp16 = mul(x = x_327_split_cast_fp16_0, y = x_327_split_1_sigmoid_cast_fp16)[name = string("x_327_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_773_cast_fp16 = select(a = var_11_to_fp16, b = x_327_cast_fp16, cond = var_328)[name = string("input_773_cast_fp16")];
+            tensor<int32, [6]> input_775_pad_0 = const()[name = string("input_775_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_775_mode_0 = const()[name = string("input_775_mode_0"), val = string("constant")];
+            fp16 const_157_to_fp16 = const()[name = string("const_157_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 72]> input_775_cast_fp16 = pad(constant_val = const_157_to_fp16, mode = input_775_mode_0, pad = input_775_pad_0, x = input_773_cast_fp16)[name = string("input_775_cast_fp16")];
+            string input_777_pad_type_0 = const()[name = string("input_777_pad_type_0"), val = string("valid")];
+            int32 input_777_groups_0 = const()[name = string("input_777_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_777_strides_0 = const()[name = string("input_777_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_777_pad_0 = const()[name = string("input_777_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_777_dilations_0 = const()[name = string("input_777_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_276_to_fp16 = const()[name = string("const_276_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190121536)))];
+            tensor<fp16, [1024]> const_277_to_fp16 = const()[name = string("const_277_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190140032)))];
+            tensor<fp16, [1, 1024, 64]> input_779_cast_fp16 = conv(bias = const_277_to_fp16, dilations = input_777_dilations_0, groups = input_777_groups_0, pad = input_777_pad_0, pad_type = input_777_pad_type_0, strides = input_777_strides_0, weight = const_276_to_fp16, x = input_775_cast_fp16)[name = string("input_779_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_781_cast_fp16 = silu(x = input_779_cast_fp16)[name = string("input_781_cast_fp16")];
+            string x_329_pad_type_0 = const()[name = string("x_329_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_329_strides_0 = const()[name = string("x_329_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_329_pad_0 = const()[name = string("x_329_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_329_dilations_0 = const()[name = string("x_329_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_329_groups_0 = const()[name = string("x_329_groups_0"), val = int32(1)];
+            tensor<uint1, [1024, 1024, 1]> module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525590]> module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190142144))), indices_nonzero_data = tensor<uint6, [525590]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190273280))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190667584))))[name = string("module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024, 1]> module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 1024, 64]> x_329_cast_fp16 = conv(dilations = x_329_dilations_0, groups = x_329_groups_0, pad = x_329_pad_0, pad_type = x_329_pad_type_0, strides = x_329_strides_0, weight = module_layers_14_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_781_cast_fp16)[name = string("x_329_cast_fp16")];
+            tensor<int32, [3]> input_783_perm_0 = const()[name = string("input_783_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 64, 1024]> input_783_cast_fp16 = transpose(perm = input_783_perm_0, x = x_329_cast_fp16)[name = string("transpose_208")];
+            tensor<fp16, [1, 64, 1024]> input_785_cast_fp16 = add(x = input_767_cast_fp16, y = input_783_cast_fp16)[name = string("input_785_cast_fp16")];
+            tensor<int32, [1]> input_787_axes_0 = const()[name = string("input_787_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_14_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_14_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190684032)))];
+            tensor<fp16, [1024]> module_layers_14_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_14_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190686144)))];
+            tensor<fp16, [1, 64, 1024]> input_787_cast_fp16 = layer_norm(axes = input_787_axes_0, beta = module_layers_14_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_feed_forward2_weight_to_fp16, x = input_785_cast_fp16)[name = string("input_787_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2073313]> module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190688256))), indices_nonzero_data = tensor<uint6, [2073313]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191212608))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192767680))))[name = string("module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_134_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_14_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_787_cast_fp16)[name = string("linear_134_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_791_cast_fp16 = silu(x = linear_134_cast_fp16)[name = string("input_791_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1989781]> module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192833280))), indices_nonzero_data = tensor<uint6, [1989781]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193357632))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194850048))))[name = string("module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_135_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_14_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_791_cast_fp16)[name = string("linear_135_cast_fp16")];
+            fp16 var_2648_to_fp16 = const()[name = string("op_2648_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_2649_cast_fp16 = mul(x = linear_135_cast_fp16, y = var_2648_to_fp16)[name = string("op_2649_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_797_cast_fp16 = add(x = input_785_cast_fp16, y = var_2649_cast_fp16)[name = string("input_797_cast_fp16")];
+            tensor<int32, [1]> input_799_axes_0 = const()[name = string("input_799_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_14_norm_out_weight_to_fp16 = const()[name = string("module_layers_14_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194866496)))];
+            tensor<fp16, [1024]> module_layers_14_norm_out_bias_to_fp16 = const()[name = string("module_layers_14_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194868608)))];
+            tensor<fp16, [1, 64, 1024]> input_799_cast_fp16 = layer_norm(axes = input_799_axes_0, beta = module_layers_14_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_out_weight_to_fp16, x = input_797_cast_fp16)[name = string("input_799_cast_fp16")];
+            tensor<int32, [1]> input_801_axes_0 = const()[name = string("input_801_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_15_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_15_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194870720)))];
+            tensor<fp16, [1024]> module_layers_15_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_15_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194872832)))];
+            tensor<fp16, [1, 64, 1024]> input_801_cast_fp16 = layer_norm(axes = input_801_axes_0, beta = module_layers_15_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_feed_forward1_weight_to_fp16, x = input_799_cast_fp16)[name = string("input_801_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2064877]> module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194874944))), indices_nonzero_data = tensor<uint6, [2064877]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195399296))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196948032))))[name = string("module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [4096, 1024]> module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 4096]> linear_136_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_15_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_801_cast_fp16)[name = string("linear_136_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_805_cast_fp16 = silu(x = linear_136_cast_fp16)[name = string("input_805_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1919271]> module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197013632))), indices_nonzero_data = tensor<uint6, [1919271]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197537984))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198977536))))[name = string("module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 4096]> module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_137_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_805_cast_fp16)[name = string("linear_137_cast_fp16")];
+            fp16 var_2677_to_fp16 = const()[name = string("op_2677_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 64, 1024]> var_2678_cast_fp16 = mul(x = linear_137_cast_fp16, y = var_2677_to_fp16)[name = string("op_2678_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_811_cast_fp16 = add(x = input_799_cast_fp16, y = var_2678_cast_fp16)[name = string("input_811_cast_fp16")];
+            tensor<int32, [1]> query_31_axes_0 = const()[name = string("query_31_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> module_layers_15_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_15_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198993984)))];
+            tensor<fp16, [1024]> module_layers_15_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_15_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198996096)))];
+            tensor<fp16, [1, 64, 1024]> query_31_cast_fp16 = layer_norm(axes = query_31_axes_0, beta = module_layers_15_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_self_att_weight_to_fp16, x = input_811_cast_fp16)[name = string("query_31_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525795]> module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198998208))), indices_nonzero_data = tensor<uint6, [525795]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199129344))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199523776))))[name = string("module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_138_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_31_cast_fp16)[name = string("linear_138_cast_fp16")];
+            tensor<int32, [4]> var_2694 = const()[name = string("op_2694"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> q_91_cast_fp16 = reshape(shape = var_2694, x = linear_138_cast_fp16)[name = string("q_91_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523094]> module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199540224))), indices_nonzero_data = tensor<uint6, [523094]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199671360))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200063808))))[name = string("module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_139_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_31_cast_fp16)[name = string("linear_139_cast_fp16")];
+            tensor<int32, [4]> var_2698 = const()[name = string("op_2698"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> k_61_cast_fp16 = reshape(shape = var_2698, x = linear_139_cast_fp16)[name = string("k_61_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523685]> module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200080256))), indices_nonzero_data = tensor<uint6, [523685]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200211392))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200604224))))[name = string("module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024, 1024]> module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue")];
+            tensor<fp16, [1, 64, 1024]> linear_140_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_31_cast_fp16)[name = string("linear_140_cast_fp16")];
+            tensor<int32, [4]> var_2702 = const()[name = string("op_2702"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 64, 8, 128]> v_31_cast_fp16 = reshape(shape = var_2702, x = linear_140_cast_fp16)[name = string("v_31_cast_fp16")];
+            tensor<int32, [4]> value_33_perm_0 = const()[name = string("value_33_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> module_layers_15_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_15_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200620672)))];
+            tensor<fp16, [1, 64, 8, 128]> var_2714_cast_fp16 = add(x = q_91_cast_fp16, y = module_layers_15_self_attn_pos_bias_u_to_fp16)[name = string("op_2714_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_15_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_15_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200622784)))];
+            tensor<fp16, [1, 64, 8, 128]> var_2716_cast_fp16 = add(x = q_91_cast_fp16, y = module_layers_15_self_attn_pos_bias_v_to_fp16)[name = string("op_2716_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_31_perm_0 = const()[name = string("q_with_bias_v_31_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_337_transpose_x_0 = const()[name = string("x_337_transpose_x_0"), val = bool(false)];
+            bool x_337_transpose_y_0 = const()[name = string("x_337_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 127]> var_2718_to_fp16 = const()[name = string("op_2718_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200624896)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_31_cast_fp16 = transpose(perm = q_with_bias_v_31_perm_0, x = var_2716_cast_fp16)[name = string("transpose_207")];
+            tensor<fp16, [1, 8, 64, 127]> x_337_cast_fp16 = matmul(transpose_x = x_337_transpose_x_0, transpose_y = x_337_transpose_y_0, x = q_with_bias_v_31_cast_fp16, y = var_2718_to_fp16)[name = string("x_337_cast_fp16")];
+            tensor<int32, [8]> x_339_pad_0 = const()[name = string("x_339_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_339_mode_0 = const()[name = string("x_339_mode_0"), val = string("constant")];
+            fp16 const_164_to_fp16 = const()[name = string("const_164_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 64, 128]> x_339_cast_fp16 = pad(constant_val = const_164_to_fp16, mode = x_339_mode_0, pad = x_339_pad_0, x = x_337_cast_fp16)[name = string("x_339_cast_fp16")];
+            tensor<int32, [4]> var_2726 = const()[name = string("op_2726"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_341_cast_fp16 = reshape(shape = var_2726, x = x_339_cast_fp16)[name = string("x_341_cast_fp16")];
+            tensor<int32, [4]> var_2730_begin_0 = const()[name = string("op_2730_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2730_end_0 = const()[name = string("op_2730_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
+            tensor<bool, [4]> var_2730_end_mask_0 = const()[name = string("op_2730_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 127, 64]> var_2730_cast_fp16 = slice_by_index(begin = var_2730_begin_0, end = var_2730_end_0, end_mask = var_2730_end_mask_0, x = x_341_cast_fp16)[name = string("op_2730_cast_fp16")];
+            tensor<int32, [4]> var_2731 = const()[name = string("op_2731"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_61_cast_fp16 = reshape(shape = var_2731, x = var_2730_cast_fp16)[name = string("matrix_bd_61_cast_fp16")];
+            bool matrix_ac_31_transpose_x_0 = const()[name = string("matrix_ac_31_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_31_transpose_y_0 = const()[name = string("matrix_ac_31_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_126_perm_0 = const()[name = string("transpose_126_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_127_perm_0 = const()[name = string("transpose_127_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 64]> transpose_127 = transpose(perm = transpose_127_perm_0, x = k_61_cast_fp16)[name = string("transpose_205")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_126 = transpose(perm = transpose_126_perm_0, x = var_2714_cast_fp16)[name = string("transpose_206")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_31_cast_fp16 = matmul(transpose_x = matrix_ac_31_transpose_x_0, transpose_y = matrix_ac_31_transpose_y_0, x = transpose_126, y = transpose_127)[name = string("matrix_ac_31_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_63_begin_0 = const()[name = string("matrix_bd_63_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_63_end_0 = const()[name = string("matrix_bd_63_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
+            tensor<bool, [4]> matrix_bd_63_end_mask_0 = const()[name = string("matrix_bd_63_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_63_cast_fp16 = slice_by_index(begin = matrix_bd_63_begin_0, end = matrix_bd_63_end_0, end_mask = matrix_bd_63_end_mask_0, x = matrix_bd_61_cast_fp16)[name = string("matrix_bd_63_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_2740_cast_fp16 = add(x = matrix_ac_31_cast_fp16, y = matrix_bd_63_cast_fp16)[name = string("op_2740_cast_fp16")];
+            fp16 _inversed_scores_61_y_0_to_fp16 = const()[name = string("_inversed_scores_61_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_61_cast_fp16 = mul(x = var_2740_cast_fp16, y = _inversed_scores_61_y_0_to_fp16)[name = string("_inversed_scores_61_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_63_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_61_cast_fp16, cond = mask_3)[name = string("scores_63_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_2746_cast_fp16 = softmax(axis = var_30, x = scores_63_cast_fp16)[name = string("op_2746_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_813_cast_fp16 = select(a = var_11_to_fp16, b = var_2746_cast_fp16, cond = mask_3)[name = string("input_813_cast_fp16")];
             bool x_343_transpose_x_0 = const()[name = string("x_343_transpose_x_0"), val = bool(false)];
             bool x_343_transpose_y_0 = const()[name = string("x_343_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_33_cast_fp16 = transpose(perm = value_33_perm_0, x = v_31_cast_fp16)[name = string("transpose_204")];
-            tensor<fp16, [1, 8, 188, 128]> x_343_cast_fp16 = matmul(transpose_x = x_343_transpose_x_0, transpose_y = x_343_transpose_y_0, x = input_813_cast_fp16, y = value_33_cast_fp16)[name = string("x_343_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 128]> value_33_cast_fp16 = transpose(perm = value_33_perm_0, x = v_31_cast_fp16)[name = string("transpose_204")];
+            tensor<fp16, [1, 8, 64, 128]> x_343_cast_fp16 = matmul(transpose_x = x_343_transpose_x_0, transpose_y = x_343_transpose_y_0, x = input_813_cast_fp16, y = value_33_cast_fp16)[name = string("x_343_cast_fp16")];
             tensor<int32, [4]> var_2750_perm_0 = const()[name = string("op_2750_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_2751 = const()[name = string("op_2751"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_2750_cast_fp16 = transpose(perm = var_2750_perm_0, x = x_343_cast_fp16)[name = string("transpose_203")];
-            tensor<fp16, [1, 188, 1024]> input_815_cast_fp16 = reshape(shape = var_2751, x = var_2750_cast_fp16)[name = string("input_815_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [526464]> module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209012032))), indices_nonzero_data = tensor<uint6, [526464]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209143168))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209538112))))[name = string("module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> var_2750_cast_fp16 = transpose(perm = var_2750_perm_0, x = x_343_cast_fp16)[name = string("transpose_203")];
+            tensor<fp16, [1, 64, 1024]> input_815_cast_fp16 = reshape(shape = var_2751, x = var_2750_cast_fp16)[name = string("input_815_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [526464]> module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200885056))), indices_nonzero_data = tensor<uint6, [526464]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201016192))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201411136))))[name = string("module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_142_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_815_cast_fp16)[name = string("linear_142_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_819_cast_fp16 = add(x = input_811_cast_fp16, y = linear_142_cast_fp16)[name = string("input_819_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_142_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_815_cast_fp16)[name = string("linear_142_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_819_cast_fp16 = add(x = input_811_cast_fp16, y = linear_142_cast_fp16)[name = string("input_819_cast_fp16")];
             tensor<int32, [1]> x_347_axes_0 = const()[name = string("x_347_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_15_norm_conv_weight_to_fp16 = const()[name = string("module_layers_15_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209554560)))];
-            tensor<fp16, [1024]> module_layers_15_norm_conv_bias_to_fp16 = const()[name = string("module_layers_15_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209556672)))];
-            tensor<fp16, [1, 188, 1024]> x_347_cast_fp16 = layer_norm(axes = x_347_axes_0, beta = module_layers_15_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_conv_weight_to_fp16, x = input_819_cast_fp16)[name = string("x_347_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_15_norm_conv_weight_to_fp16 = const()[name = string("module_layers_15_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201427584)))];
+            tensor<fp16, [1024]> module_layers_15_norm_conv_bias_to_fp16 = const()[name = string("module_layers_15_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201429696)))];
+            tensor<fp16, [1, 64, 1024]> x_347_cast_fp16 = layer_norm(axes = x_347_axes_0, beta = module_layers_15_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_conv_weight_to_fp16, x = input_819_cast_fp16)[name = string("x_347_cast_fp16")];
             tensor<int32, [3]> input_821_perm_0 = const()[name = string("input_821_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_823_pad_type_0 = const()[name = string("input_823_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_823_strides_0 = const()[name = string("input_823_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_823_pad_0 = const()[name = string("input_823_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_823_dilations_0 = const()[name = string("input_823_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_823_groups_0 = const()[name = string("input_823_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1046264]> module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209558784))), indices_nonzero_data = tensor<uint6, [1046264]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209820992))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210605760))))[name = string("module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1046264]> module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201431808))), indices_nonzero_data = tensor<uint6, [1046264]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201694016))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202478784))))[name = string("module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_821_cast_fp16 = transpose(perm = input_821_perm_0, x = x_347_cast_fp16)[name = string("transpose_202")];
-            tensor<fp16, [1, 2048, 188]> input_823_cast_fp16 = conv(dilations = input_823_dilations_0, groups = input_823_groups_0, pad = input_823_pad_0, pad_type = input_823_pad_type_0, strides = input_823_strides_0, weight = module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_821_cast_fp16)[name = string("input_823_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_821_cast_fp16 = transpose(perm = input_821_perm_0, x = x_347_cast_fp16)[name = string("transpose_202")];
+            tensor<fp16, [1, 2048, 64]> input_823_cast_fp16 = conv(dilations = input_823_dilations_0, groups = input_823_groups_0, pad = input_823_pad_0, pad_type = input_823_pad_type_0, strides = input_823_strides_0, weight = module_layers_15_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_821_cast_fp16)[name = string("input_823_cast_fp16")];
             int32 x_349_split_num_splits_0 = const()[name = string("x_349_split_num_splits_0"), val = int32(2)];
             int32 x_349_split_axis_0 = const()[name = string("x_349_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_349_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_349_split_cast_fp16_1 = split(axis = x_349_split_axis_0, num_splits = x_349_split_num_splits_0, x = input_823_cast_fp16)[name = string("x_349_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_349_split_1_sigmoid_cast_fp16 = sigmoid(x = x_349_split_cast_fp16_1)[name = string("x_349_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_349_cast_fp16 = mul(x = x_349_split_cast_fp16_0, y = x_349_split_1_sigmoid_cast_fp16)[name = string("x_349_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_825_cast_fp16 = select(a = var_11_to_fp16, b = x_349_cast_fp16, cond = var_328)[name = string("input_825_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_349_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_349_split_cast_fp16_1 = split(axis = x_349_split_axis_0, num_splits = x_349_split_num_splits_0, x = input_823_cast_fp16)[name = string("x_349_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_349_split_1_sigmoid_cast_fp16 = sigmoid(x = x_349_split_cast_fp16_1)[name = string("x_349_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_349_cast_fp16 = mul(x = x_349_split_cast_fp16_0, y = x_349_split_1_sigmoid_cast_fp16)[name = string("x_349_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_825_cast_fp16 = select(a = var_11_to_fp16, b = x_349_cast_fp16, cond = var_328)[name = string("input_825_cast_fp16")];
             tensor<int32, [6]> input_827_pad_0 = const()[name = string("input_827_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_827_mode_0 = const()[name = string("input_827_mode_0"), val = string("constant")];
             fp16 const_167_to_fp16 = const()[name = string("const_167_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_827_cast_fp16 = pad(constant_val = const_167_to_fp16, mode = input_827_mode_0, pad = input_827_pad_0, x = input_825_cast_fp16)[name = string("input_827_cast_fp16")];
+            tensor<fp16, [1, 1024, 72]> input_827_cast_fp16 = pad(constant_val = const_167_to_fp16, mode = input_827_mode_0, pad = input_827_pad_0, x = input_825_cast_fp16)[name = string("input_827_cast_fp16")];
             string input_829_pad_type_0 = const()[name = string("input_829_pad_type_0"), val = string("valid")];
             int32 input_829_groups_0 = const()[name = string("input_829_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_829_strides_0 = const()[name = string("input_829_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_829_pad_0 = const()[name = string("input_829_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_829_dilations_0 = const()[name = string("input_829_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_278_to_fp16 = const()[name = string("const_278_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210638592)))];
-            tensor<fp16, [1024]> const_279_to_fp16 = const()[name = string("const_279_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210657088)))];
-            tensor<fp16, [1, 1024, 188]> input_831_cast_fp16 = conv(bias = const_279_to_fp16, dilations = input_829_dilations_0, groups = input_829_groups_0, pad = input_829_pad_0, pad_type = input_829_pad_type_0, strides = input_829_strides_0, weight = const_278_to_fp16, x = input_827_cast_fp16)[name = string("input_831_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_833_cast_fp16 = silu(x = input_831_cast_fp16)[name = string("input_833_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_278_to_fp16 = const()[name = string("const_278_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202511616)))];
+            tensor<fp16, [1024]> const_279_to_fp16 = const()[name = string("const_279_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202530112)))];
+            tensor<fp16, [1, 1024, 64]> input_831_cast_fp16 = conv(bias = const_279_to_fp16, dilations = input_829_dilations_0, groups = input_829_groups_0, pad = input_829_pad_0, pad_type = input_829_pad_type_0, strides = input_829_strides_0, weight = const_278_to_fp16, x = input_827_cast_fp16)[name = string("input_831_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_833_cast_fp16 = silu(x = input_831_cast_fp16)[name = string("input_833_cast_fp16")];
             string x_351_pad_type_0 = const()[name = string("x_351_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_351_strides_0 = const()[name = string("x_351_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_351_pad_0 = const()[name = string("x_351_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_351_dilations_0 = const()[name = string("x_351_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_351_groups_0 = const()[name = string("x_351_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524462]> module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210659200))), indices_nonzero_data = tensor<uint6, [524462]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210790336))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211183808))))[name = string("module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524462]> module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202532224))), indices_nonzero_data = tensor<uint6, [524462]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202663360))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203056832))))[name = string("module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_351_cast_fp16 = conv(dilations = x_351_dilations_0, groups = x_351_groups_0, pad = x_351_pad_0, pad_type = x_351_pad_type_0, strides = x_351_strides_0, weight = module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_833_cast_fp16)[name = string("x_351_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_351_cast_fp16 = conv(dilations = x_351_dilations_0, groups = x_351_groups_0, pad = x_351_pad_0, pad_type = x_351_pad_type_0, strides = x_351_strides_0, weight = module_layers_15_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_833_cast_fp16)[name = string("x_351_cast_fp16")];
             tensor<int32, [3]> input_835_perm_0 = const()[name = string("input_835_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_835_cast_fp16 = transpose(perm = input_835_perm_0, x = x_351_cast_fp16)[name = string("transpose_201")];
-            tensor<fp16, [1, 188, 1024]> input_837_cast_fp16 = add(x = input_819_cast_fp16, y = input_835_cast_fp16)[name = string("input_837_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_835_cast_fp16 = transpose(perm = input_835_perm_0, x = x_351_cast_fp16)[name = string("transpose_201")];
+            tensor<fp16, [1, 64, 1024]> input_837_cast_fp16 = add(x = input_819_cast_fp16, y = input_835_cast_fp16)[name = string("input_837_cast_fp16")];
             tensor<int32, [1]> input_839_axes_0 = const()[name = string("input_839_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_15_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_15_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211200256)))];
-            tensor<fp16, [1024]> module_layers_15_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_15_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211202368)))];
-            tensor<fp16, [1, 188, 1024]> input_839_cast_fp16 = layer_norm(axes = input_839_axes_0, beta = module_layers_15_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_feed_forward2_weight_to_fp16, x = input_837_cast_fp16)[name = string("input_839_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2078588]> module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211204480))), indices_nonzero_data = tensor<uint6, [2078588]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211728832))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213287872))))[name = string("module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_15_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_15_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203073280)))];
+            tensor<fp16, [1024]> module_layers_15_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_15_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203075392)))];
+            tensor<fp16, [1, 64, 1024]> input_839_cast_fp16 = layer_norm(axes = input_839_axes_0, beta = module_layers_15_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_feed_forward2_weight_to_fp16, x = input_837_cast_fp16)[name = string("input_839_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2078588]> module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203077504))), indices_nonzero_data = tensor<uint6, [2078588]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203601856))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205160896))))[name = string("module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_143_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_839_cast_fp16)[name = string("linear_143_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_843_cast_fp16 = silu(x = linear_143_cast_fp16)[name = string("input_843_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1967312]> module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213353472))), indices_nonzero_data = tensor<uint6, [1967312]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213877824))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215353408))))[name = string("module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 4096]> linear_143_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_15_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_839_cast_fp16)[name = string("linear_143_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_843_cast_fp16 = silu(x = linear_143_cast_fp16)[name = string("input_843_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1967312]> module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205226496))), indices_nonzero_data = tensor<uint6, [1967312]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205750848))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207226432))))[name = string("module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_144_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_843_cast_fp16)[name = string("linear_144_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_144_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_15_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_843_cast_fp16)[name = string("linear_144_cast_fp16")];
             fp16 var_2811_to_fp16 = const()[name = string("op_2811_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_2812_cast_fp16 = mul(x = linear_144_cast_fp16, y = var_2811_to_fp16)[name = string("op_2812_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_849_cast_fp16 = add(x = input_837_cast_fp16, y = var_2812_cast_fp16)[name = string("input_849_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> var_2812_cast_fp16 = mul(x = linear_144_cast_fp16, y = var_2811_to_fp16)[name = string("op_2812_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_849_cast_fp16 = add(x = input_837_cast_fp16, y = var_2812_cast_fp16)[name = string("input_849_cast_fp16")];
             tensor<int32, [1]> input_851_axes_0 = const()[name = string("input_851_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_15_norm_out_weight_to_fp16 = const()[name = string("module_layers_15_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215369856)))];
-            tensor<fp16, [1024]> module_layers_15_norm_out_bias_to_fp16 = const()[name = string("module_layers_15_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215371968)))];
-            tensor<fp16, [1, 188, 1024]> input_851_cast_fp16 = layer_norm(axes = input_851_axes_0, beta = module_layers_15_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_out_weight_to_fp16, x = input_849_cast_fp16)[name = string("input_851_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_15_norm_out_weight_to_fp16 = const()[name = string("module_layers_15_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207242880)))];
+            tensor<fp16, [1024]> module_layers_15_norm_out_bias_to_fp16 = const()[name = string("module_layers_15_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207244992)))];
+            tensor<fp16, [1, 64, 1024]> input_851_cast_fp16 = layer_norm(axes = input_851_axes_0, beta = module_layers_15_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_out_weight_to_fp16, x = input_849_cast_fp16)[name = string("input_851_cast_fp16")];
             tensor<int32, [1]> input_853_axes_0 = const()[name = string("input_853_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_16_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_16_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215374080)))];
-            tensor<fp16, [1024]> module_layers_16_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_16_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215376192)))];
-            tensor<fp16, [1, 188, 1024]> input_853_cast_fp16 = layer_norm(axes = input_853_axes_0, beta = module_layers_16_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_feed_forward1_weight_to_fp16, x = input_851_cast_fp16)[name = string("input_853_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2069902]> module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215378304))), indices_nonzero_data = tensor<uint6, [2069902]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215902656))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217455168))))[name = string("module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_16_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_16_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207247104)))];
+            tensor<fp16, [1024]> module_layers_16_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_16_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207249216)))];
+            tensor<fp16, [1, 64, 1024]> input_853_cast_fp16 = layer_norm(axes = input_853_axes_0, beta = module_layers_16_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_feed_forward1_weight_to_fp16, x = input_851_cast_fp16)[name = string("input_853_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2069902]> module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207251328))), indices_nonzero_data = tensor<uint6, [2069902]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207775680))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209328192))))[name = string("module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_145_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_853_cast_fp16)[name = string("linear_145_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_857_cast_fp16 = silu(x = linear_145_cast_fp16)[name = string("input_857_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1927421]> module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217520768))), indices_nonzero_data = tensor<uint6, [1927421]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218045120))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219490752))))[name = string("module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 4096]> linear_145_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_16_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_853_cast_fp16)[name = string("linear_145_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_857_cast_fp16 = silu(x = linear_145_cast_fp16)[name = string("input_857_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1927421]> module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209393792))), indices_nonzero_data = tensor<uint6, [1927421]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209918144))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211363776))))[name = string("module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_146_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_857_cast_fp16)[name = string("linear_146_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_146_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_857_cast_fp16)[name = string("linear_146_cast_fp16")];
             fp16 var_2840_to_fp16 = const()[name = string("op_2840_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_2841_cast_fp16 = mul(x = linear_146_cast_fp16, y = var_2840_to_fp16)[name = string("op_2841_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_863_cast_fp16 = add(x = input_851_cast_fp16, y = var_2841_cast_fp16)[name = string("input_863_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> var_2841_cast_fp16 = mul(x = linear_146_cast_fp16, y = var_2840_to_fp16)[name = string("op_2841_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_863_cast_fp16 = add(x = input_851_cast_fp16, y = var_2841_cast_fp16)[name = string("input_863_cast_fp16")];
             tensor<int32, [1]> query_33_axes_0 = const()[name = string("query_33_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_16_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_16_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219507200)))];
-            tensor<fp16, [1024]> module_layers_16_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_16_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219509312)))];
-            tensor<fp16, [1, 188, 1024]> query_33_cast_fp16 = layer_norm(axes = query_33_axes_0, beta = module_layers_16_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_self_att_weight_to_fp16, x = input_863_cast_fp16)[name = string("query_33_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524478]> module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219511424))), indices_nonzero_data = tensor<uint6, [524478]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219642560))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220036032))))[name = string("module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_16_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_16_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211380224)))];
+            tensor<fp16, [1024]> module_layers_16_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_16_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211382336)))];
+            tensor<fp16, [1, 64, 1024]> query_33_cast_fp16 = layer_norm(axes = query_33_axes_0, beta = module_layers_16_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_self_att_weight_to_fp16, x = input_863_cast_fp16)[name = string("query_33_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524478]> module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211384448))), indices_nonzero_data = tensor<uint6, [524478]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211515584))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211909056))))[name = string("module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_147_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_33_cast_fp16)[name = string("linear_147_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_147_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_33_cast_fp16)[name = string("linear_147_cast_fp16")];
             tensor<int32, [4]> var_2857 = const()[name = string("op_2857"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_97_cast_fp16 = reshape(shape = var_2857, x = linear_147_cast_fp16)[name = string("q_97_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524779]> module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220052480))), indices_nonzero_data = tensor<uint6, [524779]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220183616))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220577280))))[name = string("module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> q_97_cast_fp16 = reshape(shape = var_2857, x = linear_147_cast_fp16)[name = string("q_97_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524779]> module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211925504))), indices_nonzero_data = tensor<uint6, [524779]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212056640))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212450304))))[name = string("module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_148_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_33_cast_fp16)[name = string("linear_148_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_148_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_33_cast_fp16)[name = string("linear_148_cast_fp16")];
             tensor<int32, [4]> var_2861 = const()[name = string("op_2861"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_65_cast_fp16 = reshape(shape = var_2861, x = linear_148_cast_fp16)[name = string("k_65_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525290]> module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220593728))), indices_nonzero_data = tensor<uint6, [525290]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220724864))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221118912))))[name = string("module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> k_65_cast_fp16 = reshape(shape = var_2861, x = linear_148_cast_fp16)[name = string("k_65_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525290]> module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212466752))), indices_nonzero_data = tensor<uint6, [525290]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212597888))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212991936))))[name = string("module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_149_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_33_cast_fp16)[name = string("linear_149_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_149_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_33_cast_fp16)[name = string("linear_149_cast_fp16")];
             tensor<int32, [4]> var_2865 = const()[name = string("op_2865"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_33_cast_fp16 = reshape(shape = var_2865, x = linear_149_cast_fp16)[name = string("v_33_cast_fp16")];
+            tensor<fp16, [1, 64, 8, 128]> v_33_cast_fp16 = reshape(shape = var_2865, x = linear_149_cast_fp16)[name = string("v_33_cast_fp16")];
             tensor<int32, [4]> value_35_perm_0 = const()[name = string("value_35_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_16_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_16_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221135360)))];
-            tensor<fp16, [1, 188, 8, 128]> var_2877_cast_fp16 = add(x = q_97_cast_fp16, y = module_layers_16_self_attn_pos_bias_u_to_fp16)[name = string("op_2877_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_16_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_16_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221137472)))];
-            tensor<fp16, [1, 188, 8, 128]> var_2879_cast_fp16 = add(x = q_97_cast_fp16, y = module_layers_16_self_attn_pos_bias_v_to_fp16)[name = string("op_2879_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_16_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_16_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213008384)))];
+            tensor<fp16, [1, 64, 8, 128]> var_2877_cast_fp16 = add(x = q_97_cast_fp16, y = module_layers_16_self_attn_pos_bias_u_to_fp16)[name = string("op_2877_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_16_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_16_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213010496)))];
+            tensor<fp16, [1, 64, 8, 128]> var_2879_cast_fp16 = add(x = q_97_cast_fp16, y = module_layers_16_self_attn_pos_bias_v_to_fp16)[name = string("op_2879_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_33_perm_0 = const()[name = string("q_with_bias_v_33_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_359_transpose_x_0 = const()[name = string("x_359_transpose_x_0"), val = bool(false)];
             bool x_359_transpose_y_0 = const()[name = string("x_359_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_2881_to_fp16 = const()[name = string("op_2881_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221139584)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_33_cast_fp16 = transpose(perm = q_with_bias_v_33_perm_0, x = var_2879_cast_fp16)[name = string("transpose_200")];
-            tensor<fp16, [1, 8, 188, 375]> x_359_cast_fp16 = matmul(transpose_x = x_359_transpose_x_0, transpose_y = x_359_transpose_y_0, x = q_with_bias_v_33_cast_fp16, y = var_2881_to_fp16)[name = string("x_359_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 127]> var_2881_to_fp16 = const()[name = string("op_2881_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213012608)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_33_cast_fp16 = transpose(perm = q_with_bias_v_33_perm_0, x = var_2879_cast_fp16)[name = string("transpose_200")];
+            tensor<fp16, [1, 8, 64, 127]> x_359_cast_fp16 = matmul(transpose_x = x_359_transpose_x_0, transpose_y = x_359_transpose_y_0, x = q_with_bias_v_33_cast_fp16, y = var_2881_to_fp16)[name = string("x_359_cast_fp16")];
             tensor<int32, [8]> x_361_pad_0 = const()[name = string("x_361_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_361_mode_0 = const()[name = string("x_361_mode_0"), val = string("constant")];
             fp16 const_174_to_fp16 = const()[name = string("const_174_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_361_cast_fp16 = pad(constant_val = const_174_to_fp16, mode = x_361_mode_0, pad = x_361_pad_0, x = x_359_cast_fp16)[name = string("x_361_cast_fp16")];
-            tensor<int32, [4]> var_2889 = const()[name = string("op_2889"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_363_cast_fp16 = reshape(shape = var_2889, x = x_361_cast_fp16)[name = string("x_363_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 128]> x_361_cast_fp16 = pad(constant_val = const_174_to_fp16, mode = x_361_mode_0, pad = x_361_pad_0, x = x_359_cast_fp16)[name = string("x_361_cast_fp16")];
+            tensor<int32, [4]> var_2889 = const()[name = string("op_2889"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_363_cast_fp16 = reshape(shape = var_2889, x = x_361_cast_fp16)[name = string("x_363_cast_fp16")];
             tensor<int32, [4]> var_2893_begin_0 = const()[name = string("op_2893_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_2893_end_0 = const()[name = string("op_2893_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_2893_end_0 = const()[name = string("op_2893_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
             tensor<bool, [4]> var_2893_end_mask_0 = const()[name = string("op_2893_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_2893_cast_fp16 = slice_by_index(begin = var_2893_begin_0, end = var_2893_end_0, end_mask = var_2893_end_mask_0, x = x_363_cast_fp16)[name = string("op_2893_cast_fp16")];
-            tensor<int32, [4]> var_2894 = const()[name = string("op_2894"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_65_cast_fp16 = reshape(shape = var_2894, x = var_2893_cast_fp16)[name = string("matrix_bd_65_cast_fp16")];
+            tensor<fp16, [1, 8, 127, 64]> var_2893_cast_fp16 = slice_by_index(begin = var_2893_begin_0, end = var_2893_end_0, end_mask = var_2893_end_mask_0, x = x_363_cast_fp16)[name = string("op_2893_cast_fp16")];
+            tensor<int32, [4]> var_2894 = const()[name = string("op_2894"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_65_cast_fp16 = reshape(shape = var_2894, x = var_2893_cast_fp16)[name = string("matrix_bd_65_cast_fp16")];
             bool matrix_ac_33_transpose_x_0 = const()[name = string("matrix_ac_33_transpose_x_0"), val = bool(false)];
             bool matrix_ac_33_transpose_y_0 = const()[name = string("matrix_ac_33_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_128_perm_0 = const()[name = string("transpose_128_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_129_perm_0 = const()[name = string("transpose_129_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_129 = transpose(perm = transpose_129_perm_0, x = k_65_cast_fp16)[name = string("transpose_198")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_128 = transpose(perm = transpose_128_perm_0, x = var_2877_cast_fp16)[name = string("transpose_199")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_33_cast_fp16 = matmul(transpose_x = matrix_ac_33_transpose_x_0, transpose_y = matrix_ac_33_transpose_y_0, x = transpose_128, y = transpose_129)[name = string("matrix_ac_33_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 64]> transpose_129 = transpose(perm = transpose_129_perm_0, x = k_65_cast_fp16)[name = string("transpose_198")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_128 = transpose(perm = transpose_128_perm_0, x = var_2877_cast_fp16)[name = string("transpose_199")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_33_cast_fp16 = matmul(transpose_x = matrix_ac_33_transpose_x_0, transpose_y = matrix_ac_33_transpose_y_0, x = transpose_128, y = transpose_129)[name = string("matrix_ac_33_cast_fp16")];
             tensor<int32, [4]> matrix_bd_67_begin_0 = const()[name = string("matrix_bd_67_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_67_end_0 = const()[name = string("matrix_bd_67_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_67_end_0 = const()[name = string("matrix_bd_67_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
             tensor<bool, [4]> matrix_bd_67_end_mask_0 = const()[name = string("matrix_bd_67_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_67_cast_fp16 = slice_by_index(begin = matrix_bd_67_begin_0, end = matrix_bd_67_end_0, end_mask = matrix_bd_67_end_mask_0, x = matrix_bd_65_cast_fp16)[name = string("matrix_bd_67_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_2903_cast_fp16 = add(x = matrix_ac_33_cast_fp16, y = matrix_bd_67_cast_fp16)[name = string("op_2903_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_67_cast_fp16 = slice_by_index(begin = matrix_bd_67_begin_0, end = matrix_bd_67_end_0, end_mask = matrix_bd_67_end_mask_0, x = matrix_bd_65_cast_fp16)[name = string("matrix_bd_67_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_2903_cast_fp16 = add(x = matrix_ac_33_cast_fp16, y = matrix_bd_67_cast_fp16)[name = string("op_2903_cast_fp16")];
             fp16 _inversed_scores_65_y_0_to_fp16 = const()[name = string("_inversed_scores_65_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_65_cast_fp16 = mul(x = var_2903_cast_fp16, y = _inversed_scores_65_y_0_to_fp16)[name = string("_inversed_scores_65_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_67_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_65_cast_fp16, cond = mask_3)[name = string("scores_67_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_2909_cast_fp16 = softmax(axis = var_30, x = scores_67_cast_fp16)[name = string("op_2909_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_865_cast_fp16 = select(a = var_11_to_fp16, b = var_2909_cast_fp16, cond = mask_3)[name = string("input_865_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_65_cast_fp16 = mul(x = var_2903_cast_fp16, y = _inversed_scores_65_y_0_to_fp16)[name = string("_inversed_scores_65_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_67_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_65_cast_fp16, cond = mask_3)[name = string("scores_67_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_2909_cast_fp16 = softmax(axis = var_30, x = scores_67_cast_fp16)[name = string("op_2909_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_865_cast_fp16 = select(a = var_11_to_fp16, b = var_2909_cast_fp16, cond = mask_3)[name = string("input_865_cast_fp16")];
             bool x_365_transpose_x_0 = const()[name = string("x_365_transpose_x_0"), val = bool(false)];
             bool x_365_transpose_y_0 = const()[name = string("x_365_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_35_cast_fp16 = transpose(perm = value_35_perm_0, x = v_33_cast_fp16)[name = string("transpose_197")];
-            tensor<fp16, [1, 8, 188, 128]> x_365_cast_fp16 = matmul(transpose_x = x_365_transpose_x_0, transpose_y = x_365_transpose_y_0, x = input_865_cast_fp16, y = value_35_cast_fp16)[name = string("x_365_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 128]> value_35_cast_fp16 = transpose(perm = value_35_perm_0, x = v_33_cast_fp16)[name = string("transpose_197")];
+            tensor<fp16, [1, 8, 64, 128]> x_365_cast_fp16 = matmul(transpose_x = x_365_transpose_x_0, transpose_y = x_365_transpose_y_0, x = input_865_cast_fp16, y = value_35_cast_fp16)[name = string("x_365_cast_fp16")];
             tensor<int32, [4]> var_2913_perm_0 = const()[name = string("op_2913_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_2914 = const()[name = string("op_2914"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_2913_cast_fp16 = transpose(perm = var_2913_perm_0, x = x_365_cast_fp16)[name = string("transpose_196")];
-            tensor<fp16, [1, 188, 1024]> input_867_cast_fp16 = reshape(shape = var_2914, x = var_2913_cast_fp16)[name = string("input_867_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523537]> module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221907648))), indices_nonzero_data = tensor<uint6, [523537]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222038784))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222431552))))[name = string("module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> var_2913_cast_fp16 = transpose(perm = var_2913_perm_0, x = x_365_cast_fp16)[name = string("transpose_196")];
+            tensor<fp16, [1, 64, 1024]> input_867_cast_fp16 = reshape(shape = var_2914, x = var_2913_cast_fp16)[name = string("input_867_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523537]> module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213272768))), indices_nonzero_data = tensor<uint6, [523537]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213403904))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213796672))))[name = string("module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_151_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_867_cast_fp16)[name = string("linear_151_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_871_cast_fp16 = add(x = input_863_cast_fp16, y = linear_151_cast_fp16)[name = string("input_871_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_151_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_867_cast_fp16)[name = string("linear_151_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_871_cast_fp16 = add(x = input_863_cast_fp16, y = linear_151_cast_fp16)[name = string("input_871_cast_fp16")];
             tensor<int32, [1]> x_369_axes_0 = const()[name = string("x_369_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_16_norm_conv_weight_to_fp16 = const()[name = string("module_layers_16_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222448000)))];
-            tensor<fp16, [1024]> module_layers_16_norm_conv_bias_to_fp16 = const()[name = string("module_layers_16_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222450112)))];
-            tensor<fp16, [1, 188, 1024]> x_369_cast_fp16 = layer_norm(axes = x_369_axes_0, beta = module_layers_16_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_conv_weight_to_fp16, x = input_871_cast_fp16)[name = string("x_369_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_16_norm_conv_weight_to_fp16 = const()[name = string("module_layers_16_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213813120)))];
+            tensor<fp16, [1024]> module_layers_16_norm_conv_bias_to_fp16 = const()[name = string("module_layers_16_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213815232)))];
+            tensor<fp16, [1, 64, 1024]> x_369_cast_fp16 = layer_norm(axes = x_369_axes_0, beta = module_layers_16_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_conv_weight_to_fp16, x = input_871_cast_fp16)[name = string("x_369_cast_fp16")];
             tensor<int32, [3]> input_873_perm_0 = const()[name = string("input_873_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_875_pad_type_0 = const()[name = string("input_875_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_875_strides_0 = const()[name = string("input_875_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_875_pad_0 = const()[name = string("input_875_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_875_dilations_0 = const()[name = string("input_875_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_875_groups_0 = const()[name = string("input_875_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047665]> module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222452224))), indices_nonzero_data = tensor<uint6, [1047665]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222714432))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223500288))))[name = string("module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047665]> module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213817344))), indices_nonzero_data = tensor<uint6, [1047665]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214079552))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214865408))))[name = string("module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_873_cast_fp16 = transpose(perm = input_873_perm_0, x = x_369_cast_fp16)[name = string("transpose_195")];
-            tensor<fp16, [1, 2048, 188]> input_875_cast_fp16 = conv(dilations = input_875_dilations_0, groups = input_875_groups_0, pad = input_875_pad_0, pad_type = input_875_pad_type_0, strides = input_875_strides_0, weight = module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_873_cast_fp16)[name = string("input_875_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_873_cast_fp16 = transpose(perm = input_873_perm_0, x = x_369_cast_fp16)[name = string("transpose_195")];
+            tensor<fp16, [1, 2048, 64]> input_875_cast_fp16 = conv(dilations = input_875_dilations_0, groups = input_875_groups_0, pad = input_875_pad_0, pad_type = input_875_pad_type_0, strides = input_875_strides_0, weight = module_layers_16_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_873_cast_fp16)[name = string("input_875_cast_fp16")];
             int32 x_371_split_num_splits_0 = const()[name = string("x_371_split_num_splits_0"), val = int32(2)];
             int32 x_371_split_axis_0 = const()[name = string("x_371_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_371_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_371_split_cast_fp16_1 = split(axis = x_371_split_axis_0, num_splits = x_371_split_num_splits_0, x = input_875_cast_fp16)[name = string("x_371_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_371_split_1_sigmoid_cast_fp16 = sigmoid(x = x_371_split_cast_fp16_1)[name = string("x_371_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_371_cast_fp16 = mul(x = x_371_split_cast_fp16_0, y = x_371_split_1_sigmoid_cast_fp16)[name = string("x_371_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_877_cast_fp16 = select(a = var_11_to_fp16, b = x_371_cast_fp16, cond = var_328)[name = string("input_877_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_371_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_371_split_cast_fp16_1 = split(axis = x_371_split_axis_0, num_splits = x_371_split_num_splits_0, x = input_875_cast_fp16)[name = string("x_371_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_371_split_1_sigmoid_cast_fp16 = sigmoid(x = x_371_split_cast_fp16_1)[name = string("x_371_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_371_cast_fp16 = mul(x = x_371_split_cast_fp16_0, y = x_371_split_1_sigmoid_cast_fp16)[name = string("x_371_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_877_cast_fp16 = select(a = var_11_to_fp16, b = x_371_cast_fp16, cond = var_328)[name = string("input_877_cast_fp16")];
             tensor<int32, [6]> input_879_pad_0 = const()[name = string("input_879_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_879_mode_0 = const()[name = string("input_879_mode_0"), val = string("constant")];
             fp16 const_177_to_fp16 = const()[name = string("const_177_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_879_cast_fp16 = pad(constant_val = const_177_to_fp16, mode = input_879_mode_0, pad = input_879_pad_0, x = input_877_cast_fp16)[name = string("input_879_cast_fp16")];
+            tensor<fp16, [1, 1024, 72]> input_879_cast_fp16 = pad(constant_val = const_177_to_fp16, mode = input_879_mode_0, pad = input_879_pad_0, x = input_877_cast_fp16)[name = string("input_879_cast_fp16")];
             string input_881_pad_type_0 = const()[name = string("input_881_pad_type_0"), val = string("valid")];
             int32 input_881_groups_0 = const()[name = string("input_881_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_881_strides_0 = const()[name = string("input_881_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_881_pad_0 = const()[name = string("input_881_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_881_dilations_0 = const()[name = string("input_881_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_280_to_fp16 = const()[name = string("const_280_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223533120)))];
-            tensor<fp16, [1024]> const_281_to_fp16 = const()[name = string("const_281_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223551616)))];
-            tensor<fp16, [1, 1024, 188]> input_883_cast_fp16 = conv(bias = const_281_to_fp16, dilations = input_881_dilations_0, groups = input_881_groups_0, pad = input_881_pad_0, pad_type = input_881_pad_type_0, strides = input_881_strides_0, weight = const_280_to_fp16, x = input_879_cast_fp16)[name = string("input_883_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_885_cast_fp16 = silu(x = input_883_cast_fp16)[name = string("input_885_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_280_to_fp16 = const()[name = string("const_280_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214898240)))];
+            tensor<fp16, [1024]> const_281_to_fp16 = const()[name = string("const_281_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214916736)))];
+            tensor<fp16, [1, 1024, 64]> input_883_cast_fp16 = conv(bias = const_281_to_fp16, dilations = input_881_dilations_0, groups = input_881_groups_0, pad = input_881_pad_0, pad_type = input_881_pad_type_0, strides = input_881_strides_0, weight = const_280_to_fp16, x = input_879_cast_fp16)[name = string("input_883_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_885_cast_fp16 = silu(x = input_883_cast_fp16)[name = string("input_885_cast_fp16")];
             string x_373_pad_type_0 = const()[name = string("x_373_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_373_strides_0 = const()[name = string("x_373_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_373_pad_0 = const()[name = string("x_373_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_373_dilations_0 = const()[name = string("x_373_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_373_groups_0 = const()[name = string("x_373_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523727]> module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223553728))), indices_nonzero_data = tensor<uint6, [523727]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223684864))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224077760))))[name = string("module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523727]> module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214918848))), indices_nonzero_data = tensor<uint6, [523727]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215049984))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215442880))))[name = string("module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_373_cast_fp16 = conv(dilations = x_373_dilations_0, groups = x_373_groups_0, pad = x_373_pad_0, pad_type = x_373_pad_type_0, strides = x_373_strides_0, weight = module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_885_cast_fp16)[name = string("x_373_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_373_cast_fp16 = conv(dilations = x_373_dilations_0, groups = x_373_groups_0, pad = x_373_pad_0, pad_type = x_373_pad_type_0, strides = x_373_strides_0, weight = module_layers_16_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_885_cast_fp16)[name = string("x_373_cast_fp16")];
             tensor<int32, [3]> input_887_perm_0 = const()[name = string("input_887_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_887_cast_fp16 = transpose(perm = input_887_perm_0, x = x_373_cast_fp16)[name = string("transpose_194")];
-            tensor<fp16, [1, 188, 1024]> input_889_cast_fp16 = add(x = input_871_cast_fp16, y = input_887_cast_fp16)[name = string("input_889_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_887_cast_fp16 = transpose(perm = input_887_perm_0, x = x_373_cast_fp16)[name = string("transpose_194")];
+            tensor<fp16, [1, 64, 1024]> input_889_cast_fp16 = add(x = input_871_cast_fp16, y = input_887_cast_fp16)[name = string("input_889_cast_fp16")];
             tensor<int32, [1]> input_891_axes_0 = const()[name = string("input_891_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_16_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_16_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224094208)))];
-            tensor<fp16, [1024]> module_layers_16_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_16_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224096320)))];
-            tensor<fp16, [1, 188, 1024]> input_891_cast_fp16 = layer_norm(axes = input_891_axes_0, beta = module_layers_16_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_feed_forward2_weight_to_fp16, x = input_889_cast_fp16)[name = string("input_891_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2085434]> module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224098432))), indices_nonzero_data = tensor<uint6, [2085434]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224622784))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226186944))))[name = string("module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_16_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_16_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215459328)))];
+            tensor<fp16, [1024]> module_layers_16_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_16_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215461440)))];
+            tensor<fp16, [1, 64, 1024]> input_891_cast_fp16 = layer_norm(axes = input_891_axes_0, beta = module_layers_16_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_feed_forward2_weight_to_fp16, x = input_889_cast_fp16)[name = string("input_891_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2085434]> module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215463552))), indices_nonzero_data = tensor<uint6, [2085434]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215987904))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217552064))))[name = string("module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_152_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_891_cast_fp16)[name = string("linear_152_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_895_cast_fp16 = silu(x = linear_152_cast_fp16)[name = string("input_895_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2028445]> module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226252544))), indices_nonzero_data = tensor<uint6, [2028445]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226776896))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228298304))))[name = string("module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 4096]> linear_152_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_16_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_891_cast_fp16)[name = string("linear_152_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_895_cast_fp16 = silu(x = linear_152_cast_fp16)[name = string("input_895_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2028445]> module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217617664))), indices_nonzero_data = tensor<uint6, [2028445]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218142016))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219663424))))[name = string("module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_153_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_895_cast_fp16)[name = string("linear_153_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_153_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_16_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_895_cast_fp16)[name = string("linear_153_cast_fp16")];
             fp16 var_2974_to_fp16 = const()[name = string("op_2974_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_2975_cast_fp16 = mul(x = linear_153_cast_fp16, y = var_2974_to_fp16)[name = string("op_2975_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_901_cast_fp16 = add(x = input_889_cast_fp16, y = var_2975_cast_fp16)[name = string("input_901_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> var_2975_cast_fp16 = mul(x = linear_153_cast_fp16, y = var_2974_to_fp16)[name = string("op_2975_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_901_cast_fp16 = add(x = input_889_cast_fp16, y = var_2975_cast_fp16)[name = string("input_901_cast_fp16")];
             tensor<int32, [1]> input_903_axes_0 = const()[name = string("input_903_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_16_norm_out_weight_to_fp16 = const()[name = string("module_layers_16_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228314752)))];
-            tensor<fp16, [1024]> module_layers_16_norm_out_bias_to_fp16 = const()[name = string("module_layers_16_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228316864)))];
-            tensor<fp16, [1, 188, 1024]> input_903_cast_fp16 = layer_norm(axes = input_903_axes_0, beta = module_layers_16_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_out_weight_to_fp16, x = input_901_cast_fp16)[name = string("input_903_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_16_norm_out_weight_to_fp16 = const()[name = string("module_layers_16_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219679872)))];
+            tensor<fp16, [1024]> module_layers_16_norm_out_bias_to_fp16 = const()[name = string("module_layers_16_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219681984)))];
+            tensor<fp16, [1, 64, 1024]> input_903_cast_fp16 = layer_norm(axes = input_903_axes_0, beta = module_layers_16_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_out_weight_to_fp16, x = input_901_cast_fp16)[name = string("input_903_cast_fp16")];
             tensor<int32, [1]> input_905_axes_0 = const()[name = string("input_905_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_17_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_17_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228318976)))];
-            tensor<fp16, [1024]> module_layers_17_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_17_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228321088)))];
-            tensor<fp16, [1, 188, 1024]> input_905_cast_fp16 = layer_norm(axes = input_905_axes_0, beta = module_layers_17_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_feed_forward1_weight_to_fp16, x = input_903_cast_fp16)[name = string("input_905_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2078697]> module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228323200))), indices_nonzero_data = tensor<uint6, [2078697]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228847552))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230406656))))[name = string("module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_17_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_17_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219684096)))];
+            tensor<fp16, [1024]> module_layers_17_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_17_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219686208)))];
+            tensor<fp16, [1, 64, 1024]> input_905_cast_fp16 = layer_norm(axes = input_905_axes_0, beta = module_layers_17_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_feed_forward1_weight_to_fp16, x = input_903_cast_fp16)[name = string("input_905_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2078697]> module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219688320))), indices_nonzero_data = tensor<uint6, [2078697]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220212672))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221771776))))[name = string("module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_154_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_905_cast_fp16)[name = string("linear_154_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_909_cast_fp16 = silu(x = linear_154_cast_fp16)[name = string("input_909_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2015245]> module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230472256))), indices_nonzero_data = tensor<uint6, [2015245]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230996608))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232508160))))[name = string("module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 4096]> linear_154_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_17_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_905_cast_fp16)[name = string("linear_154_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_909_cast_fp16 = silu(x = linear_154_cast_fp16)[name = string("input_909_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2015245]> module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221837376))), indices_nonzero_data = tensor<uint6, [2015245]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222361728))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223873280))))[name = string("module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_155_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_909_cast_fp16)[name = string("linear_155_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_155_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_909_cast_fp16)[name = string("linear_155_cast_fp16")];
             fp16 var_3003_to_fp16 = const()[name = string("op_3003_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_3004_cast_fp16 = mul(x = linear_155_cast_fp16, y = var_3003_to_fp16)[name = string("op_3004_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_915_cast_fp16 = add(x = input_903_cast_fp16, y = var_3004_cast_fp16)[name = string("input_915_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> var_3004_cast_fp16 = mul(x = linear_155_cast_fp16, y = var_3003_to_fp16)[name = string("op_3004_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_915_cast_fp16 = add(x = input_903_cast_fp16, y = var_3004_cast_fp16)[name = string("input_915_cast_fp16")];
             tensor<int32, [1]> query_35_axes_0 = const()[name = string("query_35_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_17_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_17_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232524608)))];
-            tensor<fp16, [1024]> module_layers_17_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_17_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232526720)))];
-            tensor<fp16, [1, 188, 1024]> query_35_cast_fp16 = layer_norm(axes = query_35_axes_0, beta = module_layers_17_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_self_att_weight_to_fp16, x = input_915_cast_fp16)[name = string("query_35_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525433]> module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232528832))), indices_nonzero_data = tensor<uint6, [525433]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232659968))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233054144))))[name = string("module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_17_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_17_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223889728)))];
+            tensor<fp16, [1024]> module_layers_17_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_17_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223891840)))];
+            tensor<fp16, [1, 64, 1024]> query_35_cast_fp16 = layer_norm(axes = query_35_axes_0, beta = module_layers_17_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_self_att_weight_to_fp16, x = input_915_cast_fp16)[name = string("query_35_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525433]> module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223893952))), indices_nonzero_data = tensor<uint6, [525433]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224025088))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224419264))))[name = string("module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_156_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_35_cast_fp16)[name = string("linear_156_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_156_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_35_cast_fp16)[name = string("linear_156_cast_fp16")];
             tensor<int32, [4]> var_3020 = const()[name = string("op_3020"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_103_cast_fp16 = reshape(shape = var_3020, x = linear_156_cast_fp16)[name = string("q_103_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521661]> module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233070592))), indices_nonzero_data = tensor<uint6, [521661]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233201728))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233593088))))[name = string("module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> q_103_cast_fp16 = reshape(shape = var_3020, x = linear_156_cast_fp16)[name = string("q_103_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521661]> module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224435712))), indices_nonzero_data = tensor<uint6, [521661]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224566848))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224958208))))[name = string("module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_157_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_35_cast_fp16)[name = string("linear_157_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_157_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_35_cast_fp16)[name = string("linear_157_cast_fp16")];
             tensor<int32, [4]> var_3024 = const()[name = string("op_3024"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_69_cast_fp16 = reshape(shape = var_3024, x = linear_157_cast_fp16)[name = string("k_69_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521931]> module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233609536))), indices_nonzero_data = tensor<uint6, [521931]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233740672))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234132224))))[name = string("module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> k_69_cast_fp16 = reshape(shape = var_3024, x = linear_157_cast_fp16)[name = string("k_69_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521931]> module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224974656))), indices_nonzero_data = tensor<uint6, [521931]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225105792))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225497344))))[name = string("module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_158_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_35_cast_fp16)[name = string("linear_158_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_158_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_35_cast_fp16)[name = string("linear_158_cast_fp16")];
             tensor<int32, [4]> var_3028 = const()[name = string("op_3028"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_35_cast_fp16 = reshape(shape = var_3028, x = linear_158_cast_fp16)[name = string("v_35_cast_fp16")];
+            tensor<fp16, [1, 64, 8, 128]> v_35_cast_fp16 = reshape(shape = var_3028, x = linear_158_cast_fp16)[name = string("v_35_cast_fp16")];
             tensor<int32, [4]> value_37_perm_0 = const()[name = string("value_37_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_17_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_17_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234148672)))];
-            tensor<fp16, [1, 188, 8, 128]> var_3040_cast_fp16 = add(x = q_103_cast_fp16, y = module_layers_17_self_attn_pos_bias_u_to_fp16)[name = string("op_3040_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_17_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_17_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234150784)))];
-            tensor<fp16, [1, 188, 8, 128]> var_3042_cast_fp16 = add(x = q_103_cast_fp16, y = module_layers_17_self_attn_pos_bias_v_to_fp16)[name = string("op_3042_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_17_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_17_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225513792)))];
+            tensor<fp16, [1, 64, 8, 128]> var_3040_cast_fp16 = add(x = q_103_cast_fp16, y = module_layers_17_self_attn_pos_bias_u_to_fp16)[name = string("op_3040_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_17_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_17_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225515904)))];
+            tensor<fp16, [1, 64, 8, 128]> var_3042_cast_fp16 = add(x = q_103_cast_fp16, y = module_layers_17_self_attn_pos_bias_v_to_fp16)[name = string("op_3042_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_35_perm_0 = const()[name = string("q_with_bias_v_35_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_381_transpose_x_0 = const()[name = string("x_381_transpose_x_0"), val = bool(false)];
             bool x_381_transpose_y_0 = const()[name = string("x_381_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_3044_to_fp16 = const()[name = string("op_3044_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234152896)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_35_cast_fp16 = transpose(perm = q_with_bias_v_35_perm_0, x = var_3042_cast_fp16)[name = string("transpose_193")];
-            tensor<fp16, [1, 8, 188, 375]> x_381_cast_fp16 = matmul(transpose_x = x_381_transpose_x_0, transpose_y = x_381_transpose_y_0, x = q_with_bias_v_35_cast_fp16, y = var_3044_to_fp16)[name = string("x_381_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 127]> var_3044_to_fp16 = const()[name = string("op_3044_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225518016)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_35_cast_fp16 = transpose(perm = q_with_bias_v_35_perm_0, x = var_3042_cast_fp16)[name = string("transpose_193")];
+            tensor<fp16, [1, 8, 64, 127]> x_381_cast_fp16 = matmul(transpose_x = x_381_transpose_x_0, transpose_y = x_381_transpose_y_0, x = q_with_bias_v_35_cast_fp16, y = var_3044_to_fp16)[name = string("x_381_cast_fp16")];
             tensor<int32, [8]> x_383_pad_0 = const()[name = string("x_383_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_383_mode_0 = const()[name = string("x_383_mode_0"), val = string("constant")];
             fp16 const_184_to_fp16 = const()[name = string("const_184_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_383_cast_fp16 = pad(constant_val = const_184_to_fp16, mode = x_383_mode_0, pad = x_383_pad_0, x = x_381_cast_fp16)[name = string("x_383_cast_fp16")];
-            tensor<int32, [4]> var_3052 = const()[name = string("op_3052"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_385_cast_fp16 = reshape(shape = var_3052, x = x_383_cast_fp16)[name = string("x_385_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 128]> x_383_cast_fp16 = pad(constant_val = const_184_to_fp16, mode = x_383_mode_0, pad = x_383_pad_0, x = x_381_cast_fp16)[name = string("x_383_cast_fp16")];
+            tensor<int32, [4]> var_3052 = const()[name = string("op_3052"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_385_cast_fp16 = reshape(shape = var_3052, x = x_383_cast_fp16)[name = string("x_385_cast_fp16")];
             tensor<int32, [4]> var_3056_begin_0 = const()[name = string("op_3056_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_3056_end_0 = const()[name = string("op_3056_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_3056_end_0 = const()[name = string("op_3056_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
             tensor<bool, [4]> var_3056_end_mask_0 = const()[name = string("op_3056_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_3056_cast_fp16 = slice_by_index(begin = var_3056_begin_0, end = var_3056_end_0, end_mask = var_3056_end_mask_0, x = x_385_cast_fp16)[name = string("op_3056_cast_fp16")];
-            tensor<int32, [4]> var_3057 = const()[name = string("op_3057"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_69_cast_fp16 = reshape(shape = var_3057, x = var_3056_cast_fp16)[name = string("matrix_bd_69_cast_fp16")];
+            tensor<fp16, [1, 8, 127, 64]> var_3056_cast_fp16 = slice_by_index(begin = var_3056_begin_0, end = var_3056_end_0, end_mask = var_3056_end_mask_0, x = x_385_cast_fp16)[name = string("op_3056_cast_fp16")];
+            tensor<int32, [4]> var_3057 = const()[name = string("op_3057"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_69_cast_fp16 = reshape(shape = var_3057, x = var_3056_cast_fp16)[name = string("matrix_bd_69_cast_fp16")];
             bool matrix_ac_35_transpose_x_0 = const()[name = string("matrix_ac_35_transpose_x_0"), val = bool(false)];
             bool matrix_ac_35_transpose_y_0 = const()[name = string("matrix_ac_35_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_130_perm_0 = const()[name = string("transpose_130_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_131_perm_0 = const()[name = string("transpose_131_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_131 = transpose(perm = transpose_131_perm_0, x = k_69_cast_fp16)[name = string("transpose_191")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_130 = transpose(perm = transpose_130_perm_0, x = var_3040_cast_fp16)[name = string("transpose_192")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_35_cast_fp16 = matmul(transpose_x = matrix_ac_35_transpose_x_0, transpose_y = matrix_ac_35_transpose_y_0, x = transpose_130, y = transpose_131)[name = string("matrix_ac_35_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 64]> transpose_131 = transpose(perm = transpose_131_perm_0, x = k_69_cast_fp16)[name = string("transpose_191")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_130 = transpose(perm = transpose_130_perm_0, x = var_3040_cast_fp16)[name = string("transpose_192")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_35_cast_fp16 = matmul(transpose_x = matrix_ac_35_transpose_x_0, transpose_y = matrix_ac_35_transpose_y_0, x = transpose_130, y = transpose_131)[name = string("matrix_ac_35_cast_fp16")];
             tensor<int32, [4]> matrix_bd_71_begin_0 = const()[name = string("matrix_bd_71_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_71_end_0 = const()[name = string("matrix_bd_71_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_71_end_0 = const()[name = string("matrix_bd_71_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
             tensor<bool, [4]> matrix_bd_71_end_mask_0 = const()[name = string("matrix_bd_71_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_71_cast_fp16 = slice_by_index(begin = matrix_bd_71_begin_0, end = matrix_bd_71_end_0, end_mask = matrix_bd_71_end_mask_0, x = matrix_bd_69_cast_fp16)[name = string("matrix_bd_71_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_3066_cast_fp16 = add(x = matrix_ac_35_cast_fp16, y = matrix_bd_71_cast_fp16)[name = string("op_3066_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_71_cast_fp16 = slice_by_index(begin = matrix_bd_71_begin_0, end = matrix_bd_71_end_0, end_mask = matrix_bd_71_end_mask_0, x = matrix_bd_69_cast_fp16)[name = string("matrix_bd_71_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_3066_cast_fp16 = add(x = matrix_ac_35_cast_fp16, y = matrix_bd_71_cast_fp16)[name = string("op_3066_cast_fp16")];
             fp16 _inversed_scores_69_y_0_to_fp16 = const()[name = string("_inversed_scores_69_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_69_cast_fp16 = mul(x = var_3066_cast_fp16, y = _inversed_scores_69_y_0_to_fp16)[name = string("_inversed_scores_69_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_71_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_69_cast_fp16, cond = mask_3)[name = string("scores_71_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_3072_cast_fp16 = softmax(axis = var_30, x = scores_71_cast_fp16)[name = string("op_3072_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_917_cast_fp16 = select(a = var_11_to_fp16, b = var_3072_cast_fp16, cond = mask_3)[name = string("input_917_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_69_cast_fp16 = mul(x = var_3066_cast_fp16, y = _inversed_scores_69_y_0_to_fp16)[name = string("_inversed_scores_69_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_71_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_69_cast_fp16, cond = mask_3)[name = string("scores_71_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_3072_cast_fp16 = softmax(axis = var_30, x = scores_71_cast_fp16)[name = string("op_3072_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_917_cast_fp16 = select(a = var_11_to_fp16, b = var_3072_cast_fp16, cond = mask_3)[name = string("input_917_cast_fp16")];
             bool x_387_transpose_x_0 = const()[name = string("x_387_transpose_x_0"), val = bool(false)];
             bool x_387_transpose_y_0 = const()[name = string("x_387_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_37_cast_fp16 = transpose(perm = value_37_perm_0, x = v_35_cast_fp16)[name = string("transpose_190")];
-            tensor<fp16, [1, 8, 188, 128]> x_387_cast_fp16 = matmul(transpose_x = x_387_transpose_x_0, transpose_y = x_387_transpose_y_0, x = input_917_cast_fp16, y = value_37_cast_fp16)[name = string("x_387_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 128]> value_37_cast_fp16 = transpose(perm = value_37_perm_0, x = v_35_cast_fp16)[name = string("transpose_190")];
+            tensor<fp16, [1, 8, 64, 128]> x_387_cast_fp16 = matmul(transpose_x = x_387_transpose_x_0, transpose_y = x_387_transpose_y_0, x = input_917_cast_fp16, y = value_37_cast_fp16)[name = string("x_387_cast_fp16")];
             tensor<int32, [4]> var_3076_perm_0 = const()[name = string("op_3076_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_3077 = const()[name = string("op_3077"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_3076_cast_fp16 = transpose(perm = var_3076_perm_0, x = x_387_cast_fp16)[name = string("transpose_189")];
-            tensor<fp16, [1, 188, 1024]> input_919_cast_fp16 = reshape(shape = var_3077, x = var_3076_cast_fp16)[name = string("input_919_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [518816]> module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234920960))), indices_nonzero_data = tensor<uint6, [518816]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235052096))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235441280))))[name = string("module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> var_3076_cast_fp16 = transpose(perm = var_3076_perm_0, x = x_387_cast_fp16)[name = string("transpose_189")];
+            tensor<fp16, [1, 64, 1024]> input_919_cast_fp16 = reshape(shape = var_3077, x = var_3076_cast_fp16)[name = string("input_919_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [518816]> module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225778176))), indices_nonzero_data = tensor<uint6, [518816]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225909312))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226298496))))[name = string("module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_160_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_919_cast_fp16)[name = string("linear_160_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_923_cast_fp16 = add(x = input_915_cast_fp16, y = linear_160_cast_fp16)[name = string("input_923_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_160_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_919_cast_fp16)[name = string("linear_160_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_923_cast_fp16 = add(x = input_915_cast_fp16, y = linear_160_cast_fp16)[name = string("input_923_cast_fp16")];
             tensor<int32, [1]> x_391_axes_0 = const()[name = string("x_391_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_17_norm_conv_weight_to_fp16 = const()[name = string("module_layers_17_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235457728)))];
-            tensor<fp16, [1024]> module_layers_17_norm_conv_bias_to_fp16 = const()[name = string("module_layers_17_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235459840)))];
-            tensor<fp16, [1, 188, 1024]> x_391_cast_fp16 = layer_norm(axes = x_391_axes_0, beta = module_layers_17_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_conv_weight_to_fp16, x = input_923_cast_fp16)[name = string("x_391_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_17_norm_conv_weight_to_fp16 = const()[name = string("module_layers_17_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226314944)))];
+            tensor<fp16, [1024]> module_layers_17_norm_conv_bias_to_fp16 = const()[name = string("module_layers_17_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226317056)))];
+            tensor<fp16, [1, 64, 1024]> x_391_cast_fp16 = layer_norm(axes = x_391_axes_0, beta = module_layers_17_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_conv_weight_to_fp16, x = input_923_cast_fp16)[name = string("x_391_cast_fp16")];
             tensor<int32, [3]> input_925_perm_0 = const()[name = string("input_925_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_927_pad_type_0 = const()[name = string("input_927_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_927_strides_0 = const()[name = string("input_927_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_927_pad_0 = const()[name = string("input_927_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_927_dilations_0 = const()[name = string("input_927_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_927_groups_0 = const()[name = string("input_927_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047749]> module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235461952))), indices_nonzero_data = tensor<uint6, [1047749]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235724160))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236510080))))[name = string("module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047749]> module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226319168))), indices_nonzero_data = tensor<uint6, [1047749]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226581376))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227367296))))[name = string("module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_925_cast_fp16 = transpose(perm = input_925_perm_0, x = x_391_cast_fp16)[name = string("transpose_188")];
-            tensor<fp16, [1, 2048, 188]> input_927_cast_fp16 = conv(dilations = input_927_dilations_0, groups = input_927_groups_0, pad = input_927_pad_0, pad_type = input_927_pad_type_0, strides = input_927_strides_0, weight = module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_925_cast_fp16)[name = string("input_927_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_925_cast_fp16 = transpose(perm = input_925_perm_0, x = x_391_cast_fp16)[name = string("transpose_188")];
+            tensor<fp16, [1, 2048, 64]> input_927_cast_fp16 = conv(dilations = input_927_dilations_0, groups = input_927_groups_0, pad = input_927_pad_0, pad_type = input_927_pad_type_0, strides = input_927_strides_0, weight = module_layers_17_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_925_cast_fp16)[name = string("input_927_cast_fp16")];
             int32 x_393_split_num_splits_0 = const()[name = string("x_393_split_num_splits_0"), val = int32(2)];
             int32 x_393_split_axis_0 = const()[name = string("x_393_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_393_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_393_split_cast_fp16_1 = split(axis = x_393_split_axis_0, num_splits = x_393_split_num_splits_0, x = input_927_cast_fp16)[name = string("x_393_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_393_split_1_sigmoid_cast_fp16 = sigmoid(x = x_393_split_cast_fp16_1)[name = string("x_393_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_393_cast_fp16 = mul(x = x_393_split_cast_fp16_0, y = x_393_split_1_sigmoid_cast_fp16)[name = string("x_393_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_929_cast_fp16 = select(a = var_11_to_fp16, b = x_393_cast_fp16, cond = var_328)[name = string("input_929_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_393_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_393_split_cast_fp16_1 = split(axis = x_393_split_axis_0, num_splits = x_393_split_num_splits_0, x = input_927_cast_fp16)[name = string("x_393_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_393_split_1_sigmoid_cast_fp16 = sigmoid(x = x_393_split_cast_fp16_1)[name = string("x_393_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_393_cast_fp16 = mul(x = x_393_split_cast_fp16_0, y = x_393_split_1_sigmoid_cast_fp16)[name = string("x_393_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_929_cast_fp16 = select(a = var_11_to_fp16, b = x_393_cast_fp16, cond = var_328)[name = string("input_929_cast_fp16")];
             tensor<int32, [6]> input_931_pad_0 = const()[name = string("input_931_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_931_mode_0 = const()[name = string("input_931_mode_0"), val = string("constant")];
             fp16 const_187_to_fp16 = const()[name = string("const_187_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_931_cast_fp16 = pad(constant_val = const_187_to_fp16, mode = input_931_mode_0, pad = input_931_pad_0, x = input_929_cast_fp16)[name = string("input_931_cast_fp16")];
+            tensor<fp16, [1, 1024, 72]> input_931_cast_fp16 = pad(constant_val = const_187_to_fp16, mode = input_931_mode_0, pad = input_931_pad_0, x = input_929_cast_fp16)[name = string("input_931_cast_fp16")];
             string input_933_pad_type_0 = const()[name = string("input_933_pad_type_0"), val = string("valid")];
             int32 input_933_groups_0 = const()[name = string("input_933_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_933_strides_0 = const()[name = string("input_933_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_933_pad_0 = const()[name = string("input_933_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_933_dilations_0 = const()[name = string("input_933_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_282_to_fp16 = const()[name = string("const_282_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236542912)))];
-            tensor<fp16, [1024]> const_283_to_fp16 = const()[name = string("const_283_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236561408)))];
-            tensor<fp16, [1, 1024, 188]> input_935_cast_fp16 = conv(bias = const_283_to_fp16, dilations = input_933_dilations_0, groups = input_933_groups_0, pad = input_933_pad_0, pad_type = input_933_pad_type_0, strides = input_933_strides_0, weight = const_282_to_fp16, x = input_931_cast_fp16)[name = string("input_935_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_937_cast_fp16 = silu(x = input_935_cast_fp16)[name = string("input_937_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_282_to_fp16 = const()[name = string("const_282_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227400128)))];
+            tensor<fp16, [1024]> const_283_to_fp16 = const()[name = string("const_283_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227418624)))];
+            tensor<fp16, [1, 1024, 64]> input_935_cast_fp16 = conv(bias = const_283_to_fp16, dilations = input_933_dilations_0, groups = input_933_groups_0, pad = input_933_pad_0, pad_type = input_933_pad_type_0, strides = input_933_strides_0, weight = const_282_to_fp16, x = input_931_cast_fp16)[name = string("input_935_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_937_cast_fp16 = silu(x = input_935_cast_fp16)[name = string("input_937_cast_fp16")];
             string x_395_pad_type_0 = const()[name = string("x_395_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_395_strides_0 = const()[name = string("x_395_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_395_pad_0 = const()[name = string("x_395_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_395_dilations_0 = const()[name = string("x_395_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_395_groups_0 = const()[name = string("x_395_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521069]> module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236563520))), indices_nonzero_data = tensor<uint6, [521069]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236694656))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237085568))))[name = string("module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521069]> module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227420736))), indices_nonzero_data = tensor<uint6, [521069]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227551872))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227942784))))[name = string("module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_395_cast_fp16 = conv(dilations = x_395_dilations_0, groups = x_395_groups_0, pad = x_395_pad_0, pad_type = x_395_pad_type_0, strides = x_395_strides_0, weight = module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_937_cast_fp16)[name = string("x_395_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_395_cast_fp16 = conv(dilations = x_395_dilations_0, groups = x_395_groups_0, pad = x_395_pad_0, pad_type = x_395_pad_type_0, strides = x_395_strides_0, weight = module_layers_17_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_937_cast_fp16)[name = string("x_395_cast_fp16")];
             tensor<int32, [3]> input_939_perm_0 = const()[name = string("input_939_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_939_cast_fp16 = transpose(perm = input_939_perm_0, x = x_395_cast_fp16)[name = string("transpose_187")];
-            tensor<fp16, [1, 188, 1024]> input_941_cast_fp16 = add(x = input_923_cast_fp16, y = input_939_cast_fp16)[name = string("input_941_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_939_cast_fp16 = transpose(perm = input_939_perm_0, x = x_395_cast_fp16)[name = string("transpose_187")];
+            tensor<fp16, [1, 64, 1024]> input_941_cast_fp16 = add(x = input_923_cast_fp16, y = input_939_cast_fp16)[name = string("input_941_cast_fp16")];
             tensor<int32, [1]> input_943_axes_0 = const()[name = string("input_943_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_17_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_17_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237102016)))];
-            tensor<fp16, [1024]> module_layers_17_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_17_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237104128)))];
-            tensor<fp16, [1, 188, 1024]> input_943_cast_fp16 = layer_norm(axes = input_943_axes_0, beta = module_layers_17_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_feed_forward2_weight_to_fp16, x = input_941_cast_fp16)[name = string("input_943_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2085405]> module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237106240))), indices_nonzero_data = tensor<uint6, [2085405]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237630592))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239194752))))[name = string("module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_17_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_17_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227959232)))];
+            tensor<fp16, [1024]> module_layers_17_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_17_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227961344)))];
+            tensor<fp16, [1, 64, 1024]> input_943_cast_fp16 = layer_norm(axes = input_943_axes_0, beta = module_layers_17_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_feed_forward2_weight_to_fp16, x = input_941_cast_fp16)[name = string("input_943_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2085405]> module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227963456))), indices_nonzero_data = tensor<uint6, [2085405]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228487808))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230051968))))[name = string("module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_161_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_943_cast_fp16)[name = string("linear_161_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_947_cast_fp16 = silu(x = linear_161_cast_fp16)[name = string("input_947_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2032104]> module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239260352))), indices_nonzero_data = tensor<uint6, [2032104]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239784704))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241308864))))[name = string("module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 4096]> linear_161_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_17_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_943_cast_fp16)[name = string("linear_161_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_947_cast_fp16 = silu(x = linear_161_cast_fp16)[name = string("input_947_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2032104]> module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230117568))), indices_nonzero_data = tensor<uint6, [2032104]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230641920))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232166080))))[name = string("module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_162_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_947_cast_fp16)[name = string("linear_162_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_162_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_17_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_947_cast_fp16)[name = string("linear_162_cast_fp16")];
             fp16 var_3137_to_fp16 = const()[name = string("op_3137_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_3138_cast_fp16 = mul(x = linear_162_cast_fp16, y = var_3137_to_fp16)[name = string("op_3138_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_953_cast_fp16 = add(x = input_941_cast_fp16, y = var_3138_cast_fp16)[name = string("input_953_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> var_3138_cast_fp16 = mul(x = linear_162_cast_fp16, y = var_3137_to_fp16)[name = string("op_3138_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_953_cast_fp16 = add(x = input_941_cast_fp16, y = var_3138_cast_fp16)[name = string("input_953_cast_fp16")];
             tensor<int32, [1]> input_955_axes_0 = const()[name = string("input_955_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_17_norm_out_weight_to_fp16 = const()[name = string("module_layers_17_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241325312)))];
-            tensor<fp16, [1024]> module_layers_17_norm_out_bias_to_fp16 = const()[name = string("module_layers_17_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241327424)))];
-            tensor<fp16, [1, 188, 1024]> input_955_cast_fp16 = layer_norm(axes = input_955_axes_0, beta = module_layers_17_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_out_weight_to_fp16, x = input_953_cast_fp16)[name = string("input_955_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_17_norm_out_weight_to_fp16 = const()[name = string("module_layers_17_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232182528)))];
+            tensor<fp16, [1024]> module_layers_17_norm_out_bias_to_fp16 = const()[name = string("module_layers_17_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232184640)))];
+            tensor<fp16, [1, 64, 1024]> input_955_cast_fp16 = layer_norm(axes = input_955_axes_0, beta = module_layers_17_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_out_weight_to_fp16, x = input_953_cast_fp16)[name = string("input_955_cast_fp16")];
             tensor<int32, [1]> input_957_axes_0 = const()[name = string("input_957_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_18_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_18_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241329536)))];
-            tensor<fp16, [1024]> module_layers_18_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_18_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241331648)))];
-            tensor<fp16, [1, 188, 1024]> input_957_cast_fp16 = layer_norm(axes = input_957_axes_0, beta = module_layers_18_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_feed_forward1_weight_to_fp16, x = input_955_cast_fp16)[name = string("input_957_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2077029]> module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241333760))), indices_nonzero_data = tensor<uint6, [2077029]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241858112))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243416000))))[name = string("module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_18_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_18_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232186752)))];
+            tensor<fp16, [1024]> module_layers_18_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_18_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232188864)))];
+            tensor<fp16, [1, 64, 1024]> input_957_cast_fp16 = layer_norm(axes = input_957_axes_0, beta = module_layers_18_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_feed_forward1_weight_to_fp16, x = input_955_cast_fp16)[name = string("input_957_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2077029]> module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232190976))), indices_nonzero_data = tensor<uint6, [2077029]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232715328))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234273216))))[name = string("module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_163_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_957_cast_fp16)[name = string("linear_163_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_961_cast_fp16 = silu(x = linear_163_cast_fp16)[name = string("input_961_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2028343]> module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243481600))), indices_nonzero_data = tensor<uint6, [2028343]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244005952))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245527296))))[name = string("module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 4096]> linear_163_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_18_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_957_cast_fp16)[name = string("linear_163_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_961_cast_fp16 = silu(x = linear_163_cast_fp16)[name = string("input_961_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2028343]> module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234338816))), indices_nonzero_data = tensor<uint6, [2028343]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234863168))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236384512))))[name = string("module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_164_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_961_cast_fp16)[name = string("linear_164_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_164_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_961_cast_fp16)[name = string("linear_164_cast_fp16")];
             fp16 var_3166_to_fp16 = const()[name = string("op_3166_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_3167_cast_fp16 = mul(x = linear_164_cast_fp16, y = var_3166_to_fp16)[name = string("op_3167_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_967_cast_fp16 = add(x = input_955_cast_fp16, y = var_3167_cast_fp16)[name = string("input_967_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> var_3167_cast_fp16 = mul(x = linear_164_cast_fp16, y = var_3166_to_fp16)[name = string("op_3167_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_967_cast_fp16 = add(x = input_955_cast_fp16, y = var_3167_cast_fp16)[name = string("input_967_cast_fp16")];
             tensor<int32, [1]> query_37_axes_0 = const()[name = string("query_37_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_18_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_18_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245543744)))];
-            tensor<fp16, [1024]> module_layers_18_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_18_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245545856)))];
-            tensor<fp16, [1, 188, 1024]> query_37_cast_fp16 = layer_norm(axes = query_37_axes_0, beta = module_layers_18_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_self_att_weight_to_fp16, x = input_967_cast_fp16)[name = string("query_37_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525392]> module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245547968))), indices_nonzero_data = tensor<uint6, [525392]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245679104))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246073216))))[name = string("module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_18_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_18_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236400960)))];
+            tensor<fp16, [1024]> module_layers_18_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_18_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236403072)))];
+            tensor<fp16, [1, 64, 1024]> query_37_cast_fp16 = layer_norm(axes = query_37_axes_0, beta = module_layers_18_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_self_att_weight_to_fp16, x = input_967_cast_fp16)[name = string("query_37_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525392]> module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236405184))), indices_nonzero_data = tensor<uint6, [525392]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236536320))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236930432))))[name = string("module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_165_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_37_cast_fp16)[name = string("linear_165_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_165_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_37_cast_fp16)[name = string("linear_165_cast_fp16")];
             tensor<int32, [4]> var_3183 = const()[name = string("op_3183"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_109_cast_fp16 = reshape(shape = var_3183, x = linear_165_cast_fp16)[name = string("q_109_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524473]> module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246089664))), indices_nonzero_data = tensor<uint6, [524473]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246220800))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246614272))))[name = string("module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> q_109_cast_fp16 = reshape(shape = var_3183, x = linear_165_cast_fp16)[name = string("q_109_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524473]> module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236946880))), indices_nonzero_data = tensor<uint6, [524473]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237078016))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237471488))))[name = string("module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_166_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_37_cast_fp16)[name = string("linear_166_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_166_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_37_cast_fp16)[name = string("linear_166_cast_fp16")];
             tensor<int32, [4]> var_3187 = const()[name = string("op_3187"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_73_cast_fp16 = reshape(shape = var_3187, x = linear_166_cast_fp16)[name = string("k_73_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523564]> module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246630720))), indices_nonzero_data = tensor<uint6, [523564]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246761856))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247154624))))[name = string("module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> k_73_cast_fp16 = reshape(shape = var_3187, x = linear_166_cast_fp16)[name = string("k_73_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523564]> module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237487936))), indices_nonzero_data = tensor<uint6, [523564]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237619072))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238011840))))[name = string("module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_167_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_37_cast_fp16)[name = string("linear_167_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_167_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_37_cast_fp16)[name = string("linear_167_cast_fp16")];
             tensor<int32, [4]> var_3191 = const()[name = string("op_3191"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_37_cast_fp16 = reshape(shape = var_3191, x = linear_167_cast_fp16)[name = string("v_37_cast_fp16")];
+            tensor<fp16, [1, 64, 8, 128]> v_37_cast_fp16 = reshape(shape = var_3191, x = linear_167_cast_fp16)[name = string("v_37_cast_fp16")];
             tensor<int32, [4]> value_39_perm_0 = const()[name = string("value_39_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_18_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_18_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247171072)))];
-            tensor<fp16, [1, 188, 8, 128]> var_3203_cast_fp16 = add(x = q_109_cast_fp16, y = module_layers_18_self_attn_pos_bias_u_to_fp16)[name = string("op_3203_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_18_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_18_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247173184)))];
-            tensor<fp16, [1, 188, 8, 128]> var_3205_cast_fp16 = add(x = q_109_cast_fp16, y = module_layers_18_self_attn_pos_bias_v_to_fp16)[name = string("op_3205_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_18_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_18_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238028288)))];
+            tensor<fp16, [1, 64, 8, 128]> var_3203_cast_fp16 = add(x = q_109_cast_fp16, y = module_layers_18_self_attn_pos_bias_u_to_fp16)[name = string("op_3203_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_18_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_18_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238030400)))];
+            tensor<fp16, [1, 64, 8, 128]> var_3205_cast_fp16 = add(x = q_109_cast_fp16, y = module_layers_18_self_attn_pos_bias_v_to_fp16)[name = string("op_3205_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_37_perm_0 = const()[name = string("q_with_bias_v_37_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_403_transpose_x_0 = const()[name = string("x_403_transpose_x_0"), val = bool(false)];
             bool x_403_transpose_y_0 = const()[name = string("x_403_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_3207_to_fp16 = const()[name = string("op_3207_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247175296)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_37_cast_fp16 = transpose(perm = q_with_bias_v_37_perm_0, x = var_3205_cast_fp16)[name = string("transpose_186")];
-            tensor<fp16, [1, 8, 188, 375]> x_403_cast_fp16 = matmul(transpose_x = x_403_transpose_x_0, transpose_y = x_403_transpose_y_0, x = q_with_bias_v_37_cast_fp16, y = var_3207_to_fp16)[name = string("x_403_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 127]> var_3207_to_fp16 = const()[name = string("op_3207_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238032512)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_37_cast_fp16 = transpose(perm = q_with_bias_v_37_perm_0, x = var_3205_cast_fp16)[name = string("transpose_186")];
+            tensor<fp16, [1, 8, 64, 127]> x_403_cast_fp16 = matmul(transpose_x = x_403_transpose_x_0, transpose_y = x_403_transpose_y_0, x = q_with_bias_v_37_cast_fp16, y = var_3207_to_fp16)[name = string("x_403_cast_fp16")];
             tensor<int32, [8]> x_405_pad_0 = const()[name = string("x_405_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_405_mode_0 = const()[name = string("x_405_mode_0"), val = string("constant")];
             fp16 const_194_to_fp16 = const()[name = string("const_194_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_405_cast_fp16 = pad(constant_val = const_194_to_fp16, mode = x_405_mode_0, pad = x_405_pad_0, x = x_403_cast_fp16)[name = string("x_405_cast_fp16")];
-            tensor<int32, [4]> var_3215 = const()[name = string("op_3215"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_407_cast_fp16 = reshape(shape = var_3215, x = x_405_cast_fp16)[name = string("x_407_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 128]> x_405_cast_fp16 = pad(constant_val = const_194_to_fp16, mode = x_405_mode_0, pad = x_405_pad_0, x = x_403_cast_fp16)[name = string("x_405_cast_fp16")];
+            tensor<int32, [4]> var_3215 = const()[name = string("op_3215"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_407_cast_fp16 = reshape(shape = var_3215, x = x_405_cast_fp16)[name = string("x_407_cast_fp16")];
             tensor<int32, [4]> var_3219_begin_0 = const()[name = string("op_3219_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_3219_end_0 = const()[name = string("op_3219_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_3219_end_0 = const()[name = string("op_3219_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
             tensor<bool, [4]> var_3219_end_mask_0 = const()[name = string("op_3219_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_3219_cast_fp16 = slice_by_index(begin = var_3219_begin_0, end = var_3219_end_0, end_mask = var_3219_end_mask_0, x = x_407_cast_fp16)[name = string("op_3219_cast_fp16")];
-            tensor<int32, [4]> var_3220 = const()[name = string("op_3220"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_73_cast_fp16 = reshape(shape = var_3220, x = var_3219_cast_fp16)[name = string("matrix_bd_73_cast_fp16")];
+            tensor<fp16, [1, 8, 127, 64]> var_3219_cast_fp16 = slice_by_index(begin = var_3219_begin_0, end = var_3219_end_0, end_mask = var_3219_end_mask_0, x = x_407_cast_fp16)[name = string("op_3219_cast_fp16")];
+            tensor<int32, [4]> var_3220 = const()[name = string("op_3220"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_73_cast_fp16 = reshape(shape = var_3220, x = var_3219_cast_fp16)[name = string("matrix_bd_73_cast_fp16")];
             bool matrix_ac_37_transpose_x_0 = const()[name = string("matrix_ac_37_transpose_x_0"), val = bool(false)];
             bool matrix_ac_37_transpose_y_0 = const()[name = string("matrix_ac_37_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_132_perm_0 = const()[name = string("transpose_132_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_133_perm_0 = const()[name = string("transpose_133_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_133 = transpose(perm = transpose_133_perm_0, x = k_73_cast_fp16)[name = string("transpose_184")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_132 = transpose(perm = transpose_132_perm_0, x = var_3203_cast_fp16)[name = string("transpose_185")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_37_cast_fp16 = matmul(transpose_x = matrix_ac_37_transpose_x_0, transpose_y = matrix_ac_37_transpose_y_0, x = transpose_132, y = transpose_133)[name = string("matrix_ac_37_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 64]> transpose_133 = transpose(perm = transpose_133_perm_0, x = k_73_cast_fp16)[name = string("transpose_184")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_132 = transpose(perm = transpose_132_perm_0, x = var_3203_cast_fp16)[name = string("transpose_185")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_37_cast_fp16 = matmul(transpose_x = matrix_ac_37_transpose_x_0, transpose_y = matrix_ac_37_transpose_y_0, x = transpose_132, y = transpose_133)[name = string("matrix_ac_37_cast_fp16")];
             tensor<int32, [4]> matrix_bd_75_begin_0 = const()[name = string("matrix_bd_75_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_75_end_0 = const()[name = string("matrix_bd_75_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_75_end_0 = const()[name = string("matrix_bd_75_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
             tensor<bool, [4]> matrix_bd_75_end_mask_0 = const()[name = string("matrix_bd_75_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_75_cast_fp16 = slice_by_index(begin = matrix_bd_75_begin_0, end = matrix_bd_75_end_0, end_mask = matrix_bd_75_end_mask_0, x = matrix_bd_73_cast_fp16)[name = string("matrix_bd_75_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_3229_cast_fp16 = add(x = matrix_ac_37_cast_fp16, y = matrix_bd_75_cast_fp16)[name = string("op_3229_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_75_cast_fp16 = slice_by_index(begin = matrix_bd_75_begin_0, end = matrix_bd_75_end_0, end_mask = matrix_bd_75_end_mask_0, x = matrix_bd_73_cast_fp16)[name = string("matrix_bd_75_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_3229_cast_fp16 = add(x = matrix_ac_37_cast_fp16, y = matrix_bd_75_cast_fp16)[name = string("op_3229_cast_fp16")];
             fp16 _inversed_scores_73_y_0_to_fp16 = const()[name = string("_inversed_scores_73_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_73_cast_fp16 = mul(x = var_3229_cast_fp16, y = _inversed_scores_73_y_0_to_fp16)[name = string("_inversed_scores_73_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_75_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_73_cast_fp16, cond = mask_3)[name = string("scores_75_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_3235_cast_fp16 = softmax(axis = var_30, x = scores_75_cast_fp16)[name = string("op_3235_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_969_cast_fp16 = select(a = var_11_to_fp16, b = var_3235_cast_fp16, cond = mask_3)[name = string("input_969_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_73_cast_fp16 = mul(x = var_3229_cast_fp16, y = _inversed_scores_73_y_0_to_fp16)[name = string("_inversed_scores_73_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_75_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_73_cast_fp16, cond = mask_3)[name = string("scores_75_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_3235_cast_fp16 = softmax(axis = var_30, x = scores_75_cast_fp16)[name = string("op_3235_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_969_cast_fp16 = select(a = var_11_to_fp16, b = var_3235_cast_fp16, cond = mask_3)[name = string("input_969_cast_fp16")];
             bool x_409_transpose_x_0 = const()[name = string("x_409_transpose_x_0"), val = bool(false)];
             bool x_409_transpose_y_0 = const()[name = string("x_409_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_39_cast_fp16 = transpose(perm = value_39_perm_0, x = v_37_cast_fp16)[name = string("transpose_183")];
-            tensor<fp16, [1, 8, 188, 128]> x_409_cast_fp16 = matmul(transpose_x = x_409_transpose_x_0, transpose_y = x_409_transpose_y_0, x = input_969_cast_fp16, y = value_39_cast_fp16)[name = string("x_409_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 128]> value_39_cast_fp16 = transpose(perm = value_39_perm_0, x = v_37_cast_fp16)[name = string("transpose_183")];
+            tensor<fp16, [1, 8, 64, 128]> x_409_cast_fp16 = matmul(transpose_x = x_409_transpose_x_0, transpose_y = x_409_transpose_y_0, x = input_969_cast_fp16, y = value_39_cast_fp16)[name = string("x_409_cast_fp16")];
             tensor<int32, [4]> var_3239_perm_0 = const()[name = string("op_3239_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_3240 = const()[name = string("op_3240"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_3239_cast_fp16 = transpose(perm = var_3239_perm_0, x = x_409_cast_fp16)[name = string("transpose_182")];
-            tensor<fp16, [1, 188, 1024]> input_971_cast_fp16 = reshape(shape = var_3240, x = var_3239_cast_fp16)[name = string("input_971_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520692]> module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247943360))), indices_nonzero_data = tensor<uint6, [520692]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248074496))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248465088))))[name = string("module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> var_3239_cast_fp16 = transpose(perm = var_3239_perm_0, x = x_409_cast_fp16)[name = string("transpose_182")];
+            tensor<fp16, [1, 64, 1024]> input_971_cast_fp16 = reshape(shape = var_3240, x = var_3239_cast_fp16)[name = string("input_971_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [520692]> module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238292672))), indices_nonzero_data = tensor<uint6, [520692]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238423808))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238814400))))[name = string("module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_169_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_971_cast_fp16)[name = string("linear_169_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_975_cast_fp16 = add(x = input_967_cast_fp16, y = linear_169_cast_fp16)[name = string("input_975_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_169_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_971_cast_fp16)[name = string("linear_169_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_975_cast_fp16 = add(x = input_967_cast_fp16, y = linear_169_cast_fp16)[name = string("input_975_cast_fp16")];
             tensor<int32, [1]> x_413_axes_0 = const()[name = string("x_413_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_18_norm_conv_weight_to_fp16 = const()[name = string("module_layers_18_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248481536)))];
-            tensor<fp16, [1024]> module_layers_18_norm_conv_bias_to_fp16 = const()[name = string("module_layers_18_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248483648)))];
-            tensor<fp16, [1, 188, 1024]> x_413_cast_fp16 = layer_norm(axes = x_413_axes_0, beta = module_layers_18_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_conv_weight_to_fp16, x = input_975_cast_fp16)[name = string("x_413_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_18_norm_conv_weight_to_fp16 = const()[name = string("module_layers_18_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238830848)))];
+            tensor<fp16, [1024]> module_layers_18_norm_conv_bias_to_fp16 = const()[name = string("module_layers_18_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238832960)))];
+            tensor<fp16, [1, 64, 1024]> x_413_cast_fp16 = layer_norm(axes = x_413_axes_0, beta = module_layers_18_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_conv_weight_to_fp16, x = input_975_cast_fp16)[name = string("x_413_cast_fp16")];
             tensor<int32, [3]> input_977_perm_0 = const()[name = string("input_977_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_979_pad_type_0 = const()[name = string("input_979_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_979_strides_0 = const()[name = string("input_979_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_979_pad_0 = const()[name = string("input_979_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_979_dilations_0 = const()[name = string("input_979_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_979_groups_0 = const()[name = string("input_979_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047234]> module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248485760))), indices_nonzero_data = tensor<uint6, [1047234]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248747968))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249533504))))[name = string("module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047234]> module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238835072))), indices_nonzero_data = tensor<uint6, [1047234]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239097280))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239882816))))[name = string("module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_977_cast_fp16 = transpose(perm = input_977_perm_0, x = x_413_cast_fp16)[name = string("transpose_181")];
-            tensor<fp16, [1, 2048, 188]> input_979_cast_fp16 = conv(dilations = input_979_dilations_0, groups = input_979_groups_0, pad = input_979_pad_0, pad_type = input_979_pad_type_0, strides = input_979_strides_0, weight = module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_977_cast_fp16)[name = string("input_979_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_977_cast_fp16 = transpose(perm = input_977_perm_0, x = x_413_cast_fp16)[name = string("transpose_181")];
+            tensor<fp16, [1, 2048, 64]> input_979_cast_fp16 = conv(dilations = input_979_dilations_0, groups = input_979_groups_0, pad = input_979_pad_0, pad_type = input_979_pad_type_0, strides = input_979_strides_0, weight = module_layers_18_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_977_cast_fp16)[name = string("input_979_cast_fp16")];
             int32 x_415_split_num_splits_0 = const()[name = string("x_415_split_num_splits_0"), val = int32(2)];
             int32 x_415_split_axis_0 = const()[name = string("x_415_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_415_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_415_split_cast_fp16_1 = split(axis = x_415_split_axis_0, num_splits = x_415_split_num_splits_0, x = input_979_cast_fp16)[name = string("x_415_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_415_split_1_sigmoid_cast_fp16 = sigmoid(x = x_415_split_cast_fp16_1)[name = string("x_415_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_415_cast_fp16 = mul(x = x_415_split_cast_fp16_0, y = x_415_split_1_sigmoid_cast_fp16)[name = string("x_415_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_981_cast_fp16 = select(a = var_11_to_fp16, b = x_415_cast_fp16, cond = var_328)[name = string("input_981_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_415_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_415_split_cast_fp16_1 = split(axis = x_415_split_axis_0, num_splits = x_415_split_num_splits_0, x = input_979_cast_fp16)[name = string("x_415_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_415_split_1_sigmoid_cast_fp16 = sigmoid(x = x_415_split_cast_fp16_1)[name = string("x_415_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_415_cast_fp16 = mul(x = x_415_split_cast_fp16_0, y = x_415_split_1_sigmoid_cast_fp16)[name = string("x_415_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_981_cast_fp16 = select(a = var_11_to_fp16, b = x_415_cast_fp16, cond = var_328)[name = string("input_981_cast_fp16")];
             tensor<int32, [6]> input_983_pad_0 = const()[name = string("input_983_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_983_mode_0 = const()[name = string("input_983_mode_0"), val = string("constant")];
             fp16 const_197_to_fp16 = const()[name = string("const_197_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_983_cast_fp16 = pad(constant_val = const_197_to_fp16, mode = input_983_mode_0, pad = input_983_pad_0, x = input_981_cast_fp16)[name = string("input_983_cast_fp16")];
+            tensor<fp16, [1, 1024, 72]> input_983_cast_fp16 = pad(constant_val = const_197_to_fp16, mode = input_983_mode_0, pad = input_983_pad_0, x = input_981_cast_fp16)[name = string("input_983_cast_fp16")];
             string input_985_pad_type_0 = const()[name = string("input_985_pad_type_0"), val = string("valid")];
             int32 input_985_groups_0 = const()[name = string("input_985_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_985_strides_0 = const()[name = string("input_985_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_985_pad_0 = const()[name = string("input_985_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_985_dilations_0 = const()[name = string("input_985_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_284_to_fp16 = const()[name = string("const_284_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249566336)))];
-            tensor<fp16, [1024]> const_285_to_fp16 = const()[name = string("const_285_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249584832)))];
-            tensor<fp16, [1, 1024, 188]> input_987_cast_fp16 = conv(bias = const_285_to_fp16, dilations = input_985_dilations_0, groups = input_985_groups_0, pad = input_985_pad_0, pad_type = input_985_pad_type_0, strides = input_985_strides_0, weight = const_284_to_fp16, x = input_983_cast_fp16)[name = string("input_987_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_989_cast_fp16 = silu(x = input_987_cast_fp16)[name = string("input_989_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_284_to_fp16 = const()[name = string("const_284_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239915648)))];
+            tensor<fp16, [1024]> const_285_to_fp16 = const()[name = string("const_285_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239934144)))];
+            tensor<fp16, [1, 1024, 64]> input_987_cast_fp16 = conv(bias = const_285_to_fp16, dilations = input_985_dilations_0, groups = input_985_groups_0, pad = input_985_pad_0, pad_type = input_985_pad_type_0, strides = input_985_strides_0, weight = const_284_to_fp16, x = input_983_cast_fp16)[name = string("input_987_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_989_cast_fp16 = silu(x = input_987_cast_fp16)[name = string("input_989_cast_fp16")];
             string x_417_pad_type_0 = const()[name = string("x_417_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_417_strides_0 = const()[name = string("x_417_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_417_pad_0 = const()[name = string("x_417_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_417_dilations_0 = const()[name = string("x_417_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_417_groups_0 = const()[name = string("x_417_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521466]> module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249586944))), indices_nonzero_data = tensor<uint6, [521466]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249718080))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250109248))))[name = string("module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [521466]> module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239936256))), indices_nonzero_data = tensor<uint6, [521466]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240067392))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240458560))))[name = string("module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_417_cast_fp16 = conv(dilations = x_417_dilations_0, groups = x_417_groups_0, pad = x_417_pad_0, pad_type = x_417_pad_type_0, strides = x_417_strides_0, weight = module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_989_cast_fp16)[name = string("x_417_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_417_cast_fp16 = conv(dilations = x_417_dilations_0, groups = x_417_groups_0, pad = x_417_pad_0, pad_type = x_417_pad_type_0, strides = x_417_strides_0, weight = module_layers_18_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_989_cast_fp16)[name = string("x_417_cast_fp16")];
             tensor<int32, [3]> input_991_perm_0 = const()[name = string("input_991_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_991_cast_fp16 = transpose(perm = input_991_perm_0, x = x_417_cast_fp16)[name = string("transpose_180")];
-            tensor<fp16, [1, 188, 1024]> input_993_cast_fp16 = add(x = input_975_cast_fp16, y = input_991_cast_fp16)[name = string("input_993_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_991_cast_fp16 = transpose(perm = input_991_perm_0, x = x_417_cast_fp16)[name = string("transpose_180")];
+            tensor<fp16, [1, 64, 1024]> input_993_cast_fp16 = add(x = input_975_cast_fp16, y = input_991_cast_fp16)[name = string("input_993_cast_fp16")];
             tensor<int32, [1]> input_995_axes_0 = const()[name = string("input_995_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_18_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_18_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250125696)))];
-            tensor<fp16, [1024]> module_layers_18_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_18_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250127808)))];
-            tensor<fp16, [1, 188, 1024]> input_995_cast_fp16 = layer_norm(axes = input_995_axes_0, beta = module_layers_18_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_feed_forward2_weight_to_fp16, x = input_993_cast_fp16)[name = string("input_995_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2081130]> module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250129920))), indices_nonzero_data = tensor<uint6, [2081130]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250654272))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252215232))))[name = string("module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_18_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_18_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240475008)))];
+            tensor<fp16, [1024]> module_layers_18_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_18_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240477120)))];
+            tensor<fp16, [1, 64, 1024]> input_995_cast_fp16 = layer_norm(axes = input_995_axes_0, beta = module_layers_18_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_feed_forward2_weight_to_fp16, x = input_993_cast_fp16)[name = string("input_995_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2081130]> module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240479232))), indices_nonzero_data = tensor<uint6, [2081130]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241003584))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242564544))))[name = string("module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_170_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_995_cast_fp16)[name = string("linear_170_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_999_cast_fp16 = silu(x = linear_170_cast_fp16)[name = string("input_999_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2068517]> module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252280832))), indices_nonzero_data = tensor<uint6, [2068517]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252805184))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254356672))))[name = string("module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 4096]> linear_170_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_18_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_995_cast_fp16)[name = string("linear_170_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_999_cast_fp16 = silu(x = linear_170_cast_fp16)[name = string("input_999_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2068517]> module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242630144))), indices_nonzero_data = tensor<uint6, [2068517]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243154496))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244705984))))[name = string("module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_171_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_999_cast_fp16)[name = string("linear_171_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_171_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_18_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_999_cast_fp16)[name = string("linear_171_cast_fp16")];
             fp16 var_3300_to_fp16 = const()[name = string("op_3300_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_3301_cast_fp16 = mul(x = linear_171_cast_fp16, y = var_3300_to_fp16)[name = string("op_3301_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_1005_cast_fp16 = add(x = input_993_cast_fp16, y = var_3301_cast_fp16)[name = string("input_1005_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> var_3301_cast_fp16 = mul(x = linear_171_cast_fp16, y = var_3300_to_fp16)[name = string("op_3301_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_1005_cast_fp16 = add(x = input_993_cast_fp16, y = var_3301_cast_fp16)[name = string("input_1005_cast_fp16")];
             tensor<int32, [1]> input_1007_axes_0 = const()[name = string("input_1007_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_18_norm_out_weight_to_fp16 = const()[name = string("module_layers_18_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254373120)))];
-            tensor<fp16, [1024]> module_layers_18_norm_out_bias_to_fp16 = const()[name = string("module_layers_18_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254375232)))];
-            tensor<fp16, [1, 188, 1024]> input_1007_cast_fp16 = layer_norm(axes = input_1007_axes_0, beta = module_layers_18_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_out_weight_to_fp16, x = input_1005_cast_fp16)[name = string("input_1007_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_18_norm_out_weight_to_fp16 = const()[name = string("module_layers_18_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244722432)))];
+            tensor<fp16, [1024]> module_layers_18_norm_out_bias_to_fp16 = const()[name = string("module_layers_18_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244724544)))];
+            tensor<fp16, [1, 64, 1024]> input_1007_cast_fp16 = layer_norm(axes = input_1007_axes_0, beta = module_layers_18_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_out_weight_to_fp16, x = input_1005_cast_fp16)[name = string("input_1007_cast_fp16")];
             tensor<int32, [1]> input_1009_axes_0 = const()[name = string("input_1009_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_19_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_19_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254377344)))];
-            tensor<fp16, [1024]> module_layers_19_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_19_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254379456)))];
-            tensor<fp16, [1, 188, 1024]> input_1009_cast_fp16 = layer_norm(axes = input_1009_axes_0, beta = module_layers_19_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_feed_forward1_weight_to_fp16, x = input_1007_cast_fp16)[name = string("input_1009_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2077149]> module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254381568))), indices_nonzero_data = tensor<uint6, [2077149]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254905920))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256463872))))[name = string("module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_19_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_19_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244726656)))];
+            tensor<fp16, [1024]> module_layers_19_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_19_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244728768)))];
+            tensor<fp16, [1, 64, 1024]> input_1009_cast_fp16 = layer_norm(axes = input_1009_axes_0, beta = module_layers_19_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_feed_forward1_weight_to_fp16, x = input_1007_cast_fp16)[name = string("input_1009_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2077149]> module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244730880))), indices_nonzero_data = tensor<uint6, [2077149]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245255232))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246813184))))[name = string("module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_172_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_1009_cast_fp16)[name = string("linear_172_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_1013_cast_fp16 = silu(x = linear_172_cast_fp16)[name = string("input_1013_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2050240]> module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256529472))), indices_nonzero_data = tensor<uint6, [2050240]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257053824))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258591616))))[name = string("module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 4096]> linear_172_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_19_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_1009_cast_fp16)[name = string("linear_172_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_1013_cast_fp16 = silu(x = linear_172_cast_fp16)[name = string("input_1013_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2050240]> module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246878784))), indices_nonzero_data = tensor<uint6, [2050240]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247403136))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248940928))))[name = string("module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_173_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_1013_cast_fp16)[name = string("linear_173_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_173_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_1013_cast_fp16)[name = string("linear_173_cast_fp16")];
             fp16 var_3329_to_fp16 = const()[name = string("op_3329_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_3330_cast_fp16 = mul(x = linear_173_cast_fp16, y = var_3329_to_fp16)[name = string("op_3330_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_1019_cast_fp16 = add(x = input_1007_cast_fp16, y = var_3330_cast_fp16)[name = string("input_1019_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> var_3330_cast_fp16 = mul(x = linear_173_cast_fp16, y = var_3329_to_fp16)[name = string("op_3330_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_1019_cast_fp16 = add(x = input_1007_cast_fp16, y = var_3330_cast_fp16)[name = string("input_1019_cast_fp16")];
             tensor<int32, [1]> query_39_axes_0 = const()[name = string("query_39_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_19_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_19_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258608064)))];
-            tensor<fp16, [1024]> module_layers_19_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_19_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258610176)))];
-            tensor<fp16, [1, 188, 1024]> query_39_cast_fp16 = layer_norm(axes = query_39_axes_0, beta = module_layers_19_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_self_att_weight_to_fp16, x = input_1019_cast_fp16)[name = string("query_39_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524663]> module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258612288))), indices_nonzero_data = tensor<uint6, [524663]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258743424))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259137024))))[name = string("module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_19_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_19_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248957376)))];
+            tensor<fp16, [1024]> module_layers_19_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_19_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248959488)))];
+            tensor<fp16, [1, 64, 1024]> query_39_cast_fp16 = layer_norm(axes = query_39_axes_0, beta = module_layers_19_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_self_att_weight_to_fp16, x = input_1019_cast_fp16)[name = string("query_39_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524663]> module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248961600))), indices_nonzero_data = tensor<uint6, [524663]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249092736))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249486336))))[name = string("module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_174_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_39_cast_fp16)[name = string("linear_174_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_174_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_39_cast_fp16)[name = string("linear_174_cast_fp16")];
             tensor<int32, [4]> var_3346 = const()[name = string("op_3346"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_115_cast_fp16 = reshape(shape = var_3346, x = linear_174_cast_fp16)[name = string("q_115_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525503]> module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259153472))), indices_nonzero_data = tensor<uint6, [525503]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259284608))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259678848))))[name = string("module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> q_115_cast_fp16 = reshape(shape = var_3346, x = linear_174_cast_fp16)[name = string("q_115_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525503]> module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249502784))), indices_nonzero_data = tensor<uint6, [525503]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249633920))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250028160))))[name = string("module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_175_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_39_cast_fp16)[name = string("linear_175_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_175_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_39_cast_fp16)[name = string("linear_175_cast_fp16")];
             tensor<int32, [4]> var_3350 = const()[name = string("op_3350"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_77_cast_fp16 = reshape(shape = var_3350, x = linear_175_cast_fp16)[name = string("k_77_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523873]> module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259695296))), indices_nonzero_data = tensor<uint6, [523873]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259826432))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260219456))))[name = string("module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> k_77_cast_fp16 = reshape(shape = var_3350, x = linear_175_cast_fp16)[name = string("k_77_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523873]> module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250044608))), indices_nonzero_data = tensor<uint6, [523873]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250175744))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250568768))))[name = string("module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_176_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_39_cast_fp16)[name = string("linear_176_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_176_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_39_cast_fp16)[name = string("linear_176_cast_fp16")];
             tensor<int32, [4]> var_3354 = const()[name = string("op_3354"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_39_cast_fp16 = reshape(shape = var_3354, x = linear_176_cast_fp16)[name = string("v_39_cast_fp16")];
+            tensor<fp16, [1, 64, 8, 128]> v_39_cast_fp16 = reshape(shape = var_3354, x = linear_176_cast_fp16)[name = string("v_39_cast_fp16")];
             tensor<int32, [4]> value_41_perm_0 = const()[name = string("value_41_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_19_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_19_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260235904)))];
-            tensor<fp16, [1, 188, 8, 128]> var_3366_cast_fp16 = add(x = q_115_cast_fp16, y = module_layers_19_self_attn_pos_bias_u_to_fp16)[name = string("op_3366_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_19_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_19_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260238016)))];
-            tensor<fp16, [1, 188, 8, 128]> var_3368_cast_fp16 = add(x = q_115_cast_fp16, y = module_layers_19_self_attn_pos_bias_v_to_fp16)[name = string("op_3368_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_19_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_19_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250585216)))];
+            tensor<fp16, [1, 64, 8, 128]> var_3366_cast_fp16 = add(x = q_115_cast_fp16, y = module_layers_19_self_attn_pos_bias_u_to_fp16)[name = string("op_3366_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_19_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_19_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250587328)))];
+            tensor<fp16, [1, 64, 8, 128]> var_3368_cast_fp16 = add(x = q_115_cast_fp16, y = module_layers_19_self_attn_pos_bias_v_to_fp16)[name = string("op_3368_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_39_perm_0 = const()[name = string("q_with_bias_v_39_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_425_transpose_x_0 = const()[name = string("x_425_transpose_x_0"), val = bool(false)];
             bool x_425_transpose_y_0 = const()[name = string("x_425_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_3370_to_fp16 = const()[name = string("op_3370_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260240128)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_39_cast_fp16 = transpose(perm = q_with_bias_v_39_perm_0, x = var_3368_cast_fp16)[name = string("transpose_179")];
-            tensor<fp16, [1, 8, 188, 375]> x_425_cast_fp16 = matmul(transpose_x = x_425_transpose_x_0, transpose_y = x_425_transpose_y_0, x = q_with_bias_v_39_cast_fp16, y = var_3370_to_fp16)[name = string("x_425_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 127]> var_3370_to_fp16 = const()[name = string("op_3370_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250589440)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_39_cast_fp16 = transpose(perm = q_with_bias_v_39_perm_0, x = var_3368_cast_fp16)[name = string("transpose_179")];
+            tensor<fp16, [1, 8, 64, 127]> x_425_cast_fp16 = matmul(transpose_x = x_425_transpose_x_0, transpose_y = x_425_transpose_y_0, x = q_with_bias_v_39_cast_fp16, y = var_3370_to_fp16)[name = string("x_425_cast_fp16")];
             tensor<int32, [8]> x_427_pad_0 = const()[name = string("x_427_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_427_mode_0 = const()[name = string("x_427_mode_0"), val = string("constant")];
             fp16 const_204_to_fp16 = const()[name = string("const_204_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_427_cast_fp16 = pad(constant_val = const_204_to_fp16, mode = x_427_mode_0, pad = x_427_pad_0, x = x_425_cast_fp16)[name = string("x_427_cast_fp16")];
-            tensor<int32, [4]> var_3378 = const()[name = string("op_3378"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_429_cast_fp16 = reshape(shape = var_3378, x = x_427_cast_fp16)[name = string("x_429_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 128]> x_427_cast_fp16 = pad(constant_val = const_204_to_fp16, mode = x_427_mode_0, pad = x_427_pad_0, x = x_425_cast_fp16)[name = string("x_427_cast_fp16")];
+            tensor<int32, [4]> var_3378 = const()[name = string("op_3378"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_429_cast_fp16 = reshape(shape = var_3378, x = x_427_cast_fp16)[name = string("x_429_cast_fp16")];
             tensor<int32, [4]> var_3382_begin_0 = const()[name = string("op_3382_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_3382_end_0 = const()[name = string("op_3382_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_3382_end_0 = const()[name = string("op_3382_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
             tensor<bool, [4]> var_3382_end_mask_0 = const()[name = string("op_3382_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_3382_cast_fp16 = slice_by_index(begin = var_3382_begin_0, end = var_3382_end_0, end_mask = var_3382_end_mask_0, x = x_429_cast_fp16)[name = string("op_3382_cast_fp16")];
-            tensor<int32, [4]> var_3383 = const()[name = string("op_3383"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_77_cast_fp16 = reshape(shape = var_3383, x = var_3382_cast_fp16)[name = string("matrix_bd_77_cast_fp16")];
+            tensor<fp16, [1, 8, 127, 64]> var_3382_cast_fp16 = slice_by_index(begin = var_3382_begin_0, end = var_3382_end_0, end_mask = var_3382_end_mask_0, x = x_429_cast_fp16)[name = string("op_3382_cast_fp16")];
+            tensor<int32, [4]> var_3383 = const()[name = string("op_3383"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_77_cast_fp16 = reshape(shape = var_3383, x = var_3382_cast_fp16)[name = string("matrix_bd_77_cast_fp16")];
             bool matrix_ac_39_transpose_x_0 = const()[name = string("matrix_ac_39_transpose_x_0"), val = bool(false)];
             bool matrix_ac_39_transpose_y_0 = const()[name = string("matrix_ac_39_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_134_perm_0 = const()[name = string("transpose_134_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_135_perm_0 = const()[name = string("transpose_135_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_135 = transpose(perm = transpose_135_perm_0, x = k_77_cast_fp16)[name = string("transpose_177")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_134 = transpose(perm = transpose_134_perm_0, x = var_3366_cast_fp16)[name = string("transpose_178")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_39_cast_fp16 = matmul(transpose_x = matrix_ac_39_transpose_x_0, transpose_y = matrix_ac_39_transpose_y_0, x = transpose_134, y = transpose_135)[name = string("matrix_ac_39_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 64]> transpose_135 = transpose(perm = transpose_135_perm_0, x = k_77_cast_fp16)[name = string("transpose_177")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_134 = transpose(perm = transpose_134_perm_0, x = var_3366_cast_fp16)[name = string("transpose_178")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_39_cast_fp16 = matmul(transpose_x = matrix_ac_39_transpose_x_0, transpose_y = matrix_ac_39_transpose_y_0, x = transpose_134, y = transpose_135)[name = string("matrix_ac_39_cast_fp16")];
             tensor<int32, [4]> matrix_bd_79_begin_0 = const()[name = string("matrix_bd_79_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_79_end_0 = const()[name = string("matrix_bd_79_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_79_end_0 = const()[name = string("matrix_bd_79_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
             tensor<bool, [4]> matrix_bd_79_end_mask_0 = const()[name = string("matrix_bd_79_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_79_cast_fp16 = slice_by_index(begin = matrix_bd_79_begin_0, end = matrix_bd_79_end_0, end_mask = matrix_bd_79_end_mask_0, x = matrix_bd_77_cast_fp16)[name = string("matrix_bd_79_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_3392_cast_fp16 = add(x = matrix_ac_39_cast_fp16, y = matrix_bd_79_cast_fp16)[name = string("op_3392_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_79_cast_fp16 = slice_by_index(begin = matrix_bd_79_begin_0, end = matrix_bd_79_end_0, end_mask = matrix_bd_79_end_mask_0, x = matrix_bd_77_cast_fp16)[name = string("matrix_bd_79_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_3392_cast_fp16 = add(x = matrix_ac_39_cast_fp16, y = matrix_bd_79_cast_fp16)[name = string("op_3392_cast_fp16")];
             fp16 _inversed_scores_77_y_0_to_fp16 = const()[name = string("_inversed_scores_77_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_77_cast_fp16 = mul(x = var_3392_cast_fp16, y = _inversed_scores_77_y_0_to_fp16)[name = string("_inversed_scores_77_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_79_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_77_cast_fp16, cond = mask_3)[name = string("scores_79_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_3398_cast_fp16 = softmax(axis = var_30, x = scores_79_cast_fp16)[name = string("op_3398_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_1021_cast_fp16 = select(a = var_11_to_fp16, b = var_3398_cast_fp16, cond = mask_3)[name = string("input_1021_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_77_cast_fp16 = mul(x = var_3392_cast_fp16, y = _inversed_scores_77_y_0_to_fp16)[name = string("_inversed_scores_77_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_79_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_77_cast_fp16, cond = mask_3)[name = string("scores_79_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_3398_cast_fp16 = softmax(axis = var_30, x = scores_79_cast_fp16)[name = string("op_3398_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_1021_cast_fp16 = select(a = var_11_to_fp16, b = var_3398_cast_fp16, cond = mask_3)[name = string("input_1021_cast_fp16")];
             bool x_431_transpose_x_0 = const()[name = string("x_431_transpose_x_0"), val = bool(false)];
             bool x_431_transpose_y_0 = const()[name = string("x_431_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_41_cast_fp16 = transpose(perm = value_41_perm_0, x = v_39_cast_fp16)[name = string("transpose_176")];
-            tensor<fp16, [1, 8, 188, 128]> x_431_cast_fp16 = matmul(transpose_x = x_431_transpose_x_0, transpose_y = x_431_transpose_y_0, x = input_1021_cast_fp16, y = value_41_cast_fp16)[name = string("x_431_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 128]> value_41_cast_fp16 = transpose(perm = value_41_perm_0, x = v_39_cast_fp16)[name = string("transpose_176")];
+            tensor<fp16, [1, 8, 64, 128]> x_431_cast_fp16 = matmul(transpose_x = x_431_transpose_x_0, transpose_y = x_431_transpose_y_0, x = input_1021_cast_fp16, y = value_41_cast_fp16)[name = string("x_431_cast_fp16")];
             tensor<int32, [4]> var_3402_perm_0 = const()[name = string("op_3402_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_3403 = const()[name = string("op_3403"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_3402_cast_fp16 = transpose(perm = var_3402_perm_0, x = x_431_cast_fp16)[name = string("transpose_175")];
-            tensor<fp16, [1, 188, 1024]> input_1023_cast_fp16 = reshape(shape = var_3403, x = var_3402_cast_fp16)[name = string("input_1023_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523881]> module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261008192))), indices_nonzero_data = tensor<uint6, [523881]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261139328))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261532352))))[name = string("module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> var_3402_cast_fp16 = transpose(perm = var_3402_perm_0, x = x_431_cast_fp16)[name = string("transpose_175")];
+            tensor<fp16, [1, 64, 1024]> input_1023_cast_fp16 = reshape(shape = var_3403, x = var_3402_cast_fp16)[name = string("input_1023_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523881]> module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250849600))), indices_nonzero_data = tensor<uint6, [523881]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250980736))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251373760))))[name = string("module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_178_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_1023_cast_fp16)[name = string("linear_178_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_1027_cast_fp16 = add(x = input_1019_cast_fp16, y = linear_178_cast_fp16)[name = string("input_1027_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_178_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_1023_cast_fp16)[name = string("linear_178_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_1027_cast_fp16 = add(x = input_1019_cast_fp16, y = linear_178_cast_fp16)[name = string("input_1027_cast_fp16")];
             tensor<int32, [1]> x_435_axes_0 = const()[name = string("x_435_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_19_norm_conv_weight_to_fp16 = const()[name = string("module_layers_19_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261548800)))];
-            tensor<fp16, [1024]> module_layers_19_norm_conv_bias_to_fp16 = const()[name = string("module_layers_19_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261550912)))];
-            tensor<fp16, [1, 188, 1024]> x_435_cast_fp16 = layer_norm(axes = x_435_axes_0, beta = module_layers_19_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_conv_weight_to_fp16, x = input_1027_cast_fp16)[name = string("x_435_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_19_norm_conv_weight_to_fp16 = const()[name = string("module_layers_19_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251390208)))];
+            tensor<fp16, [1024]> module_layers_19_norm_conv_bias_to_fp16 = const()[name = string("module_layers_19_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251392320)))];
+            tensor<fp16, [1, 64, 1024]> x_435_cast_fp16 = layer_norm(axes = x_435_axes_0, beta = module_layers_19_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_conv_weight_to_fp16, x = input_1027_cast_fp16)[name = string("x_435_cast_fp16")];
             tensor<int32, [3]> input_1029_perm_0 = const()[name = string("input_1029_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_1031_pad_type_0 = const()[name = string("input_1031_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_1031_strides_0 = const()[name = string("input_1031_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_1031_pad_0 = const()[name = string("input_1031_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_1031_dilations_0 = const()[name = string("input_1031_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_1031_groups_0 = const()[name = string("input_1031_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047423]> module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261553024))), indices_nonzero_data = tensor<uint6, [1047423]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261815232))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262600896))))[name = string("module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1047423]> module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251394432))), indices_nonzero_data = tensor<uint6, [1047423]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251656640))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252442304))))[name = string("module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_1029_cast_fp16 = transpose(perm = input_1029_perm_0, x = x_435_cast_fp16)[name = string("transpose_174")];
-            tensor<fp16, [1, 2048, 188]> input_1031_cast_fp16 = conv(dilations = input_1031_dilations_0, groups = input_1031_groups_0, pad = input_1031_pad_0, pad_type = input_1031_pad_type_0, strides = input_1031_strides_0, weight = module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_1029_cast_fp16)[name = string("input_1031_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_1029_cast_fp16 = transpose(perm = input_1029_perm_0, x = x_435_cast_fp16)[name = string("transpose_174")];
+            tensor<fp16, [1, 2048, 64]> input_1031_cast_fp16 = conv(dilations = input_1031_dilations_0, groups = input_1031_groups_0, pad = input_1031_pad_0, pad_type = input_1031_pad_type_0, strides = input_1031_strides_0, weight = module_layers_19_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_1029_cast_fp16)[name = string("input_1031_cast_fp16")];
             int32 x_437_split_num_splits_0 = const()[name = string("x_437_split_num_splits_0"), val = int32(2)];
-            int32 x_437_split_axis_0 = const()[name = string("x_437_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_437_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_437_split_cast_fp16_1 = split(axis = x_437_split_axis_0, num_splits = x_437_split_num_splits_0, x = input_1031_cast_fp16)[name = string("x_437_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_437_split_1_sigmoid_cast_fp16 = sigmoid(x = x_437_split_cast_fp16_1)[name = string("x_437_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_437_cast_fp16 = mul(x = x_437_split_cast_fp16_0, y = x_437_split_1_sigmoid_cast_fp16)[name = string("x_437_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_1033_cast_fp16 = select(a = var_11_to_fp16, b = x_437_cast_fp16, cond = var_328)[name = string("input_1033_cast_fp16")];
+            int32 x_437_split_axis_0 = const()[name = string("x_437_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 64]> x_437_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_437_split_cast_fp16_1 = split(axis = x_437_split_axis_0, num_splits = x_437_split_num_splits_0, x = input_1031_cast_fp16)[name = string("x_437_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_437_split_1_sigmoid_cast_fp16 = sigmoid(x = x_437_split_cast_fp16_1)[name = string("x_437_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_437_cast_fp16 = mul(x = x_437_split_cast_fp16_0, y = x_437_split_1_sigmoid_cast_fp16)[name = string("x_437_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_1033_cast_fp16 = select(a = var_11_to_fp16, b = x_437_cast_fp16, cond = var_328)[name = string("input_1033_cast_fp16")];
             tensor<int32, [6]> input_1035_pad_0 = const()[name = string("input_1035_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_1035_mode_0 = const()[name = string("input_1035_mode_0"), val = string("constant")];
             fp16 const_207_to_fp16 = const()[name = string("const_207_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_1035_cast_fp16 = pad(constant_val = const_207_to_fp16, mode = input_1035_mode_0, pad = input_1035_pad_0, x = input_1033_cast_fp16)[name = string("input_1035_cast_fp16")];
+            tensor<fp16, [1, 1024, 72]> input_1035_cast_fp16 = pad(constant_val = const_207_to_fp16, mode = input_1035_mode_0, pad = input_1035_pad_0, x = input_1033_cast_fp16)[name = string("input_1035_cast_fp16")];
             string input_1037_pad_type_0 = const()[name = string("input_1037_pad_type_0"), val = string("valid")];
             int32 input_1037_groups_0 = const()[name = string("input_1037_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_1037_strides_0 = const()[name = string("input_1037_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_1037_pad_0 = const()[name = string("input_1037_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_1037_dilations_0 = const()[name = string("input_1037_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_286_to_fp16 = const()[name = string("const_286_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262633728)))];
-            tensor<fp16, [1024]> const_287_to_fp16 = const()[name = string("const_287_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262652224)))];
-            tensor<fp16, [1, 1024, 188]> input_1039_cast_fp16 = conv(bias = const_287_to_fp16, dilations = input_1037_dilations_0, groups = input_1037_groups_0, pad = input_1037_pad_0, pad_type = input_1037_pad_type_0, strides = input_1037_strides_0, weight = const_286_to_fp16, x = input_1035_cast_fp16)[name = string("input_1039_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_1041_cast_fp16 = silu(x = input_1039_cast_fp16)[name = string("input_1041_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_286_to_fp16 = const()[name = string("const_286_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252475136)))];
+            tensor<fp16, [1024]> const_287_to_fp16 = const()[name = string("const_287_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252493632)))];
+            tensor<fp16, [1, 1024, 64]> input_1039_cast_fp16 = conv(bias = const_287_to_fp16, dilations = input_1037_dilations_0, groups = input_1037_groups_0, pad = input_1037_pad_0, pad_type = input_1037_pad_type_0, strides = input_1037_strides_0, weight = const_286_to_fp16, x = input_1035_cast_fp16)[name = string("input_1039_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_1041_cast_fp16 = silu(x = input_1039_cast_fp16)[name = string("input_1041_cast_fp16")];
             string x_439_pad_type_0 = const()[name = string("x_439_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_439_strides_0 = const()[name = string("x_439_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_439_pad_0 = const()[name = string("x_439_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_439_dilations_0 = const()[name = string("x_439_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_439_groups_0 = const()[name = string("x_439_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519548]> module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262654336))), indices_nonzero_data = tensor<uint6, [519548]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262785472))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263175232))))[name = string("module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519548]> module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252495744))), indices_nonzero_data = tensor<uint6, [519548]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252626880))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253016640))))[name = string("module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_439_cast_fp16 = conv(dilations = x_439_dilations_0, groups = x_439_groups_0, pad = x_439_pad_0, pad_type = x_439_pad_type_0, strides = x_439_strides_0, weight = module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_1041_cast_fp16)[name = string("x_439_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_439_cast_fp16 = conv(dilations = x_439_dilations_0, groups = x_439_groups_0, pad = x_439_pad_0, pad_type = x_439_pad_type_0, strides = x_439_strides_0, weight = module_layers_19_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_1041_cast_fp16)[name = string("x_439_cast_fp16")];
             tensor<int32, [3]> input_1043_perm_0 = const()[name = string("input_1043_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_1043_cast_fp16 = transpose(perm = input_1043_perm_0, x = x_439_cast_fp16)[name = string("transpose_173")];
-            tensor<fp16, [1, 188, 1024]> input_1045_cast_fp16 = add(x = input_1027_cast_fp16, y = input_1043_cast_fp16)[name = string("input_1045_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_1043_cast_fp16 = transpose(perm = input_1043_perm_0, x = x_439_cast_fp16)[name = string("transpose_173")];
+            tensor<fp16, [1, 64, 1024]> input_1045_cast_fp16 = add(x = input_1027_cast_fp16, y = input_1043_cast_fp16)[name = string("input_1045_cast_fp16")];
             tensor<int32, [1]> input_1047_axes_0 = const()[name = string("input_1047_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_19_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_19_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263191680)))];
-            tensor<fp16, [1024]> module_layers_19_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_19_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263193792)))];
-            tensor<fp16, [1, 188, 1024]> input_1047_cast_fp16 = layer_norm(axes = input_1047_axes_0, beta = module_layers_19_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_feed_forward2_weight_to_fp16, x = input_1045_cast_fp16)[name = string("input_1047_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2091309]> module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263195904))), indices_nonzero_data = tensor<uint6, [2091309]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263720256))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265288832))))[name = string("module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_19_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_19_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253033088)))];
+            tensor<fp16, [1024]> module_layers_19_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_19_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253035200)))];
+            tensor<fp16, [1, 64, 1024]> input_1047_cast_fp16 = layer_norm(axes = input_1047_axes_0, beta = module_layers_19_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_feed_forward2_weight_to_fp16, x = input_1045_cast_fp16)[name = string("input_1047_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2091309]> module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253037312))), indices_nonzero_data = tensor<uint6, [2091309]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253561664))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(255130240))))[name = string("module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_179_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_1047_cast_fp16)[name = string("linear_179_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_1051_cast_fp16 = silu(x = linear_179_cast_fp16)[name = string("input_1051_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2085070]> module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265354432))), indices_nonzero_data = tensor<uint6, [2085070]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265878784))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267442688))))[name = string("module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 4096]> linear_179_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_19_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_1047_cast_fp16)[name = string("linear_179_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_1051_cast_fp16 = silu(x = linear_179_cast_fp16)[name = string("input_1051_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2085070]> module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(255195840))), indices_nonzero_data = tensor<uint6, [2085070]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(255720192))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257284096))))[name = string("module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_180_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_1051_cast_fp16)[name = string("linear_180_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_180_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_19_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_1051_cast_fp16)[name = string("linear_180_cast_fp16")];
             fp16 var_3463_to_fp16 = const()[name = string("op_3463_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_3464_cast_fp16 = mul(x = linear_180_cast_fp16, y = var_3463_to_fp16)[name = string("op_3464_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_1057_cast_fp16 = add(x = input_1045_cast_fp16, y = var_3464_cast_fp16)[name = string("input_1057_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> var_3464_cast_fp16 = mul(x = linear_180_cast_fp16, y = var_3463_to_fp16)[name = string("op_3464_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_1057_cast_fp16 = add(x = input_1045_cast_fp16, y = var_3464_cast_fp16)[name = string("input_1057_cast_fp16")];
             tensor<int32, [1]> input_1059_axes_0 = const()[name = string("input_1059_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_19_norm_out_weight_to_fp16 = const()[name = string("module_layers_19_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267459136)))];
-            tensor<fp16, [1024]> module_layers_19_norm_out_bias_to_fp16 = const()[name = string("module_layers_19_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267461248)))];
-            tensor<fp16, [1, 188, 1024]> input_1059_cast_fp16 = layer_norm(axes = input_1059_axes_0, beta = module_layers_19_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_out_weight_to_fp16, x = input_1057_cast_fp16)[name = string("input_1059_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_19_norm_out_weight_to_fp16 = const()[name = string("module_layers_19_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257300544)))];
+            tensor<fp16, [1024]> module_layers_19_norm_out_bias_to_fp16 = const()[name = string("module_layers_19_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257302656)))];
+            tensor<fp16, [1, 64, 1024]> input_1059_cast_fp16 = layer_norm(axes = input_1059_axes_0, beta = module_layers_19_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_out_weight_to_fp16, x = input_1057_cast_fp16)[name = string("input_1059_cast_fp16")];
             tensor<int32, [1]> input_1061_axes_0 = const()[name = string("input_1061_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_20_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_20_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267463360)))];
-            tensor<fp16, [1024]> module_layers_20_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_20_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267465472)))];
-            tensor<fp16, [1, 188, 1024]> input_1061_cast_fp16 = layer_norm(axes = input_1061_axes_0, beta = module_layers_20_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_feed_forward1_weight_to_fp16, x = input_1059_cast_fp16)[name = string("input_1061_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2088569]> module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267467584))), indices_nonzero_data = tensor<uint6, [2088569]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267991936))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269558464))))[name = string("module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_20_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_20_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257304768)))];
+            tensor<fp16, [1024]> module_layers_20_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_20_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257306880)))];
+            tensor<fp16, [1, 64, 1024]> input_1061_cast_fp16 = layer_norm(axes = input_1061_axes_0, beta = module_layers_20_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_feed_forward1_weight_to_fp16, x = input_1059_cast_fp16)[name = string("input_1061_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2088569]> module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257308992))), indices_nonzero_data = tensor<uint6, [2088569]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257833344))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259399872))))[name = string("module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_181_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_1061_cast_fp16)[name = string("linear_181_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_1065_cast_fp16 = silu(x = linear_181_cast_fp16)[name = string("input_1065_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2053925]> module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269624064))), indices_nonzero_data = tensor<uint6, [2053925]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270148416))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271688960))))[name = string("module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 4096]> linear_181_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_20_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_1061_cast_fp16)[name = string("linear_181_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_1065_cast_fp16 = silu(x = linear_181_cast_fp16)[name = string("input_1065_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2053925]> module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259465472))), indices_nonzero_data = tensor<uint6, [2053925]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259989824))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261530368))))[name = string("module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_182_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_1065_cast_fp16)[name = string("linear_182_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_182_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_1065_cast_fp16)[name = string("linear_182_cast_fp16")];
             fp16 var_3492_to_fp16 = const()[name = string("op_3492_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_3493_cast_fp16 = mul(x = linear_182_cast_fp16, y = var_3492_to_fp16)[name = string("op_3493_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_1071_cast_fp16 = add(x = input_1059_cast_fp16, y = var_3493_cast_fp16)[name = string("input_1071_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> var_3493_cast_fp16 = mul(x = linear_182_cast_fp16, y = var_3492_to_fp16)[name = string("op_3493_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_1071_cast_fp16 = add(x = input_1059_cast_fp16, y = var_3493_cast_fp16)[name = string("input_1071_cast_fp16")];
             tensor<int32, [1]> query_41_axes_0 = const()[name = string("query_41_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_20_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_20_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271705408)))];
-            tensor<fp16, [1024]> module_layers_20_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_20_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271707520)))];
-            tensor<fp16, [1, 188, 1024]> query_41_cast_fp16 = layer_norm(axes = query_41_axes_0, beta = module_layers_20_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_self_att_weight_to_fp16, x = input_1071_cast_fp16)[name = string("query_41_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525329]> module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271709632))), indices_nonzero_data = tensor<uint6, [525329]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271840768))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272234880))))[name = string("module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_20_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_20_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261546816)))];
+            tensor<fp16, [1024]> module_layers_20_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_20_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261548928)))];
+            tensor<fp16, [1, 64, 1024]> query_41_cast_fp16 = layer_norm(axes = query_41_axes_0, beta = module_layers_20_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_self_att_weight_to_fp16, x = input_1071_cast_fp16)[name = string("query_41_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525329]> module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261551040))), indices_nonzero_data = tensor<uint6, [525329]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261682176))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262076288))))[name = string("module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_183_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_41_cast_fp16)[name = string("linear_183_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_183_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_41_cast_fp16)[name = string("linear_183_cast_fp16")];
             tensor<int32, [4]> var_3509 = const()[name = string("op_3509"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_121_cast_fp16 = reshape(shape = var_3509, x = linear_183_cast_fp16)[name = string("q_121_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525107]> module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272251328))), indices_nonzero_data = tensor<uint6, [525107]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272382464))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272776384))))[name = string("module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> q_121_cast_fp16 = reshape(shape = var_3509, x = linear_183_cast_fp16)[name = string("q_121_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525107]> module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262092736))), indices_nonzero_data = tensor<uint6, [525107]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262223872))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262617792))))[name = string("module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_184_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_41_cast_fp16)[name = string("linear_184_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_184_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_41_cast_fp16)[name = string("linear_184_cast_fp16")];
             tensor<int32, [4]> var_3513 = const()[name = string("op_3513"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_81_cast_fp16 = reshape(shape = var_3513, x = linear_184_cast_fp16)[name = string("k_81_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525707]> module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272792832))), indices_nonzero_data = tensor<uint6, [525707]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272923968))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273318336))))[name = string("module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> k_81_cast_fp16 = reshape(shape = var_3513, x = linear_184_cast_fp16)[name = string("k_81_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525707]> module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262634240))), indices_nonzero_data = tensor<uint6, [525707]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262765376))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263159744))))[name = string("module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_185_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_41_cast_fp16)[name = string("linear_185_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_185_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_41_cast_fp16)[name = string("linear_185_cast_fp16")];
             tensor<int32, [4]> var_3517 = const()[name = string("op_3517"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_41_cast_fp16 = reshape(shape = var_3517, x = linear_185_cast_fp16)[name = string("v_41_cast_fp16")];
+            tensor<fp16, [1, 64, 8, 128]> v_41_cast_fp16 = reshape(shape = var_3517, x = linear_185_cast_fp16)[name = string("v_41_cast_fp16")];
             tensor<int32, [4]> value_43_perm_0 = const()[name = string("value_43_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_20_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_20_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273334784)))];
-            tensor<fp16, [1, 188, 8, 128]> var_3529_cast_fp16 = add(x = q_121_cast_fp16, y = module_layers_20_self_attn_pos_bias_u_to_fp16)[name = string("op_3529_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_20_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_20_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273336896)))];
-            tensor<fp16, [1, 188, 8, 128]> var_3531_cast_fp16 = add(x = q_121_cast_fp16, y = module_layers_20_self_attn_pos_bias_v_to_fp16)[name = string("op_3531_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_20_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_20_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263176192)))];
+            tensor<fp16, [1, 64, 8, 128]> var_3529_cast_fp16 = add(x = q_121_cast_fp16, y = module_layers_20_self_attn_pos_bias_u_to_fp16)[name = string("op_3529_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_20_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_20_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263178304)))];
+            tensor<fp16, [1, 64, 8, 128]> var_3531_cast_fp16 = add(x = q_121_cast_fp16, y = module_layers_20_self_attn_pos_bias_v_to_fp16)[name = string("op_3531_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_41_perm_0 = const()[name = string("q_with_bias_v_41_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_447_transpose_x_0 = const()[name = string("x_447_transpose_x_0"), val = bool(false)];
             bool x_447_transpose_y_0 = const()[name = string("x_447_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_3533_to_fp16 = const()[name = string("op_3533_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273339008)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_41_cast_fp16 = transpose(perm = q_with_bias_v_41_perm_0, x = var_3531_cast_fp16)[name = string("transpose_172")];
-            tensor<fp16, [1, 8, 188, 375]> x_447_cast_fp16 = matmul(transpose_x = x_447_transpose_x_0, transpose_y = x_447_transpose_y_0, x = q_with_bias_v_41_cast_fp16, y = var_3533_to_fp16)[name = string("x_447_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 127]> var_3533_to_fp16 = const()[name = string("op_3533_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263180416)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_41_cast_fp16 = transpose(perm = q_with_bias_v_41_perm_0, x = var_3531_cast_fp16)[name = string("transpose_172")];
+            tensor<fp16, [1, 8, 64, 127]> x_447_cast_fp16 = matmul(transpose_x = x_447_transpose_x_0, transpose_y = x_447_transpose_y_0, x = q_with_bias_v_41_cast_fp16, y = var_3533_to_fp16)[name = string("x_447_cast_fp16")];
             tensor<int32, [8]> x_449_pad_0 = const()[name = string("x_449_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_449_mode_0 = const()[name = string("x_449_mode_0"), val = string("constant")];
             fp16 const_214_to_fp16 = const()[name = string("const_214_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_449_cast_fp16 = pad(constant_val = const_214_to_fp16, mode = x_449_mode_0, pad = x_449_pad_0, x = x_447_cast_fp16)[name = string("x_449_cast_fp16")];
-            tensor<int32, [4]> var_3541 = const()[name = string("op_3541"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_451_cast_fp16 = reshape(shape = var_3541, x = x_449_cast_fp16)[name = string("x_451_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 128]> x_449_cast_fp16 = pad(constant_val = const_214_to_fp16, mode = x_449_mode_0, pad = x_449_pad_0, x = x_447_cast_fp16)[name = string("x_449_cast_fp16")];
+            tensor<int32, [4]> var_3541 = const()[name = string("op_3541"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_451_cast_fp16 = reshape(shape = var_3541, x = x_449_cast_fp16)[name = string("x_451_cast_fp16")];
             tensor<int32, [4]> var_3545_begin_0 = const()[name = string("op_3545_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_3545_end_0 = const()[name = string("op_3545_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_3545_end_0 = const()[name = string("op_3545_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
             tensor<bool, [4]> var_3545_end_mask_0 = const()[name = string("op_3545_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_3545_cast_fp16 = slice_by_index(begin = var_3545_begin_0, end = var_3545_end_0, end_mask = var_3545_end_mask_0, x = x_451_cast_fp16)[name = string("op_3545_cast_fp16")];
-            tensor<int32, [4]> var_3546 = const()[name = string("op_3546"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_81_cast_fp16 = reshape(shape = var_3546, x = var_3545_cast_fp16)[name = string("matrix_bd_81_cast_fp16")];
+            tensor<fp16, [1, 8, 127, 64]> var_3545_cast_fp16 = slice_by_index(begin = var_3545_begin_0, end = var_3545_end_0, end_mask = var_3545_end_mask_0, x = x_451_cast_fp16)[name = string("op_3545_cast_fp16")];
+            tensor<int32, [4]> var_3546 = const()[name = string("op_3546"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_81_cast_fp16 = reshape(shape = var_3546, x = var_3545_cast_fp16)[name = string("matrix_bd_81_cast_fp16")];
             bool matrix_ac_41_transpose_x_0 = const()[name = string("matrix_ac_41_transpose_x_0"), val = bool(false)];
             bool matrix_ac_41_transpose_y_0 = const()[name = string("matrix_ac_41_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_136_perm_0 = const()[name = string("transpose_136_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_137_perm_0 = const()[name = string("transpose_137_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_137 = transpose(perm = transpose_137_perm_0, x = k_81_cast_fp16)[name = string("transpose_170")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_136 = transpose(perm = transpose_136_perm_0, x = var_3529_cast_fp16)[name = string("transpose_171")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_41_cast_fp16 = matmul(transpose_x = matrix_ac_41_transpose_x_0, transpose_y = matrix_ac_41_transpose_y_0, x = transpose_136, y = transpose_137)[name = string("matrix_ac_41_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 64]> transpose_137 = transpose(perm = transpose_137_perm_0, x = k_81_cast_fp16)[name = string("transpose_170")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_136 = transpose(perm = transpose_136_perm_0, x = var_3529_cast_fp16)[name = string("transpose_171")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_41_cast_fp16 = matmul(transpose_x = matrix_ac_41_transpose_x_0, transpose_y = matrix_ac_41_transpose_y_0, x = transpose_136, y = transpose_137)[name = string("matrix_ac_41_cast_fp16")];
             tensor<int32, [4]> matrix_bd_83_begin_0 = const()[name = string("matrix_bd_83_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_83_end_0 = const()[name = string("matrix_bd_83_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_83_end_0 = const()[name = string("matrix_bd_83_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
             tensor<bool, [4]> matrix_bd_83_end_mask_0 = const()[name = string("matrix_bd_83_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_83_cast_fp16 = slice_by_index(begin = matrix_bd_83_begin_0, end = matrix_bd_83_end_0, end_mask = matrix_bd_83_end_mask_0, x = matrix_bd_81_cast_fp16)[name = string("matrix_bd_83_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_3555_cast_fp16 = add(x = matrix_ac_41_cast_fp16, y = matrix_bd_83_cast_fp16)[name = string("op_3555_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_83_cast_fp16 = slice_by_index(begin = matrix_bd_83_begin_0, end = matrix_bd_83_end_0, end_mask = matrix_bd_83_end_mask_0, x = matrix_bd_81_cast_fp16)[name = string("matrix_bd_83_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_3555_cast_fp16 = add(x = matrix_ac_41_cast_fp16, y = matrix_bd_83_cast_fp16)[name = string("op_3555_cast_fp16")];
             fp16 _inversed_scores_81_y_0_to_fp16 = const()[name = string("_inversed_scores_81_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_81_cast_fp16 = mul(x = var_3555_cast_fp16, y = _inversed_scores_81_y_0_to_fp16)[name = string("_inversed_scores_81_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_83_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_81_cast_fp16, cond = mask_3)[name = string("scores_83_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_3561_cast_fp16 = softmax(axis = var_30, x = scores_83_cast_fp16)[name = string("op_3561_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_1073_cast_fp16 = select(a = var_11_to_fp16, b = var_3561_cast_fp16, cond = mask_3)[name = string("input_1073_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_81_cast_fp16 = mul(x = var_3555_cast_fp16, y = _inversed_scores_81_y_0_to_fp16)[name = string("_inversed_scores_81_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_83_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_81_cast_fp16, cond = mask_3)[name = string("scores_83_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_3561_cast_fp16 = softmax(axis = var_30, x = scores_83_cast_fp16)[name = string("op_3561_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_1073_cast_fp16 = select(a = var_11_to_fp16, b = var_3561_cast_fp16, cond = mask_3)[name = string("input_1073_cast_fp16")];
             bool x_453_transpose_x_0 = const()[name = string("x_453_transpose_x_0"), val = bool(false)];
             bool x_453_transpose_y_0 = const()[name = string("x_453_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_43_cast_fp16 = transpose(perm = value_43_perm_0, x = v_41_cast_fp16)[name = string("transpose_169")];
-            tensor<fp16, [1, 8, 188, 128]> x_453_cast_fp16 = matmul(transpose_x = x_453_transpose_x_0, transpose_y = x_453_transpose_y_0, x = input_1073_cast_fp16, y = value_43_cast_fp16)[name = string("x_453_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 128]> value_43_cast_fp16 = transpose(perm = value_43_perm_0, x = v_41_cast_fp16)[name = string("transpose_169")];
+            tensor<fp16, [1, 8, 64, 128]> x_453_cast_fp16 = matmul(transpose_x = x_453_transpose_x_0, transpose_y = x_453_transpose_y_0, x = input_1073_cast_fp16, y = value_43_cast_fp16)[name = string("x_453_cast_fp16")];
             tensor<int32, [4]> var_3565_perm_0 = const()[name = string("op_3565_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_3566 = const()[name = string("op_3566"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_3565_cast_fp16 = transpose(perm = var_3565_perm_0, x = x_453_cast_fp16)[name = string("transpose_168")];
-            tensor<fp16, [1, 188, 1024]> input_1075_cast_fp16 = reshape(shape = var_3566, x = var_3565_cast_fp16)[name = string("input_1075_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [526127]> module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274107072))), indices_nonzero_data = tensor<uint6, [526127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274238208))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274632896))))[name = string("module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> var_3565_cast_fp16 = transpose(perm = var_3565_perm_0, x = x_453_cast_fp16)[name = string("transpose_168")];
+            tensor<fp16, [1, 64, 1024]> input_1075_cast_fp16 = reshape(shape = var_3566, x = var_3565_cast_fp16)[name = string("input_1075_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [526127]> module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263440576))), indices_nonzero_data = tensor<uint6, [526127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263571712))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263966400))))[name = string("module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_187_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_1075_cast_fp16)[name = string("linear_187_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_1079_cast_fp16 = add(x = input_1071_cast_fp16, y = linear_187_cast_fp16)[name = string("input_1079_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_187_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_1075_cast_fp16)[name = string("linear_187_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_1079_cast_fp16 = add(x = input_1071_cast_fp16, y = linear_187_cast_fp16)[name = string("input_1079_cast_fp16")];
             tensor<int32, [1]> x_457_axes_0 = const()[name = string("x_457_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_20_norm_conv_weight_to_fp16 = const()[name = string("module_layers_20_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274649344)))];
-            tensor<fp16, [1024]> module_layers_20_norm_conv_bias_to_fp16 = const()[name = string("module_layers_20_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274651456)))];
-            tensor<fp16, [1, 188, 1024]> x_457_cast_fp16 = layer_norm(axes = x_457_axes_0, beta = module_layers_20_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_conv_weight_to_fp16, x = input_1079_cast_fp16)[name = string("x_457_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_20_norm_conv_weight_to_fp16 = const()[name = string("module_layers_20_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263982848)))];
+            tensor<fp16, [1024]> module_layers_20_norm_conv_bias_to_fp16 = const()[name = string("module_layers_20_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263984960)))];
+            tensor<fp16, [1, 64, 1024]> x_457_cast_fp16 = layer_norm(axes = x_457_axes_0, beta = module_layers_20_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_conv_weight_to_fp16, x = input_1079_cast_fp16)[name = string("x_457_cast_fp16")];
             tensor<int32, [3]> input_1081_perm_0 = const()[name = string("input_1081_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_1083_pad_type_0 = const()[name = string("input_1083_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_1083_strides_0 = const()[name = string("input_1083_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_1083_pad_0 = const()[name = string("input_1083_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_1083_dilations_0 = const()[name = string("input_1083_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_1083_groups_0 = const()[name = string("input_1083_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1049203]> module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274653568))), indices_nonzero_data = tensor<uint6, [1049203]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274915776))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275702784))))[name = string("module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1049203]> module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263987072))), indices_nonzero_data = tensor<uint6, [1049203]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264249280))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265036288))))[name = string("module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_1081_cast_fp16 = transpose(perm = input_1081_perm_0, x = x_457_cast_fp16)[name = string("transpose_167")];
-            tensor<fp16, [1, 2048, 188]> input_1083_cast_fp16 = conv(dilations = input_1083_dilations_0, groups = input_1083_groups_0, pad = input_1083_pad_0, pad_type = input_1083_pad_type_0, strides = input_1083_strides_0, weight = module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_1081_cast_fp16)[name = string("input_1083_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_1081_cast_fp16 = transpose(perm = input_1081_perm_0, x = x_457_cast_fp16)[name = string("transpose_167")];
+            tensor<fp16, [1, 2048, 64]> input_1083_cast_fp16 = conv(dilations = input_1083_dilations_0, groups = input_1083_groups_0, pad = input_1083_pad_0, pad_type = input_1083_pad_type_0, strides = input_1083_strides_0, weight = module_layers_20_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_1081_cast_fp16)[name = string("input_1083_cast_fp16")];
             int32 x_459_split_num_splits_0 = const()[name = string("x_459_split_num_splits_0"), val = int32(2)];
             int32 x_459_split_axis_0 = const()[name = string("x_459_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_459_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_459_split_cast_fp16_1 = split(axis = x_459_split_axis_0, num_splits = x_459_split_num_splits_0, x = input_1083_cast_fp16)[name = string("x_459_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_459_split_1_sigmoid_cast_fp16 = sigmoid(x = x_459_split_cast_fp16_1)[name = string("x_459_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_459_cast_fp16 = mul(x = x_459_split_cast_fp16_0, y = x_459_split_1_sigmoid_cast_fp16)[name = string("x_459_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_1085_cast_fp16 = select(a = var_11_to_fp16, b = x_459_cast_fp16, cond = var_328)[name = string("input_1085_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_459_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_459_split_cast_fp16_1 = split(axis = x_459_split_axis_0, num_splits = x_459_split_num_splits_0, x = input_1083_cast_fp16)[name = string("x_459_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_459_split_1_sigmoid_cast_fp16 = sigmoid(x = x_459_split_cast_fp16_1)[name = string("x_459_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_459_cast_fp16 = mul(x = x_459_split_cast_fp16_0, y = x_459_split_1_sigmoid_cast_fp16)[name = string("x_459_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_1085_cast_fp16 = select(a = var_11_to_fp16, b = x_459_cast_fp16, cond = var_328)[name = string("input_1085_cast_fp16")];
             tensor<int32, [6]> input_1087_pad_0 = const()[name = string("input_1087_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_1087_mode_0 = const()[name = string("input_1087_mode_0"), val = string("constant")];
             fp16 const_217_to_fp16 = const()[name = string("const_217_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_1087_cast_fp16 = pad(constant_val = const_217_to_fp16, mode = input_1087_mode_0, pad = input_1087_pad_0, x = input_1085_cast_fp16)[name = string("input_1087_cast_fp16")];
+            tensor<fp16, [1, 1024, 72]> input_1087_cast_fp16 = pad(constant_val = const_217_to_fp16, mode = input_1087_mode_0, pad = input_1087_pad_0, x = input_1085_cast_fp16)[name = string("input_1087_cast_fp16")];
             string input_1089_pad_type_0 = const()[name = string("input_1089_pad_type_0"), val = string("valid")];
             int32 input_1089_groups_0 = const()[name = string("input_1089_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_1089_strides_0 = const()[name = string("input_1089_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_1089_pad_0 = const()[name = string("input_1089_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_1089_dilations_0 = const()[name = string("input_1089_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_288_to_fp16 = const()[name = string("const_288_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275735616)))];
-            tensor<fp16, [1024]> const_289_to_fp16 = const()[name = string("const_289_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275754112)))];
-            tensor<fp16, [1, 1024, 188]> input_1091_cast_fp16 = conv(bias = const_289_to_fp16, dilations = input_1089_dilations_0, groups = input_1089_groups_0, pad = input_1089_pad_0, pad_type = input_1089_pad_type_0, strides = input_1089_strides_0, weight = const_288_to_fp16, x = input_1087_cast_fp16)[name = string("input_1091_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_1093_cast_fp16 = silu(x = input_1091_cast_fp16)[name = string("input_1093_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_288_to_fp16 = const()[name = string("const_288_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265069120)))];
+            tensor<fp16, [1024]> const_289_to_fp16 = const()[name = string("const_289_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265087616)))];
+            tensor<fp16, [1, 1024, 64]> input_1091_cast_fp16 = conv(bias = const_289_to_fp16, dilations = input_1089_dilations_0, groups = input_1089_groups_0, pad = input_1089_pad_0, pad_type = input_1089_pad_type_0, strides = input_1089_strides_0, weight = const_288_to_fp16, x = input_1087_cast_fp16)[name = string("input_1091_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_1093_cast_fp16 = silu(x = input_1091_cast_fp16)[name = string("input_1093_cast_fp16")];
             string x_461_pad_type_0 = const()[name = string("x_461_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_461_strides_0 = const()[name = string("x_461_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_461_pad_0 = const()[name = string("x_461_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_461_dilations_0 = const()[name = string("x_461_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_461_groups_0 = const()[name = string("x_461_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [518556]> module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275756224))), indices_nonzero_data = tensor<uint6, [518556]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275887360))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276276352))))[name = string("module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [518556]> module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265089728))), indices_nonzero_data = tensor<uint6, [518556]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265220864))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265609856))))[name = string("module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_461_cast_fp16 = conv(dilations = x_461_dilations_0, groups = x_461_groups_0, pad = x_461_pad_0, pad_type = x_461_pad_type_0, strides = x_461_strides_0, weight = module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_1093_cast_fp16)[name = string("x_461_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_461_cast_fp16 = conv(dilations = x_461_dilations_0, groups = x_461_groups_0, pad = x_461_pad_0, pad_type = x_461_pad_type_0, strides = x_461_strides_0, weight = module_layers_20_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_1093_cast_fp16)[name = string("x_461_cast_fp16")];
             tensor<int32, [3]> input_1095_perm_0 = const()[name = string("input_1095_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_1095_cast_fp16 = transpose(perm = input_1095_perm_0, x = x_461_cast_fp16)[name = string("transpose_166")];
-            tensor<fp16, [1, 188, 1024]> input_1097_cast_fp16 = add(x = input_1079_cast_fp16, y = input_1095_cast_fp16)[name = string("input_1097_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_1095_cast_fp16 = transpose(perm = input_1095_perm_0, x = x_461_cast_fp16)[name = string("transpose_166")];
+            tensor<fp16, [1, 64, 1024]> input_1097_cast_fp16 = add(x = input_1079_cast_fp16, y = input_1095_cast_fp16)[name = string("input_1097_cast_fp16")];
             tensor<int32, [1]> input_1099_axes_0 = const()[name = string("input_1099_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_20_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_20_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276292800)))];
-            tensor<fp16, [1024]> module_layers_20_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_20_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276294912)))];
-            tensor<fp16, [1, 188, 1024]> input_1099_cast_fp16 = layer_norm(axes = input_1099_axes_0, beta = module_layers_20_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_feed_forward2_weight_to_fp16, x = input_1097_cast_fp16)[name = string("input_1099_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2095351]> module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276297024))), indices_nonzero_data = tensor<uint6, [2095351]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276821376))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278392960))))[name = string("module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_20_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_20_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265626304)))];
+            tensor<fp16, [1024]> module_layers_20_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_20_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265628416)))];
+            tensor<fp16, [1, 64, 1024]> input_1099_cast_fp16 = layer_norm(axes = input_1099_axes_0, beta = module_layers_20_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_feed_forward2_weight_to_fp16, x = input_1097_cast_fp16)[name = string("input_1099_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2095351]> module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265630528))), indices_nonzero_data = tensor<uint6, [2095351]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(266154880))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267726464))))[name = string("module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_188_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_1099_cast_fp16)[name = string("linear_188_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_1103_cast_fp16 = silu(x = linear_188_cast_fp16)[name = string("input_1103_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2074717]> module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278458560))), indices_nonzero_data = tensor<uint6, [2074717]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278982912))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280539072))))[name = string("module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 4096]> linear_188_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_20_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_1099_cast_fp16)[name = string("linear_188_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_1103_cast_fp16 = silu(x = linear_188_cast_fp16)[name = string("input_1103_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2074717]> module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267792064))), indices_nonzero_data = tensor<uint6, [2074717]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268316416))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269872576))))[name = string("module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_189_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_1103_cast_fp16)[name = string("linear_189_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_189_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_20_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_1103_cast_fp16)[name = string("linear_189_cast_fp16")];
             fp16 var_3626_to_fp16 = const()[name = string("op_3626_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_3627_cast_fp16 = mul(x = linear_189_cast_fp16, y = var_3626_to_fp16)[name = string("op_3627_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_1109_cast_fp16 = add(x = input_1097_cast_fp16, y = var_3627_cast_fp16)[name = string("input_1109_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> var_3627_cast_fp16 = mul(x = linear_189_cast_fp16, y = var_3626_to_fp16)[name = string("op_3627_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_1109_cast_fp16 = add(x = input_1097_cast_fp16, y = var_3627_cast_fp16)[name = string("input_1109_cast_fp16")];
             tensor<int32, [1]> input_1111_axes_0 = const()[name = string("input_1111_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_20_norm_out_weight_to_fp16 = const()[name = string("module_layers_20_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280555520)))];
-            tensor<fp16, [1024]> module_layers_20_norm_out_bias_to_fp16 = const()[name = string("module_layers_20_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280557632)))];
-            tensor<fp16, [1, 188, 1024]> input_1111_cast_fp16 = layer_norm(axes = input_1111_axes_0, beta = module_layers_20_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_out_weight_to_fp16, x = input_1109_cast_fp16)[name = string("input_1111_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_20_norm_out_weight_to_fp16 = const()[name = string("module_layers_20_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269889024)))];
+            tensor<fp16, [1024]> module_layers_20_norm_out_bias_to_fp16 = const()[name = string("module_layers_20_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269891136)))];
+            tensor<fp16, [1, 64, 1024]> input_1111_cast_fp16 = layer_norm(axes = input_1111_axes_0, beta = module_layers_20_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_out_weight_to_fp16, x = input_1109_cast_fp16)[name = string("input_1111_cast_fp16")];
             tensor<int32, [1]> input_1113_axes_0 = const()[name = string("input_1113_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_21_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_21_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280559744)))];
-            tensor<fp16, [1024]> module_layers_21_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_21_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280561856)))];
-            tensor<fp16, [1, 188, 1024]> input_1113_cast_fp16 = layer_norm(axes = input_1113_axes_0, beta = module_layers_21_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_feed_forward1_weight_to_fp16, x = input_1111_cast_fp16)[name = string("input_1113_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2093447]> module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280563968))), indices_nonzero_data = tensor<uint6, [2093447]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281088320))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282658496))))[name = string("module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_21_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_21_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269893248)))];
+            tensor<fp16, [1024]> module_layers_21_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_21_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269895360)))];
+            tensor<fp16, [1, 64, 1024]> input_1113_cast_fp16 = layer_norm(axes = input_1113_axes_0, beta = module_layers_21_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_feed_forward1_weight_to_fp16, x = input_1111_cast_fp16)[name = string("input_1113_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2093447]> module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269897472))), indices_nonzero_data = tensor<uint6, [2093447]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270421824))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271992000))))[name = string("module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_190_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_1113_cast_fp16)[name = string("linear_190_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_1117_cast_fp16 = silu(x = linear_190_cast_fp16)[name = string("input_1117_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2029278]> module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282724096))), indices_nonzero_data = tensor<uint6, [2029278]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(283248448))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284770496))))[name = string("module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 4096]> linear_190_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_21_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_1113_cast_fp16)[name = string("linear_190_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_1117_cast_fp16 = silu(x = linear_190_cast_fp16)[name = string("input_1117_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2029278]> module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272057600))), indices_nonzero_data = tensor<uint6, [2029278]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272581952))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274104000))))[name = string("module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_191_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_1117_cast_fp16)[name = string("linear_191_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_191_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_1117_cast_fp16)[name = string("linear_191_cast_fp16")];
             fp16 var_3655_to_fp16 = const()[name = string("op_3655_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_3656_cast_fp16 = mul(x = linear_191_cast_fp16, y = var_3655_to_fp16)[name = string("op_3656_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_1123_cast_fp16 = add(x = input_1111_cast_fp16, y = var_3656_cast_fp16)[name = string("input_1123_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> var_3656_cast_fp16 = mul(x = linear_191_cast_fp16, y = var_3655_to_fp16)[name = string("op_3656_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_1123_cast_fp16 = add(x = input_1111_cast_fp16, y = var_3656_cast_fp16)[name = string("input_1123_cast_fp16")];
             tensor<int32, [1]> query_43_axes_0 = const()[name = string("query_43_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_21_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_21_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284786944)))];
-            tensor<fp16, [1024]> module_layers_21_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_21_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284789056)))];
-            tensor<fp16, [1, 188, 1024]> query_43_cast_fp16 = layer_norm(axes = query_43_axes_0, beta = module_layers_21_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_self_att_weight_to_fp16, x = input_1123_cast_fp16)[name = string("query_43_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525507]> module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284791168))), indices_nonzero_data = tensor<uint6, [525507]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284922304))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285316544))))[name = string("module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_21_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_21_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274120448)))];
+            tensor<fp16, [1024]> module_layers_21_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_21_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274122560)))];
+            tensor<fp16, [1, 64, 1024]> query_43_cast_fp16 = layer_norm(axes = query_43_axes_0, beta = module_layers_21_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_self_att_weight_to_fp16, x = input_1123_cast_fp16)[name = string("query_43_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525507]> module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274124672))), indices_nonzero_data = tensor<uint6, [525507]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274255808))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274650048))))[name = string("module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_192_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_43_cast_fp16)[name = string("linear_192_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_192_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_43_cast_fp16)[name = string("linear_192_cast_fp16")];
             tensor<int32, [4]> var_3672 = const()[name = string("op_3672"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_127_cast_fp16 = reshape(shape = var_3672, x = linear_192_cast_fp16)[name = string("q_127_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525093]> module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285332992))), indices_nonzero_data = tensor<uint6, [525093]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285464128))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285858048))))[name = string("module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> q_127_cast_fp16 = reshape(shape = var_3672, x = linear_192_cast_fp16)[name = string("q_127_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525093]> module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274666496))), indices_nonzero_data = tensor<uint6, [525093]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274797632))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275191552))))[name = string("module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_193_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_43_cast_fp16)[name = string("linear_193_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_193_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_43_cast_fp16)[name = string("linear_193_cast_fp16")];
             tensor<int32, [4]> var_3676 = const()[name = string("op_3676"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_85_cast_fp16 = reshape(shape = var_3676, x = linear_193_cast_fp16)[name = string("k_85_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525954]> module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285874496))), indices_nonzero_data = tensor<uint6, [525954]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286005632))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286400192))))[name = string("module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> k_85_cast_fp16 = reshape(shape = var_3676, x = linear_193_cast_fp16)[name = string("k_85_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525954]> module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275208000))), indices_nonzero_data = tensor<uint6, [525954]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275339136))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275733696))))[name = string("module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_194_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_43_cast_fp16)[name = string("linear_194_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_194_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_43_cast_fp16)[name = string("linear_194_cast_fp16")];
             tensor<int32, [4]> var_3680 = const()[name = string("op_3680"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_43_cast_fp16 = reshape(shape = var_3680, x = linear_194_cast_fp16)[name = string("v_43_cast_fp16")];
+            tensor<fp16, [1, 64, 8, 128]> v_43_cast_fp16 = reshape(shape = var_3680, x = linear_194_cast_fp16)[name = string("v_43_cast_fp16")];
             tensor<int32, [4]> value_45_perm_0 = const()[name = string("value_45_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_21_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_21_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286416640)))];
-            tensor<fp16, [1, 188, 8, 128]> var_3692_cast_fp16 = add(x = q_127_cast_fp16, y = module_layers_21_self_attn_pos_bias_u_to_fp16)[name = string("op_3692_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_21_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_21_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286418752)))];
-            tensor<fp16, [1, 188, 8, 128]> var_3694_cast_fp16 = add(x = q_127_cast_fp16, y = module_layers_21_self_attn_pos_bias_v_to_fp16)[name = string("op_3694_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_21_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_21_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275750144)))];
+            tensor<fp16, [1, 64, 8, 128]> var_3692_cast_fp16 = add(x = q_127_cast_fp16, y = module_layers_21_self_attn_pos_bias_u_to_fp16)[name = string("op_3692_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_21_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_21_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275752256)))];
+            tensor<fp16, [1, 64, 8, 128]> var_3694_cast_fp16 = add(x = q_127_cast_fp16, y = module_layers_21_self_attn_pos_bias_v_to_fp16)[name = string("op_3694_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_43_perm_0 = const()[name = string("q_with_bias_v_43_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_469_transpose_x_0 = const()[name = string("x_469_transpose_x_0"), val = bool(false)];
             bool x_469_transpose_y_0 = const()[name = string("x_469_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_3696_to_fp16 = const()[name = string("op_3696_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286420864)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_43_cast_fp16 = transpose(perm = q_with_bias_v_43_perm_0, x = var_3694_cast_fp16)[name = string("transpose_165")];
-            tensor<fp16, [1, 8, 188, 375]> x_469_cast_fp16 = matmul(transpose_x = x_469_transpose_x_0, transpose_y = x_469_transpose_y_0, x = q_with_bias_v_43_cast_fp16, y = var_3696_to_fp16)[name = string("x_469_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 127]> var_3696_to_fp16 = const()[name = string("op_3696_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275754368)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_43_cast_fp16 = transpose(perm = q_with_bias_v_43_perm_0, x = var_3694_cast_fp16)[name = string("transpose_165")];
+            tensor<fp16, [1, 8, 64, 127]> x_469_cast_fp16 = matmul(transpose_x = x_469_transpose_x_0, transpose_y = x_469_transpose_y_0, x = q_with_bias_v_43_cast_fp16, y = var_3696_to_fp16)[name = string("x_469_cast_fp16")];
             tensor<int32, [8]> x_471_pad_0 = const()[name = string("x_471_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_471_mode_0 = const()[name = string("x_471_mode_0"), val = string("constant")];
             fp16 const_224_to_fp16 = const()[name = string("const_224_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_471_cast_fp16 = pad(constant_val = const_224_to_fp16, mode = x_471_mode_0, pad = x_471_pad_0, x = x_469_cast_fp16)[name = string("x_471_cast_fp16")];
-            tensor<int32, [4]> var_3704 = const()[name = string("op_3704"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_473_cast_fp16 = reshape(shape = var_3704, x = x_471_cast_fp16)[name = string("x_473_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 128]> x_471_cast_fp16 = pad(constant_val = const_224_to_fp16, mode = x_471_mode_0, pad = x_471_pad_0, x = x_469_cast_fp16)[name = string("x_471_cast_fp16")];
+            tensor<int32, [4]> var_3704 = const()[name = string("op_3704"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_473_cast_fp16 = reshape(shape = var_3704, x = x_471_cast_fp16)[name = string("x_473_cast_fp16")];
             tensor<int32, [4]> var_3708_begin_0 = const()[name = string("op_3708_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_3708_end_0 = const()[name = string("op_3708_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_3708_end_0 = const()[name = string("op_3708_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
             tensor<bool, [4]> var_3708_end_mask_0 = const()[name = string("op_3708_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_3708_cast_fp16 = slice_by_index(begin = var_3708_begin_0, end = var_3708_end_0, end_mask = var_3708_end_mask_0, x = x_473_cast_fp16)[name = string("op_3708_cast_fp16")];
-            tensor<int32, [4]> var_3709 = const()[name = string("op_3709"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_85_cast_fp16 = reshape(shape = var_3709, x = var_3708_cast_fp16)[name = string("matrix_bd_85_cast_fp16")];
+            tensor<fp16, [1, 8, 127, 64]> var_3708_cast_fp16 = slice_by_index(begin = var_3708_begin_0, end = var_3708_end_0, end_mask = var_3708_end_mask_0, x = x_473_cast_fp16)[name = string("op_3708_cast_fp16")];
+            tensor<int32, [4]> var_3709 = const()[name = string("op_3709"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_85_cast_fp16 = reshape(shape = var_3709, x = var_3708_cast_fp16)[name = string("matrix_bd_85_cast_fp16")];
             bool matrix_ac_43_transpose_x_0 = const()[name = string("matrix_ac_43_transpose_x_0"), val = bool(false)];
             bool matrix_ac_43_transpose_y_0 = const()[name = string("matrix_ac_43_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_138_perm_0 = const()[name = string("transpose_138_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_139_perm_0 = const()[name = string("transpose_139_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_139 = transpose(perm = transpose_139_perm_0, x = k_85_cast_fp16)[name = string("transpose_163")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_138 = transpose(perm = transpose_138_perm_0, x = var_3692_cast_fp16)[name = string("transpose_164")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_43_cast_fp16 = matmul(transpose_x = matrix_ac_43_transpose_x_0, transpose_y = matrix_ac_43_transpose_y_0, x = transpose_138, y = transpose_139)[name = string("matrix_ac_43_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 64]> transpose_139 = transpose(perm = transpose_139_perm_0, x = k_85_cast_fp16)[name = string("transpose_163")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_138 = transpose(perm = transpose_138_perm_0, x = var_3692_cast_fp16)[name = string("transpose_164")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_43_cast_fp16 = matmul(transpose_x = matrix_ac_43_transpose_x_0, transpose_y = matrix_ac_43_transpose_y_0, x = transpose_138, y = transpose_139)[name = string("matrix_ac_43_cast_fp16")];
             tensor<int32, [4]> matrix_bd_87_begin_0 = const()[name = string("matrix_bd_87_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_87_end_0 = const()[name = string("matrix_bd_87_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_87_end_0 = const()[name = string("matrix_bd_87_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
             tensor<bool, [4]> matrix_bd_87_end_mask_0 = const()[name = string("matrix_bd_87_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_87_cast_fp16 = slice_by_index(begin = matrix_bd_87_begin_0, end = matrix_bd_87_end_0, end_mask = matrix_bd_87_end_mask_0, x = matrix_bd_85_cast_fp16)[name = string("matrix_bd_87_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_3718_cast_fp16 = add(x = matrix_ac_43_cast_fp16, y = matrix_bd_87_cast_fp16)[name = string("op_3718_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_87_cast_fp16 = slice_by_index(begin = matrix_bd_87_begin_0, end = matrix_bd_87_end_0, end_mask = matrix_bd_87_end_mask_0, x = matrix_bd_85_cast_fp16)[name = string("matrix_bd_87_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_3718_cast_fp16 = add(x = matrix_ac_43_cast_fp16, y = matrix_bd_87_cast_fp16)[name = string("op_3718_cast_fp16")];
             fp16 _inversed_scores_85_y_0_to_fp16 = const()[name = string("_inversed_scores_85_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_85_cast_fp16 = mul(x = var_3718_cast_fp16, y = _inversed_scores_85_y_0_to_fp16)[name = string("_inversed_scores_85_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_87_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_85_cast_fp16, cond = mask_3)[name = string("scores_87_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_3724_cast_fp16 = softmax(axis = var_30, x = scores_87_cast_fp16)[name = string("op_3724_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_1125_cast_fp16 = select(a = var_11_to_fp16, b = var_3724_cast_fp16, cond = mask_3)[name = string("input_1125_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_85_cast_fp16 = mul(x = var_3718_cast_fp16, y = _inversed_scores_85_y_0_to_fp16)[name = string("_inversed_scores_85_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_87_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_85_cast_fp16, cond = mask_3)[name = string("scores_87_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_3724_cast_fp16 = softmax(axis = var_30, x = scores_87_cast_fp16)[name = string("op_3724_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_1125_cast_fp16 = select(a = var_11_to_fp16, b = var_3724_cast_fp16, cond = mask_3)[name = string("input_1125_cast_fp16")];
             bool x_475_transpose_x_0 = const()[name = string("x_475_transpose_x_0"), val = bool(false)];
             bool x_475_transpose_y_0 = const()[name = string("x_475_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_45_cast_fp16 = transpose(perm = value_45_perm_0, x = v_43_cast_fp16)[name = string("transpose_162")];
-            tensor<fp16, [1, 8, 188, 128]> x_475_cast_fp16 = matmul(transpose_x = x_475_transpose_x_0, transpose_y = x_475_transpose_y_0, x = input_1125_cast_fp16, y = value_45_cast_fp16)[name = string("x_475_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 128]> value_45_cast_fp16 = transpose(perm = value_45_perm_0, x = v_43_cast_fp16)[name = string("transpose_162")];
+            tensor<fp16, [1, 8, 64, 128]> x_475_cast_fp16 = matmul(transpose_x = x_475_transpose_x_0, transpose_y = x_475_transpose_y_0, x = input_1125_cast_fp16, y = value_45_cast_fp16)[name = string("x_475_cast_fp16")];
             tensor<int32, [4]> var_3728_perm_0 = const()[name = string("op_3728_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_3729 = const()[name = string("op_3729"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_3728_cast_fp16 = transpose(perm = var_3728_perm_0, x = x_475_cast_fp16)[name = string("transpose_161")];
-            tensor<fp16, [1, 188, 1024]> input_1127_cast_fp16 = reshape(shape = var_3729, x = var_3728_cast_fp16)[name = string("input_1127_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [526102]> module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287188928))), indices_nonzero_data = tensor<uint6, [526102]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287320064))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287714752))))[name = string("module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> var_3728_cast_fp16 = transpose(perm = var_3728_perm_0, x = x_475_cast_fp16)[name = string("transpose_161")];
+            tensor<fp16, [1, 64, 1024]> input_1127_cast_fp16 = reshape(shape = var_3729, x = var_3728_cast_fp16)[name = string("input_1127_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [526102]> module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276014528))), indices_nonzero_data = tensor<uint6, [526102]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276145664))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276540352))))[name = string("module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_196_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_1127_cast_fp16)[name = string("linear_196_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_1131_cast_fp16 = add(x = input_1123_cast_fp16, y = linear_196_cast_fp16)[name = string("input_1131_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_196_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_1127_cast_fp16)[name = string("linear_196_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_1131_cast_fp16 = add(x = input_1123_cast_fp16, y = linear_196_cast_fp16)[name = string("input_1131_cast_fp16")];
             tensor<int32, [1]> x_479_axes_0 = const()[name = string("x_479_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_21_norm_conv_weight_to_fp16 = const()[name = string("module_layers_21_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287731200)))];
-            tensor<fp16, [1024]> module_layers_21_norm_conv_bias_to_fp16 = const()[name = string("module_layers_21_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287733312)))];
-            tensor<fp16, [1, 188, 1024]> x_479_cast_fp16 = layer_norm(axes = x_479_axes_0, beta = module_layers_21_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_conv_weight_to_fp16, x = input_1131_cast_fp16)[name = string("x_479_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_21_norm_conv_weight_to_fp16 = const()[name = string("module_layers_21_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276556800)))];
+            tensor<fp16, [1024]> module_layers_21_norm_conv_bias_to_fp16 = const()[name = string("module_layers_21_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276558912)))];
+            tensor<fp16, [1, 64, 1024]> x_479_cast_fp16 = layer_norm(axes = x_479_axes_0, beta = module_layers_21_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_conv_weight_to_fp16, x = input_1131_cast_fp16)[name = string("x_479_cast_fp16")];
             tensor<int32, [3]> input_1133_perm_0 = const()[name = string("input_1133_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_1135_pad_type_0 = const()[name = string("input_1135_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_1135_strides_0 = const()[name = string("input_1135_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_1135_pad_0 = const()[name = string("input_1135_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_1135_dilations_0 = const()[name = string("input_1135_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_1135_groups_0 = const()[name = string("input_1135_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1049320]> module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287735424))), indices_nonzero_data = tensor<uint6, [1049320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287997632))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288784704))))[name = string("module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1049320]> module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276561024))), indices_nonzero_data = tensor<uint6, [1049320]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276823232))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277610304))))[name = string("module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_1133_cast_fp16 = transpose(perm = input_1133_perm_0, x = x_479_cast_fp16)[name = string("transpose_160")];
-            tensor<fp16, [1, 2048, 188]> input_1135_cast_fp16 = conv(dilations = input_1135_dilations_0, groups = input_1135_groups_0, pad = input_1135_pad_0, pad_type = input_1135_pad_type_0, strides = input_1135_strides_0, weight = module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_1133_cast_fp16)[name = string("input_1135_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_1133_cast_fp16 = transpose(perm = input_1133_perm_0, x = x_479_cast_fp16)[name = string("transpose_160")];
+            tensor<fp16, [1, 2048, 64]> input_1135_cast_fp16 = conv(dilations = input_1135_dilations_0, groups = input_1135_groups_0, pad = input_1135_pad_0, pad_type = input_1135_pad_type_0, strides = input_1135_strides_0, weight = module_layers_21_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_1133_cast_fp16)[name = string("input_1135_cast_fp16")];
             int32 x_481_split_num_splits_0 = const()[name = string("x_481_split_num_splits_0"), val = int32(2)];
             int32 x_481_split_axis_0 = const()[name = string("x_481_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_481_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_481_split_cast_fp16_1 = split(axis = x_481_split_axis_0, num_splits = x_481_split_num_splits_0, x = input_1135_cast_fp16)[name = string("x_481_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_481_split_1_sigmoid_cast_fp16 = sigmoid(x = x_481_split_cast_fp16_1)[name = string("x_481_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_481_cast_fp16 = mul(x = x_481_split_cast_fp16_0, y = x_481_split_1_sigmoid_cast_fp16)[name = string("x_481_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_1137_cast_fp16 = select(a = var_11_to_fp16, b = x_481_cast_fp16, cond = var_328)[name = string("input_1137_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_481_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_481_split_cast_fp16_1 = split(axis = x_481_split_axis_0, num_splits = x_481_split_num_splits_0, x = input_1135_cast_fp16)[name = string("x_481_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_481_split_1_sigmoid_cast_fp16 = sigmoid(x = x_481_split_cast_fp16_1)[name = string("x_481_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_481_cast_fp16 = mul(x = x_481_split_cast_fp16_0, y = x_481_split_1_sigmoid_cast_fp16)[name = string("x_481_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_1137_cast_fp16 = select(a = var_11_to_fp16, b = x_481_cast_fp16, cond = var_328)[name = string("input_1137_cast_fp16")];
             tensor<int32, [6]> input_1139_pad_0 = const()[name = string("input_1139_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_1139_mode_0 = const()[name = string("input_1139_mode_0"), val = string("constant")];
             fp16 const_227_to_fp16 = const()[name = string("const_227_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_1139_cast_fp16 = pad(constant_val = const_227_to_fp16, mode = input_1139_mode_0, pad = input_1139_pad_0, x = input_1137_cast_fp16)[name = string("input_1139_cast_fp16")];
+            tensor<fp16, [1, 1024, 72]> input_1139_cast_fp16 = pad(constant_val = const_227_to_fp16, mode = input_1139_mode_0, pad = input_1139_pad_0, x = input_1137_cast_fp16)[name = string("input_1139_cast_fp16")];
             string input_1141_pad_type_0 = const()[name = string("input_1141_pad_type_0"), val = string("valid")];
             int32 input_1141_groups_0 = const()[name = string("input_1141_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_1141_strides_0 = const()[name = string("input_1141_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_1141_pad_0 = const()[name = string("input_1141_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_1141_dilations_0 = const()[name = string("input_1141_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_290_to_fp16 = const()[name = string("const_290_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288817536)))];
-            tensor<fp16, [1024]> const_291_to_fp16 = const()[name = string("const_291_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288836032)))];
-            tensor<fp16, [1, 1024, 188]> input_1143_cast_fp16 = conv(bias = const_291_to_fp16, dilations = input_1141_dilations_0, groups = input_1141_groups_0, pad = input_1141_pad_0, pad_type = input_1141_pad_type_0, strides = input_1141_strides_0, weight = const_290_to_fp16, x = input_1139_cast_fp16)[name = string("input_1143_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_1145_cast_fp16 = silu(x = input_1143_cast_fp16)[name = string("input_1145_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_290_to_fp16 = const()[name = string("const_290_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277643136)))];
+            tensor<fp16, [1024]> const_291_to_fp16 = const()[name = string("const_291_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277661632)))];
+            tensor<fp16, [1, 1024, 64]> input_1143_cast_fp16 = conv(bias = const_291_to_fp16, dilations = input_1141_dilations_0, groups = input_1141_groups_0, pad = input_1141_pad_0, pad_type = input_1141_pad_type_0, strides = input_1141_strides_0, weight = const_290_to_fp16, x = input_1139_cast_fp16)[name = string("input_1143_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_1145_cast_fp16 = silu(x = input_1143_cast_fp16)[name = string("input_1145_cast_fp16")];
             string x_483_pad_type_0 = const()[name = string("x_483_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_483_strides_0 = const()[name = string("x_483_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_483_pad_0 = const()[name = string("x_483_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_483_dilations_0 = const()[name = string("x_483_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_483_groups_0 = const()[name = string("x_483_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519430]> module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288838144))), indices_nonzero_data = tensor<uint6, [519430]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288969280))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289358976))))[name = string("module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519430]> module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277663744))), indices_nonzero_data = tensor<uint6, [519430]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277794880))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278184576))))[name = string("module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_483_cast_fp16 = conv(dilations = x_483_dilations_0, groups = x_483_groups_0, pad = x_483_pad_0, pad_type = x_483_pad_type_0, strides = x_483_strides_0, weight = module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_1145_cast_fp16)[name = string("x_483_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_483_cast_fp16 = conv(dilations = x_483_dilations_0, groups = x_483_groups_0, pad = x_483_pad_0, pad_type = x_483_pad_type_0, strides = x_483_strides_0, weight = module_layers_21_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_1145_cast_fp16)[name = string("x_483_cast_fp16")];
             tensor<int32, [3]> input_1147_perm_0 = const()[name = string("input_1147_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_1147_cast_fp16 = transpose(perm = input_1147_perm_0, x = x_483_cast_fp16)[name = string("transpose_159")];
-            tensor<fp16, [1, 188, 1024]> input_1149_cast_fp16 = add(x = input_1131_cast_fp16, y = input_1147_cast_fp16)[name = string("input_1149_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_1147_cast_fp16 = transpose(perm = input_1147_perm_0, x = x_483_cast_fp16)[name = string("transpose_159")];
+            tensor<fp16, [1, 64, 1024]> input_1149_cast_fp16 = add(x = input_1131_cast_fp16, y = input_1147_cast_fp16)[name = string("input_1149_cast_fp16")];
             tensor<int32, [1]> input_1151_axes_0 = const()[name = string("input_1151_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_21_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_21_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289375424)))];
-            tensor<fp16, [1024]> module_layers_21_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_21_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289377536)))];
-            tensor<fp16, [1, 188, 1024]> input_1151_cast_fp16 = layer_norm(axes = input_1151_axes_0, beta = module_layers_21_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_feed_forward2_weight_to_fp16, x = input_1149_cast_fp16)[name = string("input_1151_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2100215]> module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289379648))), indices_nonzero_data = tensor<uint6, [2100215]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289904000))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(291479232))))[name = string("module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_21_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_21_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278201024)))];
+            tensor<fp16, [1024]> module_layers_21_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_21_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278203136)))];
+            tensor<fp16, [1, 64, 1024]> input_1151_cast_fp16 = layer_norm(axes = input_1151_axes_0, beta = module_layers_21_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_feed_forward2_weight_to_fp16, x = input_1149_cast_fp16)[name = string("input_1151_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2100215]> module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278205248))), indices_nonzero_data = tensor<uint6, [2100215]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278729600))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280304832))))[name = string("module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_197_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_1151_cast_fp16)[name = string("linear_197_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_1155_cast_fp16 = silu(x = linear_197_cast_fp16)[name = string("input_1155_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2058868]> module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(291544832))), indices_nonzero_data = tensor<uint6, [2058868]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292069184))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293613440))))[name = string("module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 4096]> linear_197_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_21_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_1151_cast_fp16)[name = string("linear_197_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_1155_cast_fp16 = silu(x = linear_197_cast_fp16)[name = string("input_1155_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2058868]> module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280370432))), indices_nonzero_data = tensor<uint6, [2058868]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280894784))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282439040))))[name = string("module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_198_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_1155_cast_fp16)[name = string("linear_198_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_198_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_21_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_1155_cast_fp16)[name = string("linear_198_cast_fp16")];
             fp16 var_3789_to_fp16 = const()[name = string("op_3789_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_3790_cast_fp16 = mul(x = linear_198_cast_fp16, y = var_3789_to_fp16)[name = string("op_3790_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_1161_cast_fp16 = add(x = input_1149_cast_fp16, y = var_3790_cast_fp16)[name = string("input_1161_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> var_3790_cast_fp16 = mul(x = linear_198_cast_fp16, y = var_3789_to_fp16)[name = string("op_3790_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_1161_cast_fp16 = add(x = input_1149_cast_fp16, y = var_3790_cast_fp16)[name = string("input_1161_cast_fp16")];
             tensor<int32, [1]> input_1163_axes_0 = const()[name = string("input_1163_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_21_norm_out_weight_to_fp16 = const()[name = string("module_layers_21_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293629888)))];
-            tensor<fp16, [1024]> module_layers_21_norm_out_bias_to_fp16 = const()[name = string("module_layers_21_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293632000)))];
-            tensor<fp16, [1, 188, 1024]> input_1163_cast_fp16 = layer_norm(axes = input_1163_axes_0, beta = module_layers_21_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_out_weight_to_fp16, x = input_1161_cast_fp16)[name = string("input_1163_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_21_norm_out_weight_to_fp16 = const()[name = string("module_layers_21_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282455488)))];
+            tensor<fp16, [1024]> module_layers_21_norm_out_bias_to_fp16 = const()[name = string("module_layers_21_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282457600)))];
+            tensor<fp16, [1, 64, 1024]> input_1163_cast_fp16 = layer_norm(axes = input_1163_axes_0, beta = module_layers_21_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_out_weight_to_fp16, x = input_1161_cast_fp16)[name = string("input_1163_cast_fp16")];
             tensor<int32, [1]> input_1165_axes_0 = const()[name = string("input_1165_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_22_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_22_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293634112)))];
-            tensor<fp16, [1024]> module_layers_22_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_22_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293636224)))];
-            tensor<fp16, [1, 188, 1024]> input_1165_cast_fp16 = layer_norm(axes = input_1165_axes_0, beta = module_layers_22_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_feed_forward1_weight_to_fp16, x = input_1163_cast_fp16)[name = string("input_1165_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2098607]> module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293638336))), indices_nonzero_data = tensor<uint6, [2098607]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294162688))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295736768))))[name = string("module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_22_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_22_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282459712)))];
+            tensor<fp16, [1024]> module_layers_22_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_22_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282461824)))];
+            tensor<fp16, [1, 64, 1024]> input_1165_cast_fp16 = layer_norm(axes = input_1165_axes_0, beta = module_layers_22_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_feed_forward1_weight_to_fp16, x = input_1163_cast_fp16)[name = string("input_1165_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2098607]> module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282463936))), indices_nonzero_data = tensor<uint6, [2098607]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282988288))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284562368))))[name = string("module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_199_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_1165_cast_fp16)[name = string("linear_199_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_1169_cast_fp16 = silu(x = linear_199_cast_fp16)[name = string("input_1169_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2035005]> module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295802368))), indices_nonzero_data = tensor<uint6, [2035005]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296326720))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297853056))))[name = string("module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 4096]> linear_199_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_22_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_1165_cast_fp16)[name = string("linear_199_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_1169_cast_fp16 = silu(x = linear_199_cast_fp16)[name = string("input_1169_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2035005]> module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284627968))), indices_nonzero_data = tensor<uint6, [2035005]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285152320))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286678656))))[name = string("module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_200_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_1169_cast_fp16)[name = string("linear_200_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_200_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_1169_cast_fp16)[name = string("linear_200_cast_fp16")];
             fp16 var_3818_to_fp16 = const()[name = string("op_3818_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_3819_cast_fp16 = mul(x = linear_200_cast_fp16, y = var_3818_to_fp16)[name = string("op_3819_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_1175_cast_fp16 = add(x = input_1163_cast_fp16, y = var_3819_cast_fp16)[name = string("input_1175_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> var_3819_cast_fp16 = mul(x = linear_200_cast_fp16, y = var_3818_to_fp16)[name = string("op_3819_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_1175_cast_fp16 = add(x = input_1163_cast_fp16, y = var_3819_cast_fp16)[name = string("input_1175_cast_fp16")];
             tensor<int32, [1]> query_45_axes_0 = const()[name = string("query_45_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_22_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_22_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297869504)))];
-            tensor<fp16, [1024]> module_layers_22_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_22_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297871616)))];
-            tensor<fp16, [1, 188, 1024]> query_45_cast_fp16 = layer_norm(axes = query_45_axes_0, beta = module_layers_22_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_self_att_weight_to_fp16, x = input_1175_cast_fp16)[name = string("query_45_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [527072]> module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297873728))), indices_nonzero_data = tensor<uint6, [527072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298004864))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298400256))))[name = string("module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_22_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_22_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286695104)))];
+            tensor<fp16, [1024]> module_layers_22_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_22_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286697216)))];
+            tensor<fp16, [1, 64, 1024]> query_45_cast_fp16 = layer_norm(axes = query_45_axes_0, beta = module_layers_22_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_self_att_weight_to_fp16, x = input_1175_cast_fp16)[name = string("query_45_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [527072]> module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286699328))), indices_nonzero_data = tensor<uint6, [527072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286830464))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287225856))))[name = string("module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_201_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_45_cast_fp16)[name = string("linear_201_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_201_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_45_cast_fp16)[name = string("linear_201_cast_fp16")];
             tensor<int32, [4]> var_3835 = const()[name = string("op_3835"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_133_cast_fp16 = reshape(shape = var_3835, x = linear_201_cast_fp16)[name = string("q_133_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525242]> module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298416704))), indices_nonzero_data = tensor<uint6, [525242]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298547840))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298941888))))[name = string("module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> q_133_cast_fp16 = reshape(shape = var_3835, x = linear_201_cast_fp16)[name = string("q_133_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525242]> module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287242304))), indices_nonzero_data = tensor<uint6, [525242]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287373440))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287767488))))[name = string("module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_202_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_45_cast_fp16)[name = string("linear_202_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_202_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_45_cast_fp16)[name = string("linear_202_cast_fp16")];
             tensor<int32, [4]> var_3839 = const()[name = string("op_3839"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_89_cast_fp16 = reshape(shape = var_3839, x = linear_202_cast_fp16)[name = string("k_89_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523544]> module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298958336))), indices_nonzero_data = tensor<uint6, [523544]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299089472))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299482240))))[name = string("module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> k_89_cast_fp16 = reshape(shape = var_3839, x = linear_202_cast_fp16)[name = string("k_89_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523544]> module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287783936))), indices_nonzero_data = tensor<uint6, [523544]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287915072))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288307840))))[name = string("module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_203_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_45_cast_fp16)[name = string("linear_203_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_203_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_45_cast_fp16)[name = string("linear_203_cast_fp16")];
             tensor<int32, [4]> var_3843 = const()[name = string("op_3843"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_45_cast_fp16 = reshape(shape = var_3843, x = linear_203_cast_fp16)[name = string("v_45_cast_fp16")];
+            tensor<fp16, [1, 64, 8, 128]> v_45_cast_fp16 = reshape(shape = var_3843, x = linear_203_cast_fp16)[name = string("v_45_cast_fp16")];
             tensor<int32, [4]> value_47_perm_0 = const()[name = string("value_47_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_22_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_22_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299498688)))];
-            tensor<fp16, [1, 188, 8, 128]> var_3855_cast_fp16 = add(x = q_133_cast_fp16, y = module_layers_22_self_attn_pos_bias_u_to_fp16)[name = string("op_3855_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_22_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_22_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299500800)))];
-            tensor<fp16, [1, 188, 8, 128]> var_3857_cast_fp16 = add(x = q_133_cast_fp16, y = module_layers_22_self_attn_pos_bias_v_to_fp16)[name = string("op_3857_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_22_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_22_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288324288)))];
+            tensor<fp16, [1, 64, 8, 128]> var_3855_cast_fp16 = add(x = q_133_cast_fp16, y = module_layers_22_self_attn_pos_bias_u_to_fp16)[name = string("op_3855_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_22_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_22_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288326400)))];
+            tensor<fp16, [1, 64, 8, 128]> var_3857_cast_fp16 = add(x = q_133_cast_fp16, y = module_layers_22_self_attn_pos_bias_v_to_fp16)[name = string("op_3857_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_45_perm_0 = const()[name = string("q_with_bias_v_45_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_491_transpose_x_0 = const()[name = string("x_491_transpose_x_0"), val = bool(false)];
             bool x_491_transpose_y_0 = const()[name = string("x_491_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_3859_to_fp16 = const()[name = string("op_3859_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299502912)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_45_cast_fp16 = transpose(perm = q_with_bias_v_45_perm_0, x = var_3857_cast_fp16)[name = string("transpose_158")];
-            tensor<fp16, [1, 8, 188, 375]> x_491_cast_fp16 = matmul(transpose_x = x_491_transpose_x_0, transpose_y = x_491_transpose_y_0, x = q_with_bias_v_45_cast_fp16, y = var_3859_to_fp16)[name = string("x_491_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 127]> var_3859_to_fp16 = const()[name = string("op_3859_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288328512)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_45_cast_fp16 = transpose(perm = q_with_bias_v_45_perm_0, x = var_3857_cast_fp16)[name = string("transpose_158")];
+            tensor<fp16, [1, 8, 64, 127]> x_491_cast_fp16 = matmul(transpose_x = x_491_transpose_x_0, transpose_y = x_491_transpose_y_0, x = q_with_bias_v_45_cast_fp16, y = var_3859_to_fp16)[name = string("x_491_cast_fp16")];
             tensor<int32, [8]> x_493_pad_0 = const()[name = string("x_493_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_493_mode_0 = const()[name = string("x_493_mode_0"), val = string("constant")];
             fp16 const_234_to_fp16 = const()[name = string("const_234_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_493_cast_fp16 = pad(constant_val = const_234_to_fp16, mode = x_493_mode_0, pad = x_493_pad_0, x = x_491_cast_fp16)[name = string("x_493_cast_fp16")];
-            tensor<int32, [4]> var_3867 = const()[name = string("op_3867"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_495_cast_fp16 = reshape(shape = var_3867, x = x_493_cast_fp16)[name = string("x_495_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 128]> x_493_cast_fp16 = pad(constant_val = const_234_to_fp16, mode = x_493_mode_0, pad = x_493_pad_0, x = x_491_cast_fp16)[name = string("x_493_cast_fp16")];
+            tensor<int32, [4]> var_3867 = const()[name = string("op_3867"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_495_cast_fp16 = reshape(shape = var_3867, x = x_493_cast_fp16)[name = string("x_495_cast_fp16")];
             tensor<int32, [4]> var_3871_begin_0 = const()[name = string("op_3871_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_3871_end_0 = const()[name = string("op_3871_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_3871_end_0 = const()[name = string("op_3871_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
             tensor<bool, [4]> var_3871_end_mask_0 = const()[name = string("op_3871_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_3871_cast_fp16 = slice_by_index(begin = var_3871_begin_0, end = var_3871_end_0, end_mask = var_3871_end_mask_0, x = x_495_cast_fp16)[name = string("op_3871_cast_fp16")];
-            tensor<int32, [4]> var_3872 = const()[name = string("op_3872"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_89_cast_fp16 = reshape(shape = var_3872, x = var_3871_cast_fp16)[name = string("matrix_bd_89_cast_fp16")];
+            tensor<fp16, [1, 8, 127, 64]> var_3871_cast_fp16 = slice_by_index(begin = var_3871_begin_0, end = var_3871_end_0, end_mask = var_3871_end_mask_0, x = x_495_cast_fp16)[name = string("op_3871_cast_fp16")];
+            tensor<int32, [4]> var_3872 = const()[name = string("op_3872"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_89_cast_fp16 = reshape(shape = var_3872, x = var_3871_cast_fp16)[name = string("matrix_bd_89_cast_fp16")];
             bool matrix_ac_45_transpose_x_0 = const()[name = string("matrix_ac_45_transpose_x_0"), val = bool(false)];
             bool matrix_ac_45_transpose_y_0 = const()[name = string("matrix_ac_45_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_140_perm_0 = const()[name = string("transpose_140_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_141_perm_0 = const()[name = string("transpose_141_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_141 = transpose(perm = transpose_141_perm_0, x = k_89_cast_fp16)[name = string("transpose_156")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_140 = transpose(perm = transpose_140_perm_0, x = var_3855_cast_fp16)[name = string("transpose_157")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_45_cast_fp16 = matmul(transpose_x = matrix_ac_45_transpose_x_0, transpose_y = matrix_ac_45_transpose_y_0, x = transpose_140, y = transpose_141)[name = string("matrix_ac_45_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 64]> transpose_141 = transpose(perm = transpose_141_perm_0, x = k_89_cast_fp16)[name = string("transpose_156")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_140 = transpose(perm = transpose_140_perm_0, x = var_3855_cast_fp16)[name = string("transpose_157")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_45_cast_fp16 = matmul(transpose_x = matrix_ac_45_transpose_x_0, transpose_y = matrix_ac_45_transpose_y_0, x = transpose_140, y = transpose_141)[name = string("matrix_ac_45_cast_fp16")];
             tensor<int32, [4]> matrix_bd_91_begin_0 = const()[name = string("matrix_bd_91_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_91_end_0 = const()[name = string("matrix_bd_91_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_91_end_0 = const()[name = string("matrix_bd_91_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
             tensor<bool, [4]> matrix_bd_91_end_mask_0 = const()[name = string("matrix_bd_91_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_91_cast_fp16 = slice_by_index(begin = matrix_bd_91_begin_0, end = matrix_bd_91_end_0, end_mask = matrix_bd_91_end_mask_0, x = matrix_bd_89_cast_fp16)[name = string("matrix_bd_91_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_3881_cast_fp16 = add(x = matrix_ac_45_cast_fp16, y = matrix_bd_91_cast_fp16)[name = string("op_3881_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_91_cast_fp16 = slice_by_index(begin = matrix_bd_91_begin_0, end = matrix_bd_91_end_0, end_mask = matrix_bd_91_end_mask_0, x = matrix_bd_89_cast_fp16)[name = string("matrix_bd_91_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_3881_cast_fp16 = add(x = matrix_ac_45_cast_fp16, y = matrix_bd_91_cast_fp16)[name = string("op_3881_cast_fp16")];
             fp16 _inversed_scores_89_y_0_to_fp16 = const()[name = string("_inversed_scores_89_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_89_cast_fp16 = mul(x = var_3881_cast_fp16, y = _inversed_scores_89_y_0_to_fp16)[name = string("_inversed_scores_89_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_91_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_89_cast_fp16, cond = mask_3)[name = string("scores_91_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_3887_cast_fp16 = softmax(axis = var_30, x = scores_91_cast_fp16)[name = string("op_3887_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_1177_cast_fp16 = select(a = var_11_to_fp16, b = var_3887_cast_fp16, cond = mask_3)[name = string("input_1177_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_89_cast_fp16 = mul(x = var_3881_cast_fp16, y = _inversed_scores_89_y_0_to_fp16)[name = string("_inversed_scores_89_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_91_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_89_cast_fp16, cond = mask_3)[name = string("scores_91_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_3887_cast_fp16 = softmax(axis = var_30, x = scores_91_cast_fp16)[name = string("op_3887_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_1177_cast_fp16 = select(a = var_11_to_fp16, b = var_3887_cast_fp16, cond = mask_3)[name = string("input_1177_cast_fp16")];
             bool x_497_transpose_x_0 = const()[name = string("x_497_transpose_x_0"), val = bool(false)];
             bool x_497_transpose_y_0 = const()[name = string("x_497_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_47_cast_fp16 = transpose(perm = value_47_perm_0, x = v_45_cast_fp16)[name = string("transpose_155")];
-            tensor<fp16, [1, 8, 188, 128]> x_497_cast_fp16 = matmul(transpose_x = x_497_transpose_x_0, transpose_y = x_497_transpose_y_0, x = input_1177_cast_fp16, y = value_47_cast_fp16)[name = string("x_497_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 128]> value_47_cast_fp16 = transpose(perm = value_47_perm_0, x = v_45_cast_fp16)[name = string("transpose_155")];
+            tensor<fp16, [1, 8, 64, 128]> x_497_cast_fp16 = matmul(transpose_x = x_497_transpose_x_0, transpose_y = x_497_transpose_y_0, x = input_1177_cast_fp16, y = value_47_cast_fp16)[name = string("x_497_cast_fp16")];
             tensor<int32, [4]> var_3891_perm_0 = const()[name = string("op_3891_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_3892 = const()[name = string("op_3892"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_3891_cast_fp16 = transpose(perm = var_3891_perm_0, x = x_497_cast_fp16)[name = string("transpose_154")];
-            tensor<fp16, [1, 188, 1024]> input_1179_cast_fp16 = reshape(shape = var_3892, x = var_3891_cast_fp16)[name = string("input_1179_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523989]> module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300270976))), indices_nonzero_data = tensor<uint6, [523989]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300402112))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300795200))))[name = string("module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> var_3891_cast_fp16 = transpose(perm = var_3891_perm_0, x = x_497_cast_fp16)[name = string("transpose_154")];
+            tensor<fp16, [1, 64, 1024]> input_1179_cast_fp16 = reshape(shape = var_3892, x = var_3891_cast_fp16)[name = string("input_1179_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [523989]> module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288588672))), indices_nonzero_data = tensor<uint6, [523989]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288719808))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289112896))))[name = string("module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_205_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_1179_cast_fp16)[name = string("linear_205_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_1183_cast_fp16 = add(x = input_1175_cast_fp16, y = linear_205_cast_fp16)[name = string("input_1183_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_205_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_1179_cast_fp16)[name = string("linear_205_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_1183_cast_fp16 = add(x = input_1175_cast_fp16, y = linear_205_cast_fp16)[name = string("input_1183_cast_fp16")];
             tensor<int32, [1]> x_501_axes_0 = const()[name = string("x_501_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_22_norm_conv_weight_to_fp16 = const()[name = string("module_layers_22_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300811648)))];
-            tensor<fp16, [1024]> module_layers_22_norm_conv_bias_to_fp16 = const()[name = string("module_layers_22_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300813760)))];
-            tensor<fp16, [1, 188, 1024]> x_501_cast_fp16 = layer_norm(axes = x_501_axes_0, beta = module_layers_22_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_conv_weight_to_fp16, x = input_1183_cast_fp16)[name = string("x_501_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_22_norm_conv_weight_to_fp16 = const()[name = string("module_layers_22_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289129344)))];
+            tensor<fp16, [1024]> module_layers_22_norm_conv_bias_to_fp16 = const()[name = string("module_layers_22_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289131456)))];
+            tensor<fp16, [1, 64, 1024]> x_501_cast_fp16 = layer_norm(axes = x_501_axes_0, beta = module_layers_22_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_conv_weight_to_fp16, x = input_1183_cast_fp16)[name = string("x_501_cast_fp16")];
             tensor<int32, [3]> input_1185_perm_0 = const()[name = string("input_1185_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_1187_pad_type_0 = const()[name = string("input_1187_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_1187_strides_0 = const()[name = string("input_1187_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_1187_pad_0 = const()[name = string("input_1187_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_1187_dilations_0 = const()[name = string("input_1187_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_1187_groups_0 = const()[name = string("input_1187_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1048742]> module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300815872))), indices_nonzero_data = tensor<uint6, [1048742]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301078080))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301864704))))[name = string("module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1048742]> module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289133568))), indices_nonzero_data = tensor<uint6, [1048742]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289395776))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290182400))))[name = string("module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_1185_cast_fp16 = transpose(perm = input_1185_perm_0, x = x_501_cast_fp16)[name = string("transpose_153")];
-            tensor<fp16, [1, 2048, 188]> input_1187_cast_fp16 = conv(dilations = input_1187_dilations_0, groups = input_1187_groups_0, pad = input_1187_pad_0, pad_type = input_1187_pad_type_0, strides = input_1187_strides_0, weight = module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_1185_cast_fp16)[name = string("input_1187_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_1185_cast_fp16 = transpose(perm = input_1185_perm_0, x = x_501_cast_fp16)[name = string("transpose_153")];
+            tensor<fp16, [1, 2048, 64]> input_1187_cast_fp16 = conv(dilations = input_1187_dilations_0, groups = input_1187_groups_0, pad = input_1187_pad_0, pad_type = input_1187_pad_type_0, strides = input_1187_strides_0, weight = module_layers_22_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_1185_cast_fp16)[name = string("input_1187_cast_fp16")];
             int32 x_503_split_num_splits_0 = const()[name = string("x_503_split_num_splits_0"), val = int32(2)];
             int32 x_503_split_axis_0 = const()[name = string("x_503_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_503_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_503_split_cast_fp16_1 = split(axis = x_503_split_axis_0, num_splits = x_503_split_num_splits_0, x = input_1187_cast_fp16)[name = string("x_503_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_503_split_1_sigmoid_cast_fp16 = sigmoid(x = x_503_split_cast_fp16_1)[name = string("x_503_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_503_cast_fp16 = mul(x = x_503_split_cast_fp16_0, y = x_503_split_1_sigmoid_cast_fp16)[name = string("x_503_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_1189_cast_fp16 = select(a = var_11_to_fp16, b = x_503_cast_fp16, cond = var_328)[name = string("input_1189_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_503_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_503_split_cast_fp16_1 = split(axis = x_503_split_axis_0, num_splits = x_503_split_num_splits_0, x = input_1187_cast_fp16)[name = string("x_503_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_503_split_1_sigmoid_cast_fp16 = sigmoid(x = x_503_split_cast_fp16_1)[name = string("x_503_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_503_cast_fp16 = mul(x = x_503_split_cast_fp16_0, y = x_503_split_1_sigmoid_cast_fp16)[name = string("x_503_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_1189_cast_fp16 = select(a = var_11_to_fp16, b = x_503_cast_fp16, cond = var_328)[name = string("input_1189_cast_fp16")];
             tensor<int32, [6]> input_1191_pad_0 = const()[name = string("input_1191_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_1191_mode_0 = const()[name = string("input_1191_mode_0"), val = string("constant")];
             fp16 const_237_to_fp16 = const()[name = string("const_237_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_1191_cast_fp16 = pad(constant_val = const_237_to_fp16, mode = input_1191_mode_0, pad = input_1191_pad_0, x = input_1189_cast_fp16)[name = string("input_1191_cast_fp16")];
+            tensor<fp16, [1, 1024, 72]> input_1191_cast_fp16 = pad(constant_val = const_237_to_fp16, mode = input_1191_mode_0, pad = input_1191_pad_0, x = input_1189_cast_fp16)[name = string("input_1191_cast_fp16")];
             string input_1193_pad_type_0 = const()[name = string("input_1193_pad_type_0"), val = string("valid")];
             int32 input_1193_groups_0 = const()[name = string("input_1193_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_1193_strides_0 = const()[name = string("input_1193_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_1193_pad_0 = const()[name = string("input_1193_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_1193_dilations_0 = const()[name = string("input_1193_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_292_to_fp16 = const()[name = string("const_292_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301897536)))];
-            tensor<fp16, [1024]> const_293_to_fp16 = const()[name = string("const_293_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301916032)))];
-            tensor<fp16, [1, 1024, 188]> input_1195_cast_fp16 = conv(bias = const_293_to_fp16, dilations = input_1193_dilations_0, groups = input_1193_groups_0, pad = input_1193_pad_0, pad_type = input_1193_pad_type_0, strides = input_1193_strides_0, weight = const_292_to_fp16, x = input_1191_cast_fp16)[name = string("input_1195_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_1197_cast_fp16 = silu(x = input_1195_cast_fp16)[name = string("input_1197_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_292_to_fp16 = const()[name = string("const_292_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290215232)))];
+            tensor<fp16, [1024]> const_293_to_fp16 = const()[name = string("const_293_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290233728)))];
+            tensor<fp16, [1, 1024, 64]> input_1195_cast_fp16 = conv(bias = const_293_to_fp16, dilations = input_1193_dilations_0, groups = input_1193_groups_0, pad = input_1193_pad_0, pad_type = input_1193_pad_type_0, strides = input_1193_strides_0, weight = const_292_to_fp16, x = input_1191_cast_fp16)[name = string("input_1195_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_1197_cast_fp16 = silu(x = input_1195_cast_fp16)[name = string("input_1197_cast_fp16")];
             string x_505_pad_type_0 = const()[name = string("x_505_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_505_strides_0 = const()[name = string("x_505_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_505_pad_0 = const()[name = string("x_505_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_505_dilations_0 = const()[name = string("x_505_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_505_groups_0 = const()[name = string("x_505_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [518537]> module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301918144))), indices_nonzero_data = tensor<uint6, [518537]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302049280))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302438272))))[name = string("module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [518537]> module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290235840))), indices_nonzero_data = tensor<uint6, [518537]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290366976))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290755968))))[name = string("module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_505_cast_fp16 = conv(dilations = x_505_dilations_0, groups = x_505_groups_0, pad = x_505_pad_0, pad_type = x_505_pad_type_0, strides = x_505_strides_0, weight = module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_1197_cast_fp16)[name = string("x_505_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_505_cast_fp16 = conv(dilations = x_505_dilations_0, groups = x_505_groups_0, pad = x_505_pad_0, pad_type = x_505_pad_type_0, strides = x_505_strides_0, weight = module_layers_22_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_1197_cast_fp16)[name = string("x_505_cast_fp16")];
             tensor<int32, [3]> input_1199_perm_0 = const()[name = string("input_1199_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_1199_cast_fp16 = transpose(perm = input_1199_perm_0, x = x_505_cast_fp16)[name = string("transpose_152")];
-            tensor<fp16, [1, 188, 1024]> input_1201_cast_fp16 = add(x = input_1183_cast_fp16, y = input_1199_cast_fp16)[name = string("input_1201_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_1199_cast_fp16 = transpose(perm = input_1199_perm_0, x = x_505_cast_fp16)[name = string("transpose_152")];
+            tensor<fp16, [1, 64, 1024]> input_1201_cast_fp16 = add(x = input_1183_cast_fp16, y = input_1199_cast_fp16)[name = string("input_1201_cast_fp16")];
             tensor<int32, [1]> input_1203_axes_0 = const()[name = string("input_1203_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_22_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_22_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302454720)))];
-            tensor<fp16, [1024]> module_layers_22_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_22_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302456832)))];
-            tensor<fp16, [1, 188, 1024]> input_1203_cast_fp16 = layer_norm(axes = input_1203_axes_0, beta = module_layers_22_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_feed_forward2_weight_to_fp16, x = input_1201_cast_fp16)[name = string("input_1203_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2096859]> module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302458944))), indices_nonzero_data = tensor<uint6, [2096859]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302983296))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(304556032))))[name = string("module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_22_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_22_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290772416)))];
+            tensor<fp16, [1024]> module_layers_22_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_22_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290774528)))];
+            tensor<fp16, [1, 64, 1024]> input_1203_cast_fp16 = layer_norm(axes = input_1203_axes_0, beta = module_layers_22_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_feed_forward2_weight_to_fp16, x = input_1201_cast_fp16)[name = string("input_1203_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2096859]> module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290776640))), indices_nonzero_data = tensor<uint6, [2096859]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(291300992))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292873728))))[name = string("module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_206_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_1203_cast_fp16)[name = string("linear_206_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_1207_cast_fp16 = silu(x = linear_206_cast_fp16)[name = string("input_1207_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1916371]> module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(304621632))), indices_nonzero_data = tensor<uint6, [1916371]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305145984))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306583360))))[name = string("module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 4096]> linear_206_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_22_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_1203_cast_fp16)[name = string("linear_206_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_1207_cast_fp16 = silu(x = linear_206_cast_fp16)[name = string("input_1207_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1916371]> module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292939328))), indices_nonzero_data = tensor<uint6, [1916371]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293463680))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294901056))))[name = string("module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_207_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_1207_cast_fp16)[name = string("linear_207_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_207_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_22_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_1207_cast_fp16)[name = string("linear_207_cast_fp16")];
             fp16 var_3952_to_fp16 = const()[name = string("op_3952_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_3953_cast_fp16 = mul(x = linear_207_cast_fp16, y = var_3952_to_fp16)[name = string("op_3953_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_1213_cast_fp16 = add(x = input_1201_cast_fp16, y = var_3953_cast_fp16)[name = string("input_1213_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> var_3953_cast_fp16 = mul(x = linear_207_cast_fp16, y = var_3952_to_fp16)[name = string("op_3953_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_1213_cast_fp16 = add(x = input_1201_cast_fp16, y = var_3953_cast_fp16)[name = string("input_1213_cast_fp16")];
             tensor<int32, [1]> input_1215_axes_0 = const()[name = string("input_1215_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_22_norm_out_weight_to_fp16 = const()[name = string("module_layers_22_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306599808)))];
-            tensor<fp16, [1024]> module_layers_22_norm_out_bias_to_fp16 = const()[name = string("module_layers_22_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306601920)))];
-            tensor<fp16, [1, 188, 1024]> input_1215_cast_fp16 = layer_norm(axes = input_1215_axes_0, beta = module_layers_22_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_out_weight_to_fp16, x = input_1213_cast_fp16)[name = string("input_1215_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_22_norm_out_weight_to_fp16 = const()[name = string("module_layers_22_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294917504)))];
+            tensor<fp16, [1024]> module_layers_22_norm_out_bias_to_fp16 = const()[name = string("module_layers_22_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294919616)))];
+            tensor<fp16, [1, 64, 1024]> input_1215_cast_fp16 = layer_norm(axes = input_1215_axes_0, beta = module_layers_22_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_out_weight_to_fp16, x = input_1213_cast_fp16)[name = string("input_1215_cast_fp16")];
             tensor<int32, [1]> input_1217_axes_0 = const()[name = string("input_1217_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_23_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_23_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306604032)))];
-            tensor<fp16, [1024]> module_layers_23_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_23_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306606144)))];
-            tensor<fp16, [1, 188, 1024]> input_1217_cast_fp16 = layer_norm(axes = input_1217_axes_0, beta = module_layers_23_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_feed_forward1_weight_to_fp16, x = input_1215_cast_fp16)[name = string("input_1217_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2102731]> module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306608256))), indices_nonzero_data = tensor<uint6, [2102731]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307132608))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(308709760))))[name = string("module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_23_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_23_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294921728)))];
+            tensor<fp16, [1024]> module_layers_23_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_23_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294923840)))];
+            tensor<fp16, [1, 64, 1024]> input_1217_cast_fp16 = layer_norm(axes = input_1217_axes_0, beta = module_layers_23_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_feed_forward1_weight_to_fp16, x = input_1215_cast_fp16)[name = string("input_1217_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2102731]> module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294925952))), indices_nonzero_data = tensor<uint6, [2102731]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295450304))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297027456))))[name = string("module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_208_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_1217_cast_fp16)[name = string("linear_208_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_1221_cast_fp16 = silu(x = linear_208_cast_fp16)[name = string("input_1221_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1967096]> module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(308775360))), indices_nonzero_data = tensor<uint6, [1967096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(309299712))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(310775104))))[name = string("module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 4096]> linear_208_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_23_feed_forward1_linear1_weight_to_fp16_fivevalue, x = input_1217_cast_fp16)[name = string("linear_208_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_1221_cast_fp16 = silu(x = linear_208_cast_fp16)[name = string("input_1221_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1967096]> module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297093056))), indices_nonzero_data = tensor<uint6, [1967096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297617408))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299092800))))[name = string("module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_209_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_1221_cast_fp16)[name = string("linear_209_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_209_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_feed_forward1_linear2_weight_to_fp16_fivevalue, x = input_1221_cast_fp16)[name = string("linear_209_cast_fp16")];
             fp16 var_3981_to_fp16 = const()[name = string("op_3981_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_3982_cast_fp16 = mul(x = linear_209_cast_fp16, y = var_3981_to_fp16)[name = string("op_3982_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_1227_cast_fp16 = add(x = input_1215_cast_fp16, y = var_3982_cast_fp16)[name = string("input_1227_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> var_3982_cast_fp16 = mul(x = linear_209_cast_fp16, y = var_3981_to_fp16)[name = string("op_3982_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_1227_cast_fp16 = add(x = input_1215_cast_fp16, y = var_3982_cast_fp16)[name = string("input_1227_cast_fp16")];
             tensor<int32, [1]> query_axes_0 = const()[name = string("query_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_23_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_23_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(310791552)))];
-            tensor<fp16, [1024]> module_layers_23_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_23_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(310793664)))];
-            tensor<fp16, [1, 188, 1024]> query_cast_fp16 = layer_norm(axes = query_axes_0, beta = module_layers_23_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_self_att_weight_to_fp16, x = input_1227_cast_fp16)[name = string("query_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [526594]> module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(310795776))), indices_nonzero_data = tensor<uint6, [526594]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(310926912))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311321984))))[name = string("module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_23_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_23_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299109248)))];
+            tensor<fp16, [1024]> module_layers_23_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_23_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299111360)))];
+            tensor<fp16, [1, 64, 1024]> query_cast_fp16 = layer_norm(axes = query_axes_0, beta = module_layers_23_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_self_att_weight_to_fp16, x = input_1227_cast_fp16)[name = string("query_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [526594]> module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299113472))), indices_nonzero_data = tensor<uint6, [526594]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299244608))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299639680))))[name = string("module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_210_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_cast_fp16)[name = string("linear_210_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_210_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_self_attn_linear_q_weight_to_fp16_fivevalue, x = query_cast_fp16)[name = string("linear_210_cast_fp16")];
             tensor<int32, [4]> var_3998 = const()[name = string("op_3998"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> q_139_cast_fp16 = reshape(shape = var_3998, x = linear_210_cast_fp16)[name = string("q_139_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524871]> module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311338432))), indices_nonzero_data = tensor<uint6, [524871]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311469568))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311863296))))[name = string("module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> q_139_cast_fp16 = reshape(shape = var_3998, x = linear_210_cast_fp16)[name = string("q_139_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [524871]> module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299656128))), indices_nonzero_data = tensor<uint6, [524871]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299787264))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300180992))))[name = string("module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_211_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_cast_fp16)[name = string("linear_211_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_211_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_self_attn_linear_k_weight_to_fp16_fivevalue, x = query_cast_fp16)[name = string("linear_211_cast_fp16")];
             tensor<int32, [4]> var_4002 = const()[name = string("op_4002"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> k_93_cast_fp16 = reshape(shape = var_4002, x = linear_211_cast_fp16)[name = string("k_93_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525256]> module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311879744))), indices_nonzero_data = tensor<uint6, [525256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312010880))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312404928))))[name = string("module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> k_93_cast_fp16 = reshape(shape = var_4002, x = linear_211_cast_fp16)[name = string("k_93_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [525256]> module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300197440))), indices_nonzero_data = tensor<uint6, [525256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300328576))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300722624))))[name = string("module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_212_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_cast_fp16)[name = string("linear_212_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_212_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_self_attn_linear_v_weight_to_fp16_fivevalue, x = query_cast_fp16)[name = string("linear_212_cast_fp16")];
             tensor<int32, [4]> var_4006 = const()[name = string("op_4006"), val = tensor<int32, [4]>([1, -1, 8, 128])];
-            tensor<fp16, [1, 188, 8, 128]> v_cast_fp16 = reshape(shape = var_4006, x = linear_212_cast_fp16)[name = string("v_cast_fp16")];
+            tensor<fp16, [1, 64, 8, 128]> v_cast_fp16 = reshape(shape = var_4006, x = linear_212_cast_fp16)[name = string("v_cast_fp16")];
             tensor<int32, [4]> value_perm_0 = const()[name = string("value_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<fp16, [8, 128]> module_layers_23_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_23_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312421376)))];
-            tensor<fp16, [1, 188, 8, 128]> var_4018_cast_fp16 = add(x = q_139_cast_fp16, y = module_layers_23_self_attn_pos_bias_u_to_fp16)[name = string("op_4018_cast_fp16")];
-            tensor<fp16, [8, 128]> module_layers_23_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_23_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312423488)))];
-            tensor<fp16, [1, 188, 8, 128]> var_4020_cast_fp16 = add(x = q_139_cast_fp16, y = module_layers_23_self_attn_pos_bias_v_to_fp16)[name = string("op_4020_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_23_self_attn_pos_bias_u_to_fp16 = const()[name = string("module_layers_23_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300739072)))];
+            tensor<fp16, [1, 64, 8, 128]> var_4018_cast_fp16 = add(x = q_139_cast_fp16, y = module_layers_23_self_attn_pos_bias_u_to_fp16)[name = string("op_4018_cast_fp16")];
+            tensor<fp16, [8, 128]> module_layers_23_self_attn_pos_bias_v_to_fp16 = const()[name = string("module_layers_23_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300741184)))];
+            tensor<fp16, [1, 64, 8, 128]> var_4020_cast_fp16 = add(x = q_139_cast_fp16, y = module_layers_23_self_attn_pos_bias_v_to_fp16)[name = string("op_4020_cast_fp16")];
             tensor<int32, [4]> q_with_bias_v_perm_0 = const()[name = string("q_with_bias_v_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             bool x_513_transpose_x_0 = const()[name = string("x_513_transpose_x_0"), val = bool(false)];
             bool x_513_transpose_y_0 = const()[name = string("x_513_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 128, 375]> var_4022_to_fp16 = const()[name = string("op_4022_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312425600)))];
-            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_cast_fp16 = transpose(perm = q_with_bias_v_perm_0, x = var_4020_cast_fp16)[name = string("transpose_151")];
-            tensor<fp16, [1, 8, 188, 375]> x_513_cast_fp16 = matmul(transpose_x = x_513_transpose_x_0, transpose_y = x_513_transpose_y_0, x = q_with_bias_v_cast_fp16, y = var_4022_to_fp16)[name = string("x_513_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 127]> var_4022_to_fp16 = const()[name = string("op_4022_to_fp16"), val = tensor<fp16, [1, 8, 128, 127]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300743296)))];
+            tensor<fp16, [1, 8, 64, 128]> q_with_bias_v_cast_fp16 = transpose(perm = q_with_bias_v_perm_0, x = var_4020_cast_fp16)[name = string("transpose_151")];
+            tensor<fp16, [1, 8, 64, 127]> x_513_cast_fp16 = matmul(transpose_x = x_513_transpose_x_0, transpose_y = x_513_transpose_y_0, x = q_with_bias_v_cast_fp16, y = var_4022_to_fp16)[name = string("x_513_cast_fp16")];
             tensor<int32, [8]> x_515_pad_0 = const()[name = string("x_515_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
             string x_515_mode_0 = const()[name = string("x_515_mode_0"), val = string("constant")];
             fp16 const_244_to_fp16 = const()[name = string("const_244_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 8, 188, 376]> x_515_cast_fp16 = pad(constant_val = const_244_to_fp16, mode = x_515_mode_0, pad = x_515_pad_0, x = x_513_cast_fp16)[name = string("x_515_cast_fp16")];
-            tensor<int32, [4]> var_4030 = const()[name = string("op_4030"), val = tensor<int32, [4]>([1, 8, -1, 188])];
-            tensor<fp16, [1, 8, 376, 188]> x_517_cast_fp16 = reshape(shape = var_4030, x = x_515_cast_fp16)[name = string("x_517_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 128]> x_515_cast_fp16 = pad(constant_val = const_244_to_fp16, mode = x_515_mode_0, pad = x_515_pad_0, x = x_513_cast_fp16)[name = string("x_515_cast_fp16")];
+            tensor<int32, [4]> var_4030 = const()[name = string("op_4030"), val = tensor<int32, [4]>([1, 8, -1, 64])];
+            tensor<fp16, [1, 8, 128, 64]> x_517_cast_fp16 = reshape(shape = var_4030, x = x_515_cast_fp16)[name = string("x_517_cast_fp16")];
             tensor<int32, [4]> var_4034_begin_0 = const()[name = string("op_4034_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
-            tensor<int32, [4]> var_4034_end_0 = const()[name = string("op_4034_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
+            tensor<int32, [4]> var_4034_end_0 = const()[name = string("op_4034_end_0"), val = tensor<int32, [4]>([1, 8, 128, 64])];
             tensor<bool, [4]> var_4034_end_mask_0 = const()[name = string("op_4034_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
-            tensor<fp16, [1, 8, 375, 188]> var_4034_cast_fp16 = slice_by_index(begin = var_4034_begin_0, end = var_4034_end_0, end_mask = var_4034_end_mask_0, x = x_517_cast_fp16)[name = string("op_4034_cast_fp16")];
-            tensor<int32, [4]> var_4035 = const()[name = string("op_4035"), val = tensor<int32, [4]>([1, 8, 188, 375])];
-            tensor<fp16, [1, 8, 188, 375]> matrix_bd_93_cast_fp16 = reshape(shape = var_4035, x = var_4034_cast_fp16)[name = string("matrix_bd_93_cast_fp16")];
+            tensor<fp16, [1, 8, 127, 64]> var_4034_cast_fp16 = slice_by_index(begin = var_4034_begin_0, end = var_4034_end_0, end_mask = var_4034_end_mask_0, x = x_517_cast_fp16)[name = string("op_4034_cast_fp16")];
+            tensor<int32, [4]> var_4035 = const()[name = string("op_4035"), val = tensor<int32, [4]>([1, 8, 64, 127])];
+            tensor<fp16, [1, 8, 64, 127]> matrix_bd_93_cast_fp16 = reshape(shape = var_4035, x = var_4034_cast_fp16)[name = string("matrix_bd_93_cast_fp16")];
             bool matrix_ac_transpose_x_0 = const()[name = string("matrix_ac_transpose_x_0"), val = bool(false)];
             bool matrix_ac_transpose_y_0 = const()[name = string("matrix_ac_transpose_y_0"), val = bool(false)];
             tensor<int32, [4]> transpose_142_perm_0 = const()[name = string("transpose_142_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
             tensor<int32, [4]> transpose_143_perm_0 = const()[name = string("transpose_143_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 128, 188]> transpose_143 = transpose(perm = transpose_143_perm_0, x = k_93_cast_fp16)[name = string("transpose_149")];
-            tensor<fp16, [1, 8, 188, 128]> transpose_142 = transpose(perm = transpose_142_perm_0, x = var_4018_cast_fp16)[name = string("transpose_150")];
-            tensor<fp16, [1, 8, 188, 188]> matrix_ac_cast_fp16 = matmul(transpose_x = matrix_ac_transpose_x_0, transpose_y = matrix_ac_transpose_y_0, x = transpose_142, y = transpose_143)[name = string("matrix_ac_cast_fp16")];
+            tensor<fp16, [1, 8, 128, 64]> transpose_143 = transpose(perm = transpose_143_perm_0, x = k_93_cast_fp16)[name = string("transpose_149")];
+            tensor<fp16, [1, 8, 64, 128]> transpose_142 = transpose(perm = transpose_142_perm_0, x = var_4018_cast_fp16)[name = string("transpose_150")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_ac_cast_fp16 = matmul(transpose_x = matrix_ac_transpose_x_0, transpose_y = matrix_ac_transpose_y_0, x = transpose_142, y = transpose_143)[name = string("matrix_ac_cast_fp16")];
             tensor<int32, [4]> matrix_bd_begin_0 = const()[name = string("matrix_bd_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
-            tensor<int32, [4]> matrix_bd_end_0 = const()[name = string("matrix_bd_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
+            tensor<int32, [4]> matrix_bd_end_0 = const()[name = string("matrix_bd_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
             tensor<bool, [4]> matrix_bd_end_mask_0 = const()[name = string("matrix_bd_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
-            tensor<fp16, [1, 8, 188, 188]> matrix_bd_cast_fp16 = slice_by_index(begin = matrix_bd_begin_0, end = matrix_bd_end_0, end_mask = matrix_bd_end_mask_0, x = matrix_bd_93_cast_fp16)[name = string("matrix_bd_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_4044_cast_fp16 = add(x = matrix_ac_cast_fp16, y = matrix_bd_cast_fp16)[name = string("op_4044_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> matrix_bd_cast_fp16 = slice_by_index(begin = matrix_bd_begin_0, end = matrix_bd_end_0, end_mask = matrix_bd_end_mask_0, x = matrix_bd_93_cast_fp16)[name = string("matrix_bd_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_4044_cast_fp16 = add(x = matrix_ac_cast_fp16, y = matrix_bd_cast_fp16)[name = string("op_4044_cast_fp16")];
             fp16 _inversed_scores_93_y_0_to_fp16 = const()[name = string("_inversed_scores_93_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
-            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_93_cast_fp16 = mul(x = var_4044_cast_fp16, y = _inversed_scores_93_y_0_to_fp16)[name = string("_inversed_scores_93_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> scores_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_93_cast_fp16, cond = mask_3)[name = string("scores_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> var_4050_cast_fp16 = softmax(axis = var_30, x = scores_cast_fp16)[name = string("op_4050_cast_fp16")];
-            tensor<fp16, [1, 8, 188, 188]> input_1229_cast_fp16 = select(a = var_11_to_fp16, b = var_4050_cast_fp16, cond = mask_3)[name = string("input_1229_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> _inversed_scores_93_cast_fp16 = mul(x = var_4044_cast_fp16, y = _inversed_scores_93_y_0_to_fp16)[name = string("_inversed_scores_93_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> scores_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_93_cast_fp16, cond = mask_3)[name = string("scores_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> var_4050_cast_fp16 = softmax(axis = var_30, x = scores_cast_fp16)[name = string("op_4050_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 64]> input_1229_cast_fp16 = select(a = var_11_to_fp16, b = var_4050_cast_fp16, cond = mask_3)[name = string("input_1229_cast_fp16")];
             bool x_519_transpose_x_0 = const()[name = string("x_519_transpose_x_0"), val = bool(false)];
             bool x_519_transpose_y_0 = const()[name = string("x_519_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 188, 128]> value_cast_fp16 = transpose(perm = value_perm_0, x = v_cast_fp16)[name = string("transpose_148")];
-            tensor<fp16, [1, 8, 188, 128]> x_519_cast_fp16 = matmul(transpose_x = x_519_transpose_x_0, transpose_y = x_519_transpose_y_0, x = input_1229_cast_fp16, y = value_cast_fp16)[name = string("x_519_cast_fp16")];
+            tensor<fp16, [1, 8, 64, 128]> value_cast_fp16 = transpose(perm = value_perm_0, x = v_cast_fp16)[name = string("transpose_148")];
+            tensor<fp16, [1, 8, 64, 128]> x_519_cast_fp16 = matmul(transpose_x = x_519_transpose_x_0, transpose_y = x_519_transpose_y_0, x = input_1229_cast_fp16, y = value_cast_fp16)[name = string("x_519_cast_fp16")];
             tensor<int32, [4]> var_4054_perm_0 = const()[name = string("op_4054_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
             tensor<int32, [3]> var_4055 = const()[name = string("op_4055"), val = tensor<int32, [3]>([1, -1, 1024])];
-            tensor<fp16, [1, 188, 8, 128]> var_4054_cast_fp16 = transpose(perm = var_4054_perm_0, x = x_519_cast_fp16)[name = string("transpose_147")];
-            tensor<fp16, [1, 188, 1024]> input_1231_cast_fp16 = reshape(shape = var_4055, x = var_4054_cast_fp16)[name = string("input_1231_cast_fp16")];
-            tensor<uint1, [1024, 1024]> module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [527827]> module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313193664))), indices_nonzero_data = tensor<uint6, [527827]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313324800))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313720768))))[name = string("module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 8, 128]> var_4054_cast_fp16 = transpose(perm = var_4054_perm_0, x = x_519_cast_fp16)[name = string("transpose_147")];
+            tensor<fp16, [1, 64, 1024]> input_1231_cast_fp16 = reshape(shape = var_4055, x = var_4054_cast_fp16)[name = string("input_1231_cast_fp16")];
+            tensor<uint1, [1024, 1024]> module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [527827]> module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301003456))), indices_nonzero_data = tensor<uint6, [527827]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301134592))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301530560))))[name = string("module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024]> module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_214_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_1231_cast_fp16)[name = string("linear_214_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_1235_cast_fp16 = add(x = input_1227_cast_fp16, y = linear_214_cast_fp16)[name = string("input_1235_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_214_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_self_attn_linear_out_weight_to_fp16_fivevalue, x = input_1231_cast_fp16)[name = string("linear_214_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_1235_cast_fp16 = add(x = input_1227_cast_fp16, y = linear_214_cast_fp16)[name = string("input_1235_cast_fp16")];
             tensor<int32, [1]> x_523_axes_0 = const()[name = string("x_523_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_23_norm_conv_weight_to_fp16 = const()[name = string("module_layers_23_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313737216)))];
-            tensor<fp16, [1024]> module_layers_23_norm_conv_bias_to_fp16 = const()[name = string("module_layers_23_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313739328)))];
-            tensor<fp16, [1, 188, 1024]> x_523_cast_fp16 = layer_norm(axes = x_523_axes_0, beta = module_layers_23_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_conv_weight_to_fp16, x = input_1235_cast_fp16)[name = string("x_523_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_23_norm_conv_weight_to_fp16 = const()[name = string("module_layers_23_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301547008)))];
+            tensor<fp16, [1024]> module_layers_23_norm_conv_bias_to_fp16 = const()[name = string("module_layers_23_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301549120)))];
+            tensor<fp16, [1, 64, 1024]> x_523_cast_fp16 = layer_norm(axes = x_523_axes_0, beta = module_layers_23_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_conv_weight_to_fp16, x = input_1235_cast_fp16)[name = string("x_523_cast_fp16")];
             tensor<int32, [3]> input_1237_perm_0 = const()[name = string("input_1237_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string input_1239_pad_type_0 = const()[name = string("input_1239_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> input_1239_strides_0 = const()[name = string("input_1239_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_1239_pad_0 = const()[name = string("input_1239_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_1239_dilations_0 = const()[name = string("input_1239_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 input_1239_groups_0 = const()[name = string("input_1239_groups_0"), val = int32(1)];
-            tensor<uint1, [2048, 1024, 1]> module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1051649]> module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313741440))), indices_nonzero_data = tensor<uint6, [1051649]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314003648))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314792512))))[name = string("module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [2048, 1024, 1]> module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [1051649]> module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301551232))), indices_nonzero_data = tensor<uint6, [1051649]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301813440))), lut = tensor<fp16, [256, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302602304))))[name = string("module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [2048, 1024, 1]> module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> input_1237_cast_fp16 = transpose(perm = input_1237_perm_0, x = x_523_cast_fp16)[name = string("transpose_146")];
-            tensor<fp16, [1, 2048, 188]> input_1239_cast_fp16 = conv(dilations = input_1239_dilations_0, groups = input_1239_groups_0, pad = input_1239_pad_0, pad_type = input_1239_pad_type_0, strides = input_1239_strides_0, weight = module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_1237_cast_fp16)[name = string("input_1239_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_1237_cast_fp16 = transpose(perm = input_1237_perm_0, x = x_523_cast_fp16)[name = string("transpose_146")];
+            tensor<fp16, [1, 2048, 64]> input_1239_cast_fp16 = conv(dilations = input_1239_dilations_0, groups = input_1239_groups_0, pad = input_1239_pad_0, pad_type = input_1239_pad_type_0, strides = input_1239_strides_0, weight = module_layers_23_conv_pointwise_conv1_weight_to_fp16_fivevalue, x = input_1237_cast_fp16)[name = string("input_1239_cast_fp16")];
             int32 x_525_split_num_splits_0 = const()[name = string("x_525_split_num_splits_0"), val = int32(2)];
             int32 x_525_split_axis_0 = const()[name = string("x_525_split_axis_0"), val = int32(1)];
-            tensor<fp16, [1, 1024, 188]> x_525_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_525_split_cast_fp16_1 = split(axis = x_525_split_axis_0, num_splits = x_525_split_num_splits_0, x = input_1239_cast_fp16)[name = string("x_525_split_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_525_split_1_sigmoid_cast_fp16 = sigmoid(x = x_525_split_cast_fp16_1)[name = string("x_525_split_1_sigmoid_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> x_525_cast_fp16 = mul(x = x_525_split_cast_fp16_0, y = x_525_split_1_sigmoid_cast_fp16)[name = string("x_525_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_1241_cast_fp16 = select(a = var_11_to_fp16, b = x_525_cast_fp16, cond = var_328)[name = string("input_1241_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_525_split_cast_fp16_0, tensor<fp16, [1, 1024, 64]> x_525_split_cast_fp16_1 = split(axis = x_525_split_axis_0, num_splits = x_525_split_num_splits_0, x = input_1239_cast_fp16)[name = string("x_525_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_525_split_1_sigmoid_cast_fp16 = sigmoid(x = x_525_split_cast_fp16_1)[name = string("x_525_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_525_cast_fp16 = mul(x = x_525_split_cast_fp16_0, y = x_525_split_1_sigmoid_cast_fp16)[name = string("x_525_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_1241_cast_fp16 = select(a = var_11_to_fp16, b = x_525_cast_fp16, cond = var_328)[name = string("input_1241_cast_fp16")];
             tensor<int32, [6]> input_1243_pad_0 = const()[name = string("input_1243_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
             string input_1243_mode_0 = const()[name = string("input_1243_mode_0"), val = string("constant")];
             fp16 const_247_to_fp16 = const()[name = string("const_247_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 1024, 196]> input_1243_cast_fp16 = pad(constant_val = const_247_to_fp16, mode = input_1243_mode_0, pad = input_1243_pad_0, x = input_1241_cast_fp16)[name = string("input_1243_cast_fp16")];
+            tensor<fp16, [1, 1024, 72]> input_1243_cast_fp16 = pad(constant_val = const_247_to_fp16, mode = input_1243_mode_0, pad = input_1243_pad_0, x = input_1241_cast_fp16)[name = string("input_1243_cast_fp16")];
             string input_1245_pad_type_0 = const()[name = string("input_1245_pad_type_0"), val = string("valid")];
             int32 input_1245_groups_0 = const()[name = string("input_1245_groups_0"), val = int32(1024)];
             tensor<int32, [1]> input_1245_strides_0 = const()[name = string("input_1245_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> input_1245_pad_0 = const()[name = string("input_1245_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> input_1245_dilations_0 = const()[name = string("input_1245_dilations_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp16, [1024, 1, 9]> const_294_to_fp16 = const()[name = string("const_294_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314825344)))];
-            tensor<fp16, [1024]> const_295_to_fp16 = const()[name = string("const_295_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314843840)))];
-            tensor<fp16, [1, 1024, 188]> input_1247_cast_fp16 = conv(bias = const_295_to_fp16, dilations = input_1245_dilations_0, groups = input_1245_groups_0, pad = input_1245_pad_0, pad_type = input_1245_pad_type_0, strides = input_1245_strides_0, weight = const_294_to_fp16, x = input_1243_cast_fp16)[name = string("input_1247_cast_fp16")];
-            tensor<fp16, [1, 1024, 188]> input_1249_cast_fp16 = silu(x = input_1247_cast_fp16)[name = string("input_1249_cast_fp16")];
+            tensor<fp16, [1024, 1, 9]> const_294_to_fp16 = const()[name = string("const_294_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302635136)))];
+            tensor<fp16, [1024]> const_295_to_fp16 = const()[name = string("const_295_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302653632)))];
+            tensor<fp16, [1, 1024, 64]> input_1247_cast_fp16 = conv(bias = const_295_to_fp16, dilations = input_1245_dilations_0, groups = input_1245_groups_0, pad = input_1245_pad_0, pad_type = input_1245_pad_type_0, strides = input_1245_strides_0, weight = const_294_to_fp16, x = input_1243_cast_fp16)[name = string("input_1247_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> input_1249_cast_fp16 = silu(x = input_1247_cast_fp16)[name = string("input_1249_cast_fp16")];
             string x_527_pad_type_0 = const()[name = string("x_527_pad_type_0"), val = string("valid")];
             tensor<int32, [1]> x_527_strides_0 = const()[name = string("x_527_strides_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [2]> x_527_pad_0 = const()[name = string("x_527_pad_0"), val = tensor<int32, [2]>([0, 0])];
             tensor<int32, [1]> x_527_dilations_0 = const()[name = string("x_527_dilations_0"), val = tensor<int32, [1]>([1])];
             int32 x_527_groups_0 = const()[name = string("x_527_groups_0"), val = int32(1)];
-            tensor<uint1, [1024, 1024, 1]> module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519222]> module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314845952))), indices_nonzero_data = tensor<uint6, [519222]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314977088))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315366592))))[name = string("module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
+            tensor<uint1, [1024, 1024, 1]> module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [519222]> module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302655744))), indices_nonzero_data = tensor<uint6, [519222]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302786880))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303176384))))[name = string("module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 1024, 1]> module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 1024, 188]> x_527_cast_fp16 = conv(dilations = x_527_dilations_0, groups = x_527_groups_0, pad = x_527_pad_0, pad_type = x_527_pad_type_0, strides = x_527_strides_0, weight = module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_1249_cast_fp16)[name = string("x_527_cast_fp16")];
+            tensor<fp16, [1, 1024, 64]> x_527_cast_fp16 = conv(dilations = x_527_dilations_0, groups = x_527_groups_0, pad = x_527_pad_0, pad_type = x_527_pad_type_0, strides = x_527_strides_0, weight = module_layers_23_conv_pointwise_conv2_weight_to_fp16_fivevalue, x = input_1249_cast_fp16)[name = string("x_527_cast_fp16")];
             tensor<int32, [3]> input_1251_perm_0 = const()[name = string("input_1251_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
-            tensor<fp16, [1, 188, 1024]> input_1251_cast_fp16 = transpose(perm = input_1251_perm_0, x = x_527_cast_fp16)[name = string("transpose_145")];
-            tensor<fp16, [1, 188, 1024]> input_1253_cast_fp16 = add(x = input_1235_cast_fp16, y = input_1251_cast_fp16)[name = string("input_1253_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_1251_cast_fp16 = transpose(perm = input_1251_perm_0, x = x_527_cast_fp16)[name = string("transpose_145")];
+            tensor<fp16, [1, 64, 1024]> input_1253_cast_fp16 = add(x = input_1235_cast_fp16, y = input_1251_cast_fp16)[name = string("input_1253_cast_fp16")];
             tensor<int32, [1]> input_1255_axes_0 = const()[name = string("input_1255_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_23_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_23_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315383040)))];
-            tensor<fp16, [1024]> module_layers_23_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_23_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315385152)))];
-            tensor<fp16, [1, 188, 1024]> input_1255_cast_fp16 = layer_norm(axes = input_1255_axes_0, beta = module_layers_23_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_feed_forward2_weight_to_fp16, x = input_1253_cast_fp16)[name = string("input_1255_cast_fp16")];
-            tensor<uint1, [4096, 1024]> module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2106054]> module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315387264))), indices_nonzero_data = tensor<uint6, [2106054]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315911616))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(317491264))))[name = string("module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1024]> module_layers_23_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_23_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303192832)))];
+            tensor<fp16, [1024]> module_layers_23_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_23_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303194944)))];
+            tensor<fp16, [1, 64, 1024]> input_1255_cast_fp16 = layer_norm(axes = input_1255_axes_0, beta = module_layers_23_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_feed_forward2_weight_to_fp16, x = input_1253_cast_fp16)[name = string("input_1255_cast_fp16")];
+            tensor<uint1, [4096, 1024]> module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2106054]> module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303197056))), indices_nonzero_data = tensor<uint6, [2106054]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303721408))), lut = tensor<fp16, [512, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305301056))))[name = string("module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [4096, 1024]> module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 4096]> linear_215_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_1255_cast_fp16)[name = string("linear_215_cast_fp16")];
-            tensor<fp16, [1, 188, 4096]> input_1259_cast_fp16 = silu(x = linear_215_cast_fp16)[name = string("input_1259_cast_fp16")];
-            tensor<uint1, [1024, 4096]> module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2012639]> module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(317556864))), indices_nonzero_data = tensor<uint6, [2012639]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(318081216))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(319590784))))[name = string("module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
+            tensor<fp16, [1, 64, 4096]> linear_215_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = module_layers_23_feed_forward2_linear1_weight_to_fp16_fivevalue, x = input_1255_cast_fp16)[name = string("linear_215_cast_fp16")];
+            tensor<fp16, [1, 64, 4096]> input_1259_cast_fp16 = silu(x = linear_215_cast_fp16)[name = string("input_1259_cast_fp16")];
+            tensor<uint1, [1024, 4096]> module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, tensor<fp16, [2012639]> module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1 = constexpr_lut_to_sparse(indices_mask = tensor<uint1, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305366656))), indices_nonzero_data = tensor<uint6, [2012639]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305891008))), lut = tensor<fp16, [128, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307400576))))[name = string("module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue_lut")];
             tensor<fp16, [1024, 4096]> module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue = constexpr_sparse_to_dense(mask = module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_0, nonzero_data = module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue_lut_1)[name = string("module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue")];
-            tensor<fp16, [1, 188, 1024]> linear_216_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_1259_cast_fp16)[name = string("linear_216_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> linear_216_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = module_layers_23_feed_forward2_linear2_weight_to_fp16_fivevalue, x = input_1259_cast_fp16)[name = string("linear_216_cast_fp16")];
             fp16 var_4115_to_fp16 = const()[name = string("op_4115_to_fp16"), val = fp16(0x1p-1)];
-            tensor<fp16, [1, 188, 1024]> var_4116_cast_fp16 = mul(x = linear_216_cast_fp16, y = var_4115_to_fp16)[name = string("op_4116_cast_fp16")];
-            tensor<fp16, [1, 188, 1024]> input_cast_fp16 = add(x = input_1253_cast_fp16, y = var_4116_cast_fp16)[name = string("input_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> var_4116_cast_fp16 = mul(x = linear_216_cast_fp16, y = var_4115_to_fp16)[name = string("op_4116_cast_fp16")];
+            tensor<fp16, [1, 64, 1024]> input_cast_fp16 = add(x = input_1253_cast_fp16, y = var_4116_cast_fp16)[name = string("input_cast_fp16")];
             tensor<int32, [1]> audio_signal_axes_0 = const()[name = string("audio_signal_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1024]> module_layers_23_norm_out_weight_to_fp16 = const()[name = string("module_layers_23_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(319607232)))];
-            tensor<fp16, [1024]> module_layers_23_norm_out_bias_to_fp16 = const()[name = string("module_layers_23_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(319609344)))];
-            tensor<fp16, [1, 188, 1024]> audio_signal_cast_fp16 = layer_norm(axes = audio_signal_axes_0, beta = module_layers_23_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_out_weight_to_fp16, x = input_cast_fp16)[name = string("audio_signal_cast_fp16")];
+            tensor<fp16, [1024]> module_layers_23_norm_out_weight_to_fp16 = const()[name = string("module_layers_23_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307417024)))];
+            tensor<fp16, [1024]> module_layers_23_norm_out_bias_to_fp16 = const()[name = string("module_layers_23_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307419136)))];
+            tensor<fp16, [1, 64, 1024]> audio_signal_cast_fp16 = layer_norm(axes = audio_signal_axes_0, beta = module_layers_23_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_out_weight_to_fp16, x = input_cast_fp16)[name = string("audio_signal_cast_fp16")];
             tensor<int32, [3]> obj_1_perm_0 = const()[name = string("obj_1_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             string obj_1_cast_fp16_to_fp32_dtype_0 = const()[name = string("obj_1_cast_fp16_to_fp32_dtype_0"), val = string("fp32")];
-            tensor<fp16, [1, 1024, 188]> obj_1_cast_fp16 = transpose(perm = obj_1_perm_0, x = audio_signal_cast_fp16)[name = string("transpose_144")];
-            tensor<fp32, [1, 1024, 188]> encoder = cast(dtype = obj_1_cast_fp16_to_fp32_dtype_0, x = obj_1_cast_fp16)[name = string("cast_0")];
+            tensor<fp16, [1, 1024, 64]> obj_1_cast_fp16 = transpose(perm = obj_1_perm_0, x = audio_signal_cast_fp16)[name = string("transpose_144")];
+            tensor<fp32, [1, 1024, 64]> encoder = cast(dtype = obj_1_cast_fp16_to_fp32_dtype_0, x = obj_1_cast_fp16)[name = string("cast_0")];
         } -> (encoder, encoder_length);
 }
\ No newline at end of file