program(1.3)
[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3600.16.1"}, {"coremlc-version", "3600.25.1"}})]
{
    func main<ios18>(tensor<fp32, [1, 128, 1501]> mel, tensor<int32, [1]> mel_length) {
            tensor<int32, [3]> var_2237_perm_0 = const()[name = string("op_2237_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")];
            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_6")];
            tensor<fp16, [1, 1501, 128]> var_2237_cast_fp16 = transpose(perm = var_2237_perm_0, x = mel_to_fp16)[name = string("transpose_482")];
            tensor<fp16, [1, 1, 1501, 128]> input_1_cast_fp16 = expand_dims(axes = input_1_axes_0, x = var_2237_cast_fp16)[name = string("input_1_cast_fp16")];
            string x_1_pad_type_0 = const()[name = string("x_1_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> x_1_pad_0 = const()[name = string("x_1_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> x_1_strides_0 = const()[name = string("x_1_strides_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> x_1_dilations_0 = const()[name = string("x_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 x_1_groups_0 = const()[name = string("x_1_groups_0"), val = int32(1)];
            tensor<fp16, [256, 1, 3, 3]> enc_subsampling_layers_0_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64))), scale = tensor<fp16, [256, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2432))))[name = string("enc_subsampling_layers_0_weight_to_fp16_quantized")];
            tensor<fp16, [256]> enc_subsampling_layers_0_bias_to_fp16 = const()[name = string("enc_subsampling_layers_0_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3008)))];
            tensor<fp16, [1, 256, 751, 64]> x_1_cast_fp16 = conv(bias = enc_subsampling_layers_0_bias_to_fp16, dilations = x_1_dilations_0, groups = x_1_groups_0, pad = x_1_pad_0, pad_type = x_1_pad_type_0, strides = x_1_strides_0, weight = enc_subsampling_layers_0_weight_to_fp16_quantized, x = input_1_cast_fp16)[name = string("x_1_cast_fp16")];
            int32 var_2253 = const()[name = string("op_2253"), val = int32(2)];
            tensor<int32, [1]> var_2254 = add(x = mel_length, y = var_2253)[name = string("op_2254")];
            int32 var_2256 = const()[name = string("op_2256"), val = int32(3)];
            tensor<int32, [1]> var_2257 = sub(x = var_2254, y = var_2256)[name = string("op_2257")];
            string cast_0_to_fp16_dtype_0 = const()[name = string("cast_0_to_fp16_dtype_0"), val = string("fp16")];
            fp16 cast_1_to_fp16 = const()[name = string("cast_1_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1]> var_2257_to_fp16 = cast(dtype = cast_0_to_fp16_dtype_0, x = var_2257)[name = string("cast_5")];
            tensor<fp16, [1]> var_2260_cast_fp16 = floor_div(x = var_2257_to_fp16, y = cast_1_to_fp16)[name = string("op_2260_cast_fp16")];
            fp16 var_2262_promoted_to_fp16 = const()[name = string("op_2262_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> length_1_cast_fp16 = add(x = var_2260_cast_fp16, y = var_2262_promoted_to_fp16)[name = string("length_1_cast_fp16")];
            tensor<int32, [2]> var_2274 = const()[name = string("op_2274"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1, 1]> var_2275_cast_fp16 = reshape(shape = var_2274, x = length_1_cast_fp16)[name = string("op_2275_cast_fp16")];
            tensor<fp16, [1, 751]> var_2271_promoted_to_fp16 = const()[name = string("op_2271_promoted_to_fp16"), val = tensor<fp16, [1, 751]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3584)))];
            tensor<bool, [1, 751]> var_2276_cast_fp16 = less(x = var_2271_promoted_to_fp16, y = var_2275_cast_fp16)[name = string("op_2276_cast_fp16")];
            tensor<int32, [4]> var_2286 = const()[name = string("op_2286"), val = tensor<int32, [4]>([1, 1, -1, 1])];
            string keep_1_to_fp16_dtype_0 = const()[name = string("keep_1_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 751]> var_2276_cast_fp16_to_fp16 = cast(dtype = keep_1_to_fp16_dtype_0, x = var_2276_cast_fp16)[name = string("cast_4")];
            tensor<fp16, [1, 1, 751, 1]> var_2287_cast_fp16 = reshape(shape = var_2286, x = var_2276_cast_fp16_to_fp16)[name = string("op_2287_cast_fp16")];
            tensor<fp16, [1, 256, 751, 64]> input_3_cast_fp16 = mul(x = x_1_cast_fp16, y = var_2287_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 x_3_pad_type_0 = const()[name = string("x_3_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> x_3_pad_0 = const()[name = string("x_3_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> x_3_strides_0 = const()[name = string("x_3_strides_0"), val = tensor<int32, [2]>([2, 2])];
            int32 x_3_groups_0 = const()[name = string("x_3_groups_0"), val = int32(256)];
            tensor<int32, [2]> x_3_dilations_0 = const()[name = string("x_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [256, 1, 3, 3]> enc_subsampling_layers_2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5184))), scale = tensor<fp16, [256, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7552))))[name = string("enc_subsampling_layers_2_weight_to_fp16_quantized")];
            tensor<fp16, [256]> enc_subsampling_layers_2_bias_to_fp16 = const()[name = string("enc_subsampling_layers_2_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8128)))];
            tensor<fp16, [1, 256, 376, 32]> x_3_cast_fp16 = conv(bias = enc_subsampling_layers_2_bias_to_fp16, dilations = x_3_dilations_0, groups = x_3_groups_0, pad = x_3_pad_0, pad_type = x_3_pad_type_0, strides = x_3_strides_0, weight = enc_subsampling_layers_2_weight_to_fp16_quantized, x = input_5_cast_fp16)[name = string("x_3_cast_fp16")];
            fp16 var_2304_promoted_to_fp16 = const()[name = string("op_2304_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1]> var_2305_cast_fp16 = add(x = length_1_cast_fp16, y = var_2304_promoted_to_fp16)[name = string("op_2305_cast_fp16")];
            fp16 var_2307_promoted_to_fp16 = const()[name = string("op_2307_promoted_to_fp16"), val = fp16(0x1.8p+1)];
            tensor<fp16, [1]> var_2308_cast_fp16 = sub(x = var_2305_cast_fp16, y = var_2307_promoted_to_fp16)[name = string("op_2308_cast_fp16")];
            fp16 cast_5_to_fp16 = const()[name = string("cast_5_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1]> var_2311_cast_fp16 = floor_div(x = var_2308_cast_fp16, y = cast_5_to_fp16)[name = string("op_2311_cast_fp16")];
            fp16 var_2313_promoted_to_fp16 = const()[name = string("op_2313_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> length_3_cast_fp16 = add(x = var_2311_cast_fp16, y = var_2313_promoted_to_fp16)[name = string("length_3_cast_fp16")];
            tensor<int32, [2]> var_2325 = const()[name = string("op_2325"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1, 1]> var_2326_cast_fp16 = reshape(shape = var_2325, x = length_3_cast_fp16)[name = string("op_2326_cast_fp16")];
            tensor<fp16, [1, 376]> var_2322_promoted_to_fp16 = const()[name = string("op_2322_promoted_to_fp16"), val = tensor<fp16, [1, 376]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8704)))];
            tensor<bool, [1, 376]> var_2327_cast_fp16 = less(x = var_2322_promoted_to_fp16, y = var_2326_cast_fp16)[name = string("op_2327_cast_fp16")];
            tensor<int32, [4]> var_2337 = const()[name = string("op_2337"), val = tensor<int32, [4]>([1, 1, -1, 1])];
            string keep_3_to_fp16_dtype_0 = const()[name = string("keep_3_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 376]> var_2327_cast_fp16_to_fp16 = cast(dtype = keep_3_to_fp16_dtype_0, x = var_2327_cast_fp16)[name = string("cast_3")];
            tensor<fp16, [1, 1, 376, 1]> var_2338_cast_fp16 = reshape(shape = var_2337, x = var_2327_cast_fp16_to_fp16)[name = string("op_2338_cast_fp16")];
            tensor<fp16, [1, 256, 376, 32]> input_7_cast_fp16 = mul(x = x_3_cast_fp16, y = var_2338_cast_fp16)[name = string("input_7_cast_fp16")];
            string x_5_pad_type_0 = const()[name = string("x_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> x_5_strides_0 = const()[name = string("x_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_5_pad_0 = const()[name = string("x_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_5_dilations_0 = const()[name = string("x_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 x_5_groups_0 = const()[name = string("x_5_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1, 1]> enc_subsampling_layers_3_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9536))), scale = tensor<fp16, [256, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75136))))[name = string("enc_subsampling_layers_3_weight_to_fp16_quantized")];
            tensor<fp16, [256]> enc_subsampling_layers_3_bias_to_fp16 = const()[name = string("enc_subsampling_layers_3_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75712)))];
            tensor<fp16, [1, 256, 376, 32]> x_5_cast_fp16 = conv(bias = enc_subsampling_layers_3_bias_to_fp16, dilations = x_5_dilations_0, groups = x_5_groups_0, pad = x_5_pad_0, pad_type = x_5_pad_type_0, strides = x_5_strides_0, weight = enc_subsampling_layers_3_weight_to_fp16_quantized, x = input_7_cast_fp16)[name = string("x_5_cast_fp16")];
            tensor<fp16, [1, 256, 376, 32]> input_9_cast_fp16 = mul(x = x_5_cast_fp16, y = var_2338_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 x_7_pad_type_0 = const()[name = string("x_7_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> x_7_pad_0 = const()[name = string("x_7_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> x_7_strides_0 = const()[name = string("x_7_strides_0"), val = tensor<int32, [2]>([2, 2])];
            int32 x_7_groups_0 = const()[name = string("x_7_groups_0"), val = int32(256)];
            tensor<int32, [2]> x_7_dilations_0 = const()[name = string("x_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [256, 1, 3, 3]> enc_subsampling_layers_5_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76288))), scale = tensor<fp16, [256, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78656))))[name = string("enc_subsampling_layers_5_weight_to_fp16_quantized")];
            tensor<fp16, [256]> enc_subsampling_layers_5_bias_to_fp16 = const()[name = string("enc_subsampling_layers_5_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79232)))];
            tensor<fp16, [1, 256, 188, 16]> x_7_cast_fp16 = conv(bias = enc_subsampling_layers_5_bias_to_fp16, dilations = x_7_dilations_0, groups = x_7_groups_0, pad = x_7_pad_0, pad_type = x_7_pad_type_0, strides = x_7_strides_0, weight = enc_subsampling_layers_5_weight_to_fp16_quantized, x = input_11_cast_fp16)[name = string("x_7_cast_fp16")];
            fp16 var_2391_promoted_to_fp16 = const()[name = string("op_2391_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1]> var_2392_cast_fp16 = add(x = length_3_cast_fp16, y = var_2391_promoted_to_fp16)[name = string("op_2392_cast_fp16")];
            fp16 var_2394_promoted_to_fp16 = const()[name = string("op_2394_promoted_to_fp16"), val = fp16(0x1.8p+1)];
            tensor<fp16, [1]> var_2395_cast_fp16 = sub(x = var_2392_cast_fp16, y = var_2394_promoted_to_fp16)[name = string("op_2395_cast_fp16")];
            fp16 cast_11_to_fp16 = const()[name = string("cast_11_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1]> var_2398_cast_fp16 = floor_div(x = var_2395_cast_fp16, y = cast_11_to_fp16)[name = string("op_2398_cast_fp16")];
            fp16 var_2400_promoted_to_fp16 = const()[name = string("op_2400_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> length_cast_fp16 = add(x = var_2398_cast_fp16, y = var_2400_promoted_to_fp16)[name = string("length_cast_fp16")];
            tensor<int32, [2]> var_2412 = const()[name = string("op_2412"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1, 1]> var_2413_cast_fp16 = reshape(shape = var_2412, x = length_cast_fp16)[name = string("op_2413_cast_fp16")];
            tensor<fp16, [1, 188]> var_2409_promoted_to_fp16 = const()[name = string("op_2409_promoted_to_fp16"), val = tensor<fp16, [1, 188]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79808)))];
            tensor<bool, [1, 188]> var_2414_cast_fp16 = less(x = var_2409_promoted_to_fp16, y = var_2413_cast_fp16)[name = string("op_2414_cast_fp16")];
            tensor<int32, [4]> var_2424 = const()[name = string("op_2424"), val = tensor<int32, [4]>([1, 1, -1, 1])];
            string keep_7_to_fp16_dtype_0 = const()[name = string("keep_7_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 188]> var_2414_cast_fp16_to_fp16 = cast(dtype = keep_7_to_fp16_dtype_0, x = var_2414_cast_fp16)[name = string("cast_2")];
            tensor<fp16, [1, 1, 188, 1]> var_2425_cast_fp16 = reshape(shape = var_2424, x = var_2414_cast_fp16_to_fp16)[name = string("op_2425_cast_fp16")];
            tensor<fp16, [1, 256, 188, 16]> input_13_cast_fp16 = mul(x = x_7_cast_fp16, y = var_2425_cast_fp16)[name = string("input_13_cast_fp16")];
            string x_9_pad_type_0 = const()[name = string("x_9_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> x_9_strides_0 = const()[name = string("x_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_9_pad_0 = const()[name = string("x_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_9_dilations_0 = const()[name = string("x_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 x_9_groups_0 = const()[name = string("x_9_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1, 1]> enc_subsampling_layers_6_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80256))), scale = tensor<fp16, [256, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145856))))[name = string("enc_subsampling_layers_6_weight_to_fp16_quantized")];
            tensor<fp16, [256]> enc_subsampling_layers_6_bias_to_fp16 = const()[name = string("enc_subsampling_layers_6_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146432)))];
            tensor<fp16, [1, 256, 188, 16]> x_9_cast_fp16 = conv(bias = enc_subsampling_layers_6_bias_to_fp16, dilations = x_9_dilations_0, groups = x_9_groups_0, pad = x_9_pad_0, pad_type = x_9_pad_type_0, strides = x_9_strides_0, weight = enc_subsampling_layers_6_weight_to_fp16_quantized, x = input_13_cast_fp16)[name = string("x_9_cast_fp16")];
            tensor<fp16, [1, 256, 188, 16]> input_15_cast_fp16 = mul(x = x_9_cast_fp16, y = var_2425_cast_fp16)[name = string("input_15_cast_fp16")];
            tensor<fp16, [1, 256, 188, 16]> x_11_cast_fp16 = relu(x = input_15_cast_fp16)[name = string("x_11_cast_fp16")];
            tensor<int32, [4]> var_2466_perm_0 = const()[name = string("op_2466_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2470 = const()[name = string("op_2470"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 256, 16]> var_2466_cast_fp16 = transpose(perm = var_2466_perm_0, x = x_11_cast_fp16)[name = string("transpose_481")];
            tensor<fp16, [1, 188, 4096]> input_17_cast_fp16 = reshape(shape = var_2470, x = var_2466_cast_fp16)[name = string("input_17_cast_fp16")];
            tensor<fp16, [1024, 4096]> enc_subsampling_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147008))), scale = tensor<fp16, [1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4341376))))[name = string("enc_subsampling_linear_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> enc_subsampling_linear_bias_to_fp16 = const()[name = string("enc_subsampling_linear_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4343488)))];
            tensor<fp16, [1, 188, 1024]> linear_0_cast_fp16 = linear(bias = enc_subsampling_linear_bias_to_fp16, weight = enc_subsampling_linear_weight_to_fp16_quantized, x = input_17_cast_fp16)[name = string("linear_0_cast_fp16")];
            fp16 var_2495_to_fp16 = const()[name = string("op_2495_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 188]> var_2497_cast_fp16 = sub(x = var_2495_to_fp16, y = var_2414_cast_fp16_to_fp16)[name = string("op_2497_cast_fp16")];
            fp16 var_2498_to_fp16 = const()[name = string("op_2498_to_fp16"), val = fp16(-0x1.388p+13)];
            tensor<fp16, [1, 188]> var_2499_cast_fp16 = mul(x = var_2497_cast_fp16, y = var_2498_to_fp16)[name = string("op_2499_cast_fp16")];
            tensor<int32, [4]> var_2504 = const()[name = string("op_2504"), val = tensor<int32, [4]>([1, 1, 1, 188])];
            tensor<fp16, [1, 1, 1, 188]> key_bias_cast_fp16 = reshape(shape = var_2504, x = var_2499_cast_fp16)[name = string("key_bias_cast_fp16")];
            tensor<int32, [1]> valid_axes_0 = const()[name = string("valid_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 188, 1]> valid_cast_fp16 = expand_dims(axes = valid_axes_0, x = var_2414_cast_fp16_to_fp16)[name = string("valid_cast_fp16")];
            fp16 var_2513_promoted_to_fp16 = const()[name = string("op_2513_promoted_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1]> var_2514_cast_fp16 = mul(x = length_cast_fp16, y = var_2513_promoted_to_fp16)[name = string("op_2514_cast_fp16")];
            fp16 var_2516_promoted_to_fp16 = const()[name = string("op_2516_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> var_2517_cast_fp16 = add(x = var_2514_cast_fp16, y = var_2516_promoted_to_fp16)[name = string("op_2517_cast_fp16")];
            tensor<int32, [3]> var_2521 = const()[name = string("op_2521"), val = tensor<int32, [3]>([1, 1, 1])];
            tensor<fp16, [1, 1, 1]> var_2522_cast_fp16 = reshape(shape = var_2521, x = var_2517_cast_fp16)[name = string("op_2522_cast_fp16")];
            tensor<fp16, [1, 375, 1024]> pos_emb_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1, 375, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4345600))), scale = tensor<fp16, [1, 1, 1]>([[[0x1.02p-7]]]))[name = string("pos_emb_to_fp16_quantized")];
            tensor<fp16, [1, 375, 1024]> x_21_cast_fp16 = mul(x = pos_emb_to_fp16_quantized, y = var_2522_cast_fp16)[name = string("x_21_cast_fp16")];
            int32 var_2534 = const()[name = string("op_2534"), val = int32(-1)];
            tensor<int32, [1]> x_15_axes_0 = const()[name = string("x_15_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_0_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_0_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4729664)))];
            tensor<fp16, [1024]> enc_layers_0_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_0_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4731776)))];
            fp16 var_2541_to_fp16 = const()[name = string("op_2541_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_15_cast_fp16 = layer_norm(axes = x_15_axes_0, beta = enc_layers_0_norm_feed_forward1_bias_to_fp16, epsilon = var_2541_to_fp16, gamma = enc_layers_0_norm_feed_forward1_weight_to_fp16, x = linear_0_cast_fp16)[name = string("x_15_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_0_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4733888))), lut = tensor<fp16, [256, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5782528))))[name = string("blocks_0_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> blocks_0_ff1_1_bias_to_fp16 = const()[name = string("blocks_0_ff1_1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5784640)))];
            tensor<fp16, [1, 188, 4096]> linear_1_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_0_ff1_1_weight_to_fp16_palettized, x = x_15_cast_fp16)[name = string("linear_1_cast_fp16")];
            tensor<fp16, [4096]> blocks_0_ff1_1_scale_to_fp16 = const()[name = string("blocks_0_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5792896)))];
            tensor<fp16, [1, 188, 4096]> input_21_cast_fp16 = mul(x = linear_1_cast_fp16, y = blocks_0_ff1_1_scale_to_fp16)[name = string("input_21_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_17_cast_fp16 = silu(x = input_21_cast_fp16)[name = string("x_17_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_0_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5801152))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6849792))))[name = string("blocks_0_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> blocks_0_ff1_2_bias_to_fp16 = const()[name = string("blocks_0_ff1_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6850368)))];
            tensor<fp16, [1, 188, 1024]> linear_2_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_0_ff1_2_weight_to_fp16_palettized, x = x_17_cast_fp16)[name = string("linear_2_cast_fp16")];
            tensor<fp16, [1024]> blocks_0_ff1_2_scale_to_fp16 = const()[name = string("blocks_0_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6852480)))];
            tensor<fp16, [1, 188, 1024]> var_2566_cast_fp16 = mul(x = linear_2_cast_fp16, y = blocks_0_ff1_2_scale_to_fp16)[name = string("op_2566_cast_fp16")];
            fp16 var_2567_to_fp16 = const()[name = string("op_2567_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2568_cast_fp16 = mul(x = var_2566_cast_fp16, y = var_2567_to_fp16)[name = string("op_2568_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_23_cast_fp16 = add(x = linear_0_cast_fp16, y = var_2568_cast_fp16)[name = string("input_23_cast_fp16")];
            tensor<int32, [1]> x_19_axes_0 = const()[name = string("x_19_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_0_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_0_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6854592)))];
            tensor<fp16, [1024]> enc_layers_0_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_0_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6856704)))];
            tensor<fp16, [1, 188, 1024]> x_19_cast_fp16 = layer_norm(axes = x_19_axes_0, beta = enc_layers_0_norm_self_att_bias_to_fp16, epsilon = var_2541_to_fp16, gamma = enc_layers_0_norm_self_att_weight_to_fp16, x = input_23_cast_fp16)[name = string("x_19_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_0_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6858816))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7121024))))[name = string("blocks_0_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_3_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_0_q_weight_to_fp16_palettized, x = x_19_cast_fp16)[name = string("linear_3_cast_fp16")];
            tensor<fp16, [1024]> blocks_0_q_scale_to_fp16 = const()[name = string("blocks_0_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7121600)))];
            tensor<fp16, [1, 188, 1024]> var_2576_cast_fp16 = mul(x = linear_3_cast_fp16, y = blocks_0_q_scale_to_fp16)[name = string("op_2576_cast_fp16")];
            tensor<int32, [4]> var_2577 = const()[name = string("op_2577"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_2578_cast_fp16 = reshape(shape = var_2577, x = var_2576_cast_fp16)[name = string("op_2578_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_0_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7123712))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7385920))))[name = string("blocks_0_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_4_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_0_k_weight_to_fp16_palettized, x = x_19_cast_fp16)[name = string("linear_4_cast_fp16")];
            tensor<fp16, [1024]> blocks_0_k_scale_to_fp16 = const()[name = string("blocks_0_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7386496)))];
            tensor<fp16, [1, 188, 1024]> var_2584_cast_fp16 = mul(x = linear_4_cast_fp16, y = blocks_0_k_scale_to_fp16)[name = string("op_2584_cast_fp16")];
            tensor<int32, [4]> var_2585 = const()[name = string("op_2585"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_2586_cast_fp16 = reshape(shape = var_2585, x = var_2584_cast_fp16)[name = string("op_2586_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_0_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7388608))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7650816))))[name = string("blocks_0_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_5_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_0_v_weight_to_fp16_palettized, x = x_19_cast_fp16)[name = string("linear_5_cast_fp16")];
            tensor<fp16, [1024]> blocks_0_v_scale_to_fp16 = const()[name = string("blocks_0_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7651392)))];
            tensor<fp16, [1, 188, 1024]> var_2592_cast_fp16 = mul(x = linear_5_cast_fp16, y = blocks_0_v_scale_to_fp16)[name = string("op_2592_cast_fp16")];
            tensor<int32, [4]> var_2593 = const()[name = string("op_2593"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_2594_cast_fp16 = reshape(shape = var_2593, x = var_2592_cast_fp16)[name = string("op_2594_cast_fp16")];
            tensor<int32, [4]> v_1_perm_0 = const()[name = string("v_1_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_84_to_fp16 = const()[name = string("const_84_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7653504)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_1_cast_fp16 = add(x = var_2578_cast_fp16, y = const_84_to_fp16)[name = string("q_u_1_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_85_to_fp16 = const()[name = string("const_85_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7655616)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_1_cast_fp16 = add(x = var_2578_cast_fp16, y = const_85_to_fp16)[name = string("q_v_1_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_0_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7657728))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7919936))))[name = string("blocks_0_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_6_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_0_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_6_cast_fp16")];
            tensor<fp16, [1024]> blocks_0_pos_scale_to_fp16 = const()[name = string("blocks_0_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7920512)))];
            tensor<fp16, [1, 375, 1024]> var_2606_cast_fp16 = mul(x = linear_6_cast_fp16, y = blocks_0_pos_scale_to_fp16)[name = string("op_2606_cast_fp16")];
            tensor<int32, [4]> var_2607 = const()[name = string("op_2607"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_1_cast_fp16 = reshape(shape = var_2607, x = var_2606_cast_fp16)[name = string("p_1_cast_fp16")];
            bool input_25_transpose_x_0 = const()[name = string("input_25_transpose_x_0"), val = bool(false)];
            bool input_25_transpose_y_0 = const()[name = string("input_25_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_192_perm_0 = const()[name = string("transpose_192_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_193_perm_0 = const()[name = string("transpose_193_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_193 = transpose(perm = transpose_193_perm_0, x = p_1_cast_fp16)[name = string("transpose_479")];
            tensor<fp16, [1, 8, 188, 128]> transpose_192 = transpose(perm = transpose_192_perm_0, x = q_v_1_cast_fp16)[name = string("transpose_480")];
            tensor<fp16, [1, 8, 188, 375]> input_25_cast_fp16 = matmul(transpose_x = input_25_transpose_x_0, transpose_y = input_25_transpose_y_0, x = transpose_192, y = transpose_193)[name = string("input_25_cast_fp16")];
            tensor<int32, [8]> var_2613_pad_0 = const()[name = string("op_2613_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_2613_mode_0 = const()[name = string("op_2613_mode_0"), val = string("constant")];
            fp16 const_12_to_fp16 = const()[name = string("const_12_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_2613_cast_fp16 = pad(constant_val = const_12_to_fp16, mode = var_2613_mode_0, pad = var_2613_pad_0, x = input_25_cast_fp16)[name = string("op_2613_cast_fp16")];
            tensor<int32, [4]> var_2614 = const()[name = string("op_2614"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_1_cast_fp16 = reshape(shape = var_2614, x = var_2613_cast_fp16)[name = string("scores_1_cast_fp16")];
            tensor<int32, [4]> var_2618_begin_0 = const()[name = string("op_2618_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_2618_end_0 = const()[name = string("op_2618_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_2618_end_mask_0 = const()[name = string("op_2618_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_2618_cast_fp16 = slice_by_index(begin = var_2618_begin_0, end = var_2618_end_0, end_mask = var_2618_end_mask_0, x = scores_1_cast_fp16)[name = string("op_2618_cast_fp16")];
            tensor<int32, [4]> var_2619 = const()[name = string("op_2619"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_2620_cast_fp16 = reshape(shape = var_2619, x = var_2618_cast_fp16)[name = string("op_2620_cast_fp16")];
            tensor<int32, [4]> bd_1_begin_0 = const()[name = string("bd_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_1_end_0 = const()[name = string("bd_1_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_1_end_mask_0 = const()[name = string("bd_1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_1_cast_fp16 = slice_by_index(begin = bd_1_begin_0, end = bd_1_end_0, end_mask = bd_1_end_mask_0, x = var_2620_cast_fp16)[name = string("bd_1_cast_fp16")];
            bool var_2623_transpose_x_0 = const()[name = string("op_2623_transpose_x_0"), val = bool(false)];
            bool var_2623_transpose_y_0 = const()[name = string("op_2623_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_194_perm_0 = const()[name = string("transpose_194_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_195_perm_0 = const()[name = string("transpose_195_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_195 = transpose(perm = transpose_195_perm_0, x = var_2586_cast_fp16)[name = string("transpose_477")];
            tensor<fp16, [1, 8, 188, 128]> transpose_194 = transpose(perm = transpose_194_perm_0, x = q_u_1_cast_fp16)[name = string("transpose_478")];
            tensor<fp16, [1, 8, 188, 188]> var_2623_cast_fp16 = matmul(transpose_x = var_2623_transpose_x_0, transpose_y = var_2623_transpose_y_0, x = transpose_194, y = transpose_195)[name = string("op_2623_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2624_cast_fp16 = add(x = var_2623_cast_fp16, y = bd_1_cast_fp16)[name = string("op_2624_cast_fp16")];
            fp16 var_2625_to_fp16 = const()[name = string("op_2625_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_2626_cast_fp16 = mul(x = var_2624_cast_fp16, y = var_2625_to_fp16)[name = string("op_2626_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_3_cast_fp16 = add(x = var_2626_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_3_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2628_cast_fp16 = softmax(axis = var_2534, x = scores_3_cast_fp16)[name = string("op_2628_cast_fp16")];
            bool out_1_transpose_x_0 = const()[name = string("out_1_transpose_x_0"), val = bool(false)];
            bool out_1_transpose_y_0 = const()[name = string("out_1_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_1_cast_fp16 = transpose(perm = v_1_perm_0, x = var_2594_cast_fp16)[name = string("transpose_476")];
            tensor<fp16, [1, 8, 188, 128]> out_1_cast_fp16 = matmul(transpose_x = out_1_transpose_x_0, transpose_y = out_1_transpose_y_0, x = var_2628_cast_fp16, y = v_1_cast_fp16)[name = string("out_1_cast_fp16")];
            tensor<int32, [4]> var_2630_perm_0 = const()[name = string("op_2630_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2631 = const()[name = string("op_2631"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_2630_cast_fp16 = transpose(perm = var_2630_perm_0, x = out_1_cast_fp16)[name = string("transpose_475")];
            tensor<fp16, [1, 188, 1024]> x_23_cast_fp16 = reshape(shape = var_2631, x = var_2630_cast_fp16)[name = string("x_23_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_0_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7922624))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8184832))))[name = string("blocks_0_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_7_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_0_o_weight_to_fp16_palettized, x = x_23_cast_fp16)[name = string("linear_7_cast_fp16")];
            tensor<fp16, [1024]> blocks_0_o_scale_to_fp16 = const()[name = string("blocks_0_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8185408)))];
            tensor<fp16, [1, 188, 1024]> var_2637_cast_fp16 = mul(x = linear_7_cast_fp16, y = blocks_0_o_scale_to_fp16)[name = string("op_2637_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_27_cast_fp16 = add(x = input_23_cast_fp16, y = var_2637_cast_fp16)[name = string("input_27_cast_fp16")];
            tensor<int32, [1]> x_25_axes_0 = const()[name = string("x_25_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_0_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_0_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8187520)))];
            tensor<fp16, [1024]> enc_layers_0_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_0_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8189632)))];
            tensor<fp16, [1, 188, 1024]> x_25_cast_fp16 = layer_norm(axes = x_25_axes_0, beta = enc_layers_0_norm_conv_bias_to_fp16, epsilon = var_2541_to_fp16, gamma = enc_layers_0_norm_conv_weight_to_fp16, x = input_27_cast_fp16)[name = string("x_25_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_0_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8191744))), lut = tensor<fp16, [128, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8716096))))[name = string("blocks_0_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [2048]> blocks_0_pw1_bias_to_fp16 = const()[name = string("blocks_0_pw1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8717184)))];
            tensor<fp16, [1, 188, 2048]> linear_8_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_0_pw1_weight_to_fp16_palettized, x = x_25_cast_fp16)[name = string("linear_8_cast_fp16")];
            tensor<fp16, [2048]> blocks_0_pw1_scale_to_fp16 = const()[name = string("blocks_0_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8721344)))];
            tensor<fp16, [1, 188, 2048]> input_29_cast_fp16 = mul(x = linear_8_cast_fp16, y = blocks_0_pw1_scale_to_fp16)[name = string("input_29_cast_fp16")];
            int32 var_2646_split_num_splits_0 = const()[name = string("op_2646_split_num_splits_0"), val = int32(2)];
            int32 var_2646_split_axis_0 = const()[name = string("op_2646_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_2646_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_2646_split_cast_fp16_1 = split(axis = var_2646_split_axis_0, num_splits = var_2646_split_num_splits_0, x = input_29_cast_fp16)[name = string("op_2646_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_2646_split_1_sigmoid_cast_fp16 = sigmoid(x = var_2646_split_cast_fp16_1)[name = string("op_2646_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_2646_cast_fp16 = mul(x = var_2646_split_cast_fp16_0, y = var_2646_split_1_sigmoid_cast_fp16)[name = string("op_2646_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_1_cast_fp16 = mul(x = var_2646_cast_fp16, y = valid_cast_fp16)[name = string("h_1_cast_fp16")];
            tensor<int32, [3]> input_31_perm_0 = const()[name = string("input_31_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_33_pad_type_0 = const()[name = string("input_33_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_33_pad_0 = const()[name = string("input_33_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_33_groups_0 = const()[name = string("input_33_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_33_strides_0 = const()[name = string("input_33_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_33_dilations_0 = const()[name = string("input_33_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_36_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8725504))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8734784))))[name = string("const_36_to_fp16_quantized")];
            tensor<fp16, [1024]> const_37_to_fp16 = const()[name = string("const_37_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8736896)))];
            tensor<fp16, [1, 1024, 188]> input_31_cast_fp16 = transpose(perm = input_31_perm_0, x = h_1_cast_fp16)[name = string("transpose_474")];
            tensor<fp16, [1, 1024, 188]> input_35_cast_fp16 = conv(bias = const_37_to_fp16, dilations = input_33_dilations_0, groups = input_33_groups_0, pad = input_33_pad_0, pad_type = input_33_pad_type_0, strides = input_33_strides_0, weight = const_36_to_fp16_quantized, x = input_31_cast_fp16)[name = string("input_35_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_3_cast_fp16 = silu(x = input_35_cast_fp16)[name = string("h_3_cast_fp16")];
            tensor<int32, [3]> x_27_perm_0 = const()[name = string("x_27_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_0_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8739008))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9001216))))[name = string("blocks_0_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_27_cast_fp16 = transpose(perm = x_27_perm_0, x = h_3_cast_fp16)[name = string("transpose_473")];
            tensor<fp16, [1, 188, 1024]> linear_9_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_0_pw2_weight_to_fp16_palettized, x = x_27_cast_fp16)[name = string("linear_9_cast_fp16")];
            tensor<fp16, [1024]> blocks_0_pw2_scale_to_fp16 = const()[name = string("blocks_0_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9001792)))];
            tensor<fp16, [1, 188, 1024]> var_2661_cast_fp16 = mul(x = linear_9_cast_fp16, y = blocks_0_pw2_scale_to_fp16)[name = string("op_2661_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_37_cast_fp16 = add(x = input_27_cast_fp16, y = var_2661_cast_fp16)[name = string("input_37_cast_fp16")];
            tensor<int32, [1]> x_29_axes_0 = const()[name = string("x_29_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_0_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_0_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9003904)))];
            tensor<fp16, [1024]> enc_layers_0_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_0_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9006016)))];
            tensor<fp16, [1, 188, 1024]> x_29_cast_fp16 = layer_norm(axes = x_29_axes_0, beta = enc_layers_0_norm_feed_forward2_bias_to_fp16, epsilon = var_2541_to_fp16, gamma = enc_layers_0_norm_feed_forward2_weight_to_fp16, x = input_37_cast_fp16)[name = string("x_29_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_0_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9008128))), lut = tensor<fp16, [256, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10056768))))[name = string("blocks_0_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_10_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_0_ff2_1_weight_to_fp16_palettized, x = x_29_cast_fp16)[name = string("linear_10_cast_fp16")];
            tensor<fp16, [4096]> blocks_0_ff2_1_scale_to_fp16 = const()[name = string("blocks_0_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10058880)))];
            tensor<fp16, [1, 188, 4096]> input_39_cast_fp16 = mul(x = linear_10_cast_fp16, y = blocks_0_ff2_1_scale_to_fp16)[name = string("input_39_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_31_cast_fp16 = silu(x = input_39_cast_fp16)[name = string("x_31_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_0_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10067136))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11115776))))[name = string("blocks_0_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_11_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_0_ff2_2_weight_to_fp16_palettized, x = x_31_cast_fp16)[name = string("linear_11_cast_fp16")];
            tensor<fp16, [1024]> blocks_0_ff2_2_scale_to_fp16 = const()[name = string("blocks_0_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11116352)))];
            tensor<fp16, [1, 188, 1024]> var_2675_cast_fp16 = mul(x = linear_11_cast_fp16, y = blocks_0_ff2_2_scale_to_fp16)[name = string("op_2675_cast_fp16")];
            fp16 var_2676_to_fp16 = const()[name = string("op_2676_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2677_cast_fp16 = mul(x = var_2675_cast_fp16, y = var_2676_to_fp16)[name = string("op_2677_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_41_cast_fp16 = add(x = input_37_cast_fp16, y = var_2677_cast_fp16)[name = string("input_41_cast_fp16")];
            tensor<int32, [1]> input_43_axes_0 = const()[name = string("input_43_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_0_norm_out_weight_to_fp16 = const()[name = string("enc_layers_0_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11118464)))];
            tensor<fp16, [1024]> enc_layers_0_norm_out_bias_to_fp16 = const()[name = string("enc_layers_0_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11120576)))];
            tensor<fp16, [1, 188, 1024]> input_43_cast_fp16 = layer_norm(axes = input_43_axes_0, beta = enc_layers_0_norm_out_bias_to_fp16, epsilon = var_2541_to_fp16, gamma = enc_layers_0_norm_out_weight_to_fp16, x = input_41_cast_fp16)[name = string("input_43_cast_fp16")];
            int32 var_2691 = const()[name = string("op_2691"), val = int32(-1)];
            tensor<int32, [1]> x_33_axes_0 = const()[name = string("x_33_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_1_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_1_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11122688)))];
            tensor<fp16, [1024]> enc_layers_1_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_1_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11124800)))];
            fp16 var_2698_to_fp16 = const()[name = string("op_2698_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_33_cast_fp16 = layer_norm(axes = x_33_axes_0, beta = enc_layers_1_norm_feed_forward1_bias_to_fp16, epsilon = var_2698_to_fp16, gamma = enc_layers_1_norm_feed_forward1_weight_to_fp16, x = input_43_cast_fp16)[name = string("x_33_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_1_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11126912))), lut = tensor<fp16, [256, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12699840))))[name = string("blocks_1_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_12_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_1_ff1_1_weight_to_fp16_palettized, x = x_33_cast_fp16)[name = string("linear_12_cast_fp16")];
            tensor<fp16, [4096]> blocks_1_ff1_1_scale_to_fp16 = const()[name = string("blocks_1_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12704000)))];
            tensor<fp16, [1, 188, 4096]> input_45_cast_fp16 = mul(x = linear_12_cast_fp16, y = blocks_1_ff1_1_scale_to_fp16)[name = string("input_45_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_35_cast_fp16 = silu(x = input_45_cast_fp16)[name = string("x_35_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_1_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12712256))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14285184))))[name = string("blocks_1_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_13_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_1_ff1_2_weight_to_fp16_palettized, x = x_35_cast_fp16)[name = string("linear_13_cast_fp16")];
            tensor<fp16, [1024]> blocks_1_ff1_2_scale_to_fp16 = const()[name = string("blocks_1_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14286272)))];
            tensor<fp16, [1, 188, 1024]> var_2723_cast_fp16 = mul(x = linear_13_cast_fp16, y = blocks_1_ff1_2_scale_to_fp16)[name = string("op_2723_cast_fp16")];
            fp16 var_2724_to_fp16 = const()[name = string("op_2724_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2725_cast_fp16 = mul(x = var_2723_cast_fp16, y = var_2724_to_fp16)[name = string("op_2725_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_47_cast_fp16 = add(x = input_43_cast_fp16, y = var_2725_cast_fp16)[name = string("input_47_cast_fp16")];
            tensor<int32, [1]> x_37_axes_0 = const()[name = string("x_37_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_1_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_1_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14288384)))];
            tensor<fp16, [1024]> enc_layers_1_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_1_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14290496)))];
            tensor<fp16, [1, 188, 1024]> x_37_cast_fp16 = layer_norm(axes = x_37_axes_0, beta = enc_layers_1_norm_self_att_bias_to_fp16, epsilon = var_2698_to_fp16, gamma = enc_layers_1_norm_self_att_weight_to_fp16, x = input_47_cast_fp16)[name = string("x_37_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_1_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14292608))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14554816))))[name = string("blocks_1_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_14_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_1_q_weight_to_fp16_palettized, x = x_37_cast_fp16)[name = string("linear_14_cast_fp16")];
            tensor<fp16, [1024]> blocks_1_q_scale_to_fp16 = const()[name = string("blocks_1_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14555392)))];
            tensor<fp16, [1, 188, 1024]> var_2733_cast_fp16 = mul(x = linear_14_cast_fp16, y = blocks_1_q_scale_to_fp16)[name = string("op_2733_cast_fp16")];
            tensor<int32, [4]> var_2734 = const()[name = string("op_2734"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_2735_cast_fp16 = reshape(shape = var_2734, x = var_2733_cast_fp16)[name = string("op_2735_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_1_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14557504))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14819712))))[name = string("blocks_1_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_15_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_1_k_weight_to_fp16_palettized, x = x_37_cast_fp16)[name = string("linear_15_cast_fp16")];
            tensor<fp16, [1024]> blocks_1_k_scale_to_fp16 = const()[name = string("blocks_1_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14820288)))];
            tensor<fp16, [1, 188, 1024]> var_2741_cast_fp16 = mul(x = linear_15_cast_fp16, y = blocks_1_k_scale_to_fp16)[name = string("op_2741_cast_fp16")];
            tensor<int32, [4]> var_2742 = const()[name = string("op_2742"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_2743_cast_fp16 = reshape(shape = var_2742, x = var_2741_cast_fp16)[name = string("op_2743_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_1_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14822400))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15084608))))[name = string("blocks_1_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_16_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_1_v_weight_to_fp16_palettized, x = x_37_cast_fp16)[name = string("linear_16_cast_fp16")];
            tensor<fp16, [1024]> blocks_1_v_scale_to_fp16 = const()[name = string("blocks_1_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15085184)))];
            tensor<fp16, [1, 188, 1024]> var_2749_cast_fp16 = mul(x = linear_16_cast_fp16, y = blocks_1_v_scale_to_fp16)[name = string("op_2749_cast_fp16")];
            tensor<int32, [4]> var_2750 = const()[name = string("op_2750"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_2751_cast_fp16 = reshape(shape = var_2750, x = var_2749_cast_fp16)[name = string("op_2751_cast_fp16")];
            tensor<int32, [4]> v_3_perm_0 = const()[name = string("v_3_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_86_to_fp16 = const()[name = string("const_86_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15087296)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_3_cast_fp16 = add(x = var_2735_cast_fp16, y = const_86_to_fp16)[name = string("q_u_3_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_87_to_fp16 = const()[name = string("const_87_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15089408)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_3_cast_fp16 = add(x = var_2735_cast_fp16, y = const_87_to_fp16)[name = string("q_v_3_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_1_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15091520))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15353728))))[name = string("blocks_1_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_17_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_1_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_17_cast_fp16")];
            tensor<fp16, [1024]> blocks_1_pos_scale_to_fp16 = const()[name = string("blocks_1_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15354304)))];
            tensor<fp16, [1, 375, 1024]> var_2763_cast_fp16 = mul(x = linear_17_cast_fp16, y = blocks_1_pos_scale_to_fp16)[name = string("op_2763_cast_fp16")];
            tensor<int32, [4]> var_2764 = const()[name = string("op_2764"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_3_cast_fp16 = reshape(shape = var_2764, x = var_2763_cast_fp16)[name = string("p_3_cast_fp16")];
            bool input_49_transpose_x_0 = const()[name = string("input_49_transpose_x_0"), val = bool(false)];
            bool input_49_transpose_y_0 = const()[name = string("input_49_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_196_perm_0 = const()[name = string("transpose_196_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_197_perm_0 = const()[name = string("transpose_197_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_197 = transpose(perm = transpose_197_perm_0, x = p_3_cast_fp16)[name = string("transpose_471")];
            tensor<fp16, [1, 8, 188, 128]> transpose_196 = transpose(perm = transpose_196_perm_0, x = q_v_3_cast_fp16)[name = string("transpose_472")];
            tensor<fp16, [1, 8, 188, 375]> input_49_cast_fp16 = matmul(transpose_x = input_49_transpose_x_0, transpose_y = input_49_transpose_y_0, x = transpose_196, y = transpose_197)[name = string("input_49_cast_fp16")];
            tensor<int32, [8]> var_2770_pad_0 = const()[name = string("op_2770_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_2770_mode_0 = const()[name = string("op_2770_mode_0"), val = string("constant")];
            fp16 const_13_to_fp16 = const()[name = string("const_13_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_2770_cast_fp16 = pad(constant_val = const_13_to_fp16, mode = var_2770_mode_0, pad = var_2770_pad_0, x = input_49_cast_fp16)[name = string("op_2770_cast_fp16")];
            tensor<int32, [4]> var_2771 = const()[name = string("op_2771"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_5_cast_fp16 = reshape(shape = var_2771, x = var_2770_cast_fp16)[name = string("scores_5_cast_fp16")];
            tensor<int32, [4]> var_2775_begin_0 = const()[name = string("op_2775_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_2775_end_0 = const()[name = string("op_2775_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_2775_end_mask_0 = const()[name = string("op_2775_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_2775_cast_fp16 = slice_by_index(begin = var_2775_begin_0, end = var_2775_end_0, end_mask = var_2775_end_mask_0, x = scores_5_cast_fp16)[name = string("op_2775_cast_fp16")];
            tensor<int32, [4]> var_2776 = const()[name = string("op_2776"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_2777_cast_fp16 = reshape(shape = var_2776, x = var_2775_cast_fp16)[name = string("op_2777_cast_fp16")];
            tensor<int32, [4]> bd_3_begin_0 = const()[name = string("bd_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_3_end_0 = const()[name = string("bd_3_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_3_end_mask_0 = const()[name = string("bd_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_3_cast_fp16 = slice_by_index(begin = bd_3_begin_0, end = bd_3_end_0, end_mask = bd_3_end_mask_0, x = var_2777_cast_fp16)[name = string("bd_3_cast_fp16")];
            bool var_2780_transpose_x_0 = const()[name = string("op_2780_transpose_x_0"), val = bool(false)];
            bool var_2780_transpose_y_0 = const()[name = string("op_2780_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_198_perm_0 = const()[name = string("transpose_198_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_199_perm_0 = const()[name = string("transpose_199_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_199 = transpose(perm = transpose_199_perm_0, x = var_2743_cast_fp16)[name = string("transpose_469")];
            tensor<fp16, [1, 8, 188, 128]> transpose_198 = transpose(perm = transpose_198_perm_0, x = q_u_3_cast_fp16)[name = string("transpose_470")];
            tensor<fp16, [1, 8, 188, 188]> var_2780_cast_fp16 = matmul(transpose_x = var_2780_transpose_x_0, transpose_y = var_2780_transpose_y_0, x = transpose_198, y = transpose_199)[name = string("op_2780_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2781_cast_fp16 = add(x = var_2780_cast_fp16, y = bd_3_cast_fp16)[name = string("op_2781_cast_fp16")];
            fp16 var_2782_to_fp16 = const()[name = string("op_2782_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_2783_cast_fp16 = mul(x = var_2781_cast_fp16, y = var_2782_to_fp16)[name = string("op_2783_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_7_cast_fp16 = add(x = var_2783_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_7_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2785_cast_fp16 = softmax(axis = var_2691, x = scores_7_cast_fp16)[name = string("op_2785_cast_fp16")];
            bool out_3_transpose_x_0 = const()[name = string("out_3_transpose_x_0"), val = bool(false)];
            bool out_3_transpose_y_0 = const()[name = string("out_3_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_3_cast_fp16 = transpose(perm = v_3_perm_0, x = var_2751_cast_fp16)[name = string("transpose_468")];
            tensor<fp16, [1, 8, 188, 128]> out_3_cast_fp16 = matmul(transpose_x = out_3_transpose_x_0, transpose_y = out_3_transpose_y_0, x = var_2785_cast_fp16, y = v_3_cast_fp16)[name = string("out_3_cast_fp16")];
            tensor<int32, [4]> var_2787_perm_0 = const()[name = string("op_2787_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2788 = const()[name = string("op_2788"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_2787_cast_fp16 = transpose(perm = var_2787_perm_0, x = out_3_cast_fp16)[name = string("transpose_467")];
            tensor<fp16, [1, 188, 1024]> x_39_cast_fp16 = reshape(shape = var_2788, x = var_2787_cast_fp16)[name = string("x_39_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_1_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15356416))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15618624))))[name = string("blocks_1_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_18_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_1_o_weight_to_fp16_palettized, x = x_39_cast_fp16)[name = string("linear_18_cast_fp16")];
            tensor<fp16, [1024]> blocks_1_o_scale_to_fp16 = const()[name = string("blocks_1_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15619200)))];
            tensor<fp16, [1, 188, 1024]> var_2794_cast_fp16 = mul(x = linear_18_cast_fp16, y = blocks_1_o_scale_to_fp16)[name = string("op_2794_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_51_cast_fp16 = add(x = input_47_cast_fp16, y = var_2794_cast_fp16)[name = string("input_51_cast_fp16")];
            tensor<int32, [1]> x_41_axes_0 = const()[name = string("x_41_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_1_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_1_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15621312)))];
            tensor<fp16, [1024]> enc_layers_1_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_1_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15623424)))];
            tensor<fp16, [1, 188, 1024]> x_41_cast_fp16 = layer_norm(axes = x_41_axes_0, beta = enc_layers_1_norm_conv_bias_to_fp16, epsilon = var_2698_to_fp16, gamma = enc_layers_1_norm_conv_weight_to_fp16, x = input_51_cast_fp16)[name = string("x_41_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_1_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15625536))), lut = tensor<fp16, [128, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16412032))))[name = string("blocks_1_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_19_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_1_pw1_weight_to_fp16_palettized, x = x_41_cast_fp16)[name = string("linear_19_cast_fp16")];
            tensor<fp16, [2048]> blocks_1_pw1_scale_to_fp16 = const()[name = string("blocks_1_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16414144)))];
            tensor<fp16, [1, 188, 2048]> input_53_cast_fp16 = mul(x = linear_19_cast_fp16, y = blocks_1_pw1_scale_to_fp16)[name = string("input_53_cast_fp16")];
            int32 var_2803_split_num_splits_0 = const()[name = string("op_2803_split_num_splits_0"), val = int32(2)];
            int32 var_2803_split_axis_0 = const()[name = string("op_2803_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_2803_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_2803_split_cast_fp16_1 = split(axis = var_2803_split_axis_0, num_splits = var_2803_split_num_splits_0, x = input_53_cast_fp16)[name = string("op_2803_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_2803_split_1_sigmoid_cast_fp16 = sigmoid(x = var_2803_split_cast_fp16_1)[name = string("op_2803_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_2803_cast_fp16 = mul(x = var_2803_split_cast_fp16_0, y = var_2803_split_1_sigmoid_cast_fp16)[name = string("op_2803_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_5_cast_fp16 = mul(x = var_2803_cast_fp16, y = valid_cast_fp16)[name = string("h_5_cast_fp16")];
            tensor<int32, [3]> input_55_perm_0 = const()[name = string("input_55_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_57_pad_type_0 = const()[name = string("input_57_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_57_pad_0 = const()[name = string("input_57_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_57_groups_0 = const()[name = string("input_57_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_57_strides_0 = const()[name = string("input_57_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_57_dilations_0 = const()[name = string("input_57_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_38_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16418304))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16427584))))[name = string("const_38_to_fp16_quantized")];
            tensor<fp16, [1024]> const_39_to_fp16 = const()[name = string("const_39_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16429696)))];
            tensor<fp16, [1, 1024, 188]> input_55_cast_fp16 = transpose(perm = input_55_perm_0, x = h_5_cast_fp16)[name = string("transpose_466")];
            tensor<fp16, [1, 1024, 188]> input_59_cast_fp16 = conv(bias = const_39_to_fp16, dilations = input_57_dilations_0, groups = input_57_groups_0, pad = input_57_pad_0, pad_type = input_57_pad_type_0, strides = input_57_strides_0, weight = const_38_to_fp16_quantized, x = input_55_cast_fp16)[name = string("input_59_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_7_cast_fp16 = silu(x = input_59_cast_fp16)[name = string("h_7_cast_fp16")];
            tensor<int32, [3]> x_43_perm_0 = const()[name = string("x_43_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_1_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16431808))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16825088))))[name = string("blocks_1_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_43_cast_fp16 = transpose(perm = x_43_perm_0, x = h_7_cast_fp16)[name = string("transpose_465")];
            tensor<fp16, [1, 188, 1024]> linear_20_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_1_pw2_weight_to_fp16_palettized, x = x_43_cast_fp16)[name = string("linear_20_cast_fp16")];
            tensor<fp16, [1024]> blocks_1_pw2_scale_to_fp16 = const()[name = string("blocks_1_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16826176)))];
            tensor<fp16, [1, 188, 1024]> var_2818_cast_fp16 = mul(x = linear_20_cast_fp16, y = blocks_1_pw2_scale_to_fp16)[name = string("op_2818_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_61_cast_fp16 = add(x = input_51_cast_fp16, y = var_2818_cast_fp16)[name = string("input_61_cast_fp16")];
            tensor<int32, [1]> x_45_axes_0 = const()[name = string("x_45_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_1_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_1_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16828288)))];
            tensor<fp16, [1024]> enc_layers_1_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_1_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16830400)))];
            tensor<fp16, [1, 188, 1024]> x_45_cast_fp16 = layer_norm(axes = x_45_axes_0, beta = enc_layers_1_norm_feed_forward2_bias_to_fp16, epsilon = var_2698_to_fp16, gamma = enc_layers_1_norm_feed_forward2_weight_to_fp16, x = input_61_cast_fp16)[name = string("x_45_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_1_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16832512))), lut = tensor<fp16, [256, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18405440))))[name = string("blocks_1_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_21_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_1_ff2_1_weight_to_fp16_palettized, x = x_45_cast_fp16)[name = string("linear_21_cast_fp16")];
            tensor<fp16, [4096]> blocks_1_ff2_1_scale_to_fp16 = const()[name = string("blocks_1_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18409600)))];
            tensor<fp16, [1, 188, 4096]> input_63_cast_fp16 = mul(x = linear_21_cast_fp16, y = blocks_1_ff2_1_scale_to_fp16)[name = string("input_63_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_47_cast_fp16 = silu(x = input_63_cast_fp16)[name = string("x_47_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_1_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18417856))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19990784))))[name = string("blocks_1_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_22_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_1_ff2_2_weight_to_fp16_palettized, x = x_47_cast_fp16)[name = string("linear_22_cast_fp16")];
            tensor<fp16, [1024]> blocks_1_ff2_2_scale_to_fp16 = const()[name = string("blocks_1_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19991872)))];
            tensor<fp16, [1, 188, 1024]> var_2832_cast_fp16 = mul(x = linear_22_cast_fp16, y = blocks_1_ff2_2_scale_to_fp16)[name = string("op_2832_cast_fp16")];
            fp16 var_2833_to_fp16 = const()[name = string("op_2833_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2834_cast_fp16 = mul(x = var_2832_cast_fp16, y = var_2833_to_fp16)[name = string("op_2834_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_65_cast_fp16 = add(x = input_61_cast_fp16, y = var_2834_cast_fp16)[name = string("input_65_cast_fp16")];
            tensor<int32, [1]> input_67_axes_0 = const()[name = string("input_67_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_1_norm_out_weight_to_fp16 = const()[name = string("enc_layers_1_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19993984)))];
            tensor<fp16, [1024]> enc_layers_1_norm_out_bias_to_fp16 = const()[name = string("enc_layers_1_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19996096)))];
            tensor<fp16, [1, 188, 1024]> input_67_cast_fp16 = layer_norm(axes = input_67_axes_0, beta = enc_layers_1_norm_out_bias_to_fp16, epsilon = var_2698_to_fp16, gamma = enc_layers_1_norm_out_weight_to_fp16, x = input_65_cast_fp16)[name = string("input_67_cast_fp16")];
            int32 var_2848 = const()[name = string("op_2848"), val = int32(-1)];
            tensor<int32, [1]> x_49_axes_0 = const()[name = string("x_49_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_2_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_2_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19998208)))];
            tensor<fp16, [1024]> enc_layers_2_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_2_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20000320)))];
            fp16 var_2855_to_fp16 = const()[name = string("op_2855_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_49_cast_fp16 = layer_norm(axes = x_49_axes_0, beta = enc_layers_2_norm_feed_forward1_bias_to_fp16, epsilon = var_2855_to_fp16, gamma = enc_layers_2_norm_feed_forward1_weight_to_fp16, x = input_67_cast_fp16)[name = string("x_49_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_2_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20002432))), lut = tensor<fp16, [256, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21575360))))[name = string("blocks_2_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_23_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_2_ff1_1_weight_to_fp16_palettized, x = x_49_cast_fp16)[name = string("linear_23_cast_fp16")];
            tensor<fp16, [4096]> blocks_2_ff1_1_scale_to_fp16 = const()[name = string("blocks_2_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21579520)))];
            tensor<fp16, [1, 188, 4096]> input_69_cast_fp16 = mul(x = linear_23_cast_fp16, y = blocks_2_ff1_1_scale_to_fp16)[name = string("input_69_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_51_cast_fp16 = silu(x = input_69_cast_fp16)[name = string("x_51_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_2_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21587776))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23160704))))[name = string("blocks_2_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_24_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_2_ff1_2_weight_to_fp16_palettized, x = x_51_cast_fp16)[name = string("linear_24_cast_fp16")];
            tensor<fp16, [1024]> blocks_2_ff1_2_scale_to_fp16 = const()[name = string("blocks_2_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23161792)))];
            tensor<fp16, [1, 188, 1024]> var_2880_cast_fp16 = mul(x = linear_24_cast_fp16, y = blocks_2_ff1_2_scale_to_fp16)[name = string("op_2880_cast_fp16")];
            fp16 var_2881_to_fp16 = const()[name = string("op_2881_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2882_cast_fp16 = mul(x = var_2880_cast_fp16, y = var_2881_to_fp16)[name = string("op_2882_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_71_cast_fp16 = add(x = input_67_cast_fp16, y = var_2882_cast_fp16)[name = string("input_71_cast_fp16")];
            tensor<int32, [1]> x_53_axes_0 = const()[name = string("x_53_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_2_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_2_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23163904)))];
            tensor<fp16, [1024]> enc_layers_2_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_2_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23166016)))];
            tensor<fp16, [1, 188, 1024]> x_53_cast_fp16 = layer_norm(axes = x_53_axes_0, beta = enc_layers_2_norm_self_att_bias_to_fp16, epsilon = var_2855_to_fp16, gamma = enc_layers_2_norm_self_att_weight_to_fp16, x = input_71_cast_fp16)[name = string("x_53_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_2_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23168128))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23430336))))[name = string("blocks_2_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_25_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_2_q_weight_to_fp16_palettized, x = x_53_cast_fp16)[name = string("linear_25_cast_fp16")];
            tensor<fp16, [1024]> blocks_2_q_scale_to_fp16 = const()[name = string("blocks_2_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23430912)))];
            tensor<fp16, [1, 188, 1024]> var_2890_cast_fp16 = mul(x = linear_25_cast_fp16, y = blocks_2_q_scale_to_fp16)[name = string("op_2890_cast_fp16")];
            tensor<int32, [4]> var_2891 = const()[name = string("op_2891"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_2892_cast_fp16 = reshape(shape = var_2891, x = var_2890_cast_fp16)[name = string("op_2892_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_2_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23433024))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23695232))))[name = string("blocks_2_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_26_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_2_k_weight_to_fp16_palettized, x = x_53_cast_fp16)[name = string("linear_26_cast_fp16")];
            tensor<fp16, [1024]> blocks_2_k_scale_to_fp16 = const()[name = string("blocks_2_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23695808)))];
            tensor<fp16, [1, 188, 1024]> var_2898_cast_fp16 = mul(x = linear_26_cast_fp16, y = blocks_2_k_scale_to_fp16)[name = string("op_2898_cast_fp16")];
            tensor<int32, [4]> var_2899 = const()[name = string("op_2899"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_2900_cast_fp16 = reshape(shape = var_2899, x = var_2898_cast_fp16)[name = string("op_2900_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_2_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23697920))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23960128))))[name = string("blocks_2_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_27_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_2_v_weight_to_fp16_palettized, x = x_53_cast_fp16)[name = string("linear_27_cast_fp16")];
            tensor<fp16, [1024]> blocks_2_v_scale_to_fp16 = const()[name = string("blocks_2_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23960704)))];
            tensor<fp16, [1, 188, 1024]> var_2906_cast_fp16 = mul(x = linear_27_cast_fp16, y = blocks_2_v_scale_to_fp16)[name = string("op_2906_cast_fp16")];
            tensor<int32, [4]> var_2907 = const()[name = string("op_2907"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_2908_cast_fp16 = reshape(shape = var_2907, x = var_2906_cast_fp16)[name = string("op_2908_cast_fp16")];
            tensor<int32, [4]> v_5_perm_0 = const()[name = string("v_5_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_88_to_fp16 = const()[name = string("const_88_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23962816)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_5_cast_fp16 = add(x = var_2892_cast_fp16, y = const_88_to_fp16)[name = string("q_u_5_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_89_to_fp16 = const()[name = string("const_89_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23964928)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_5_cast_fp16 = add(x = var_2892_cast_fp16, y = const_89_to_fp16)[name = string("q_v_5_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_2_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23967040))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24229248))))[name = string("blocks_2_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_28_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_2_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_28_cast_fp16")];
            tensor<fp16, [1024]> blocks_2_pos_scale_to_fp16 = const()[name = string("blocks_2_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24229824)))];
            tensor<fp16, [1, 375, 1024]> var_2920_cast_fp16 = mul(x = linear_28_cast_fp16, y = blocks_2_pos_scale_to_fp16)[name = string("op_2920_cast_fp16")];
            tensor<int32, [4]> var_2921 = const()[name = string("op_2921"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_5_cast_fp16 = reshape(shape = var_2921, x = var_2920_cast_fp16)[name = string("p_5_cast_fp16")];
            bool input_73_transpose_x_0 = const()[name = string("input_73_transpose_x_0"), val = bool(false)];
            bool input_73_transpose_y_0 = const()[name = string("input_73_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_200_perm_0 = const()[name = string("transpose_200_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_201_perm_0 = const()[name = string("transpose_201_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_201 = transpose(perm = transpose_201_perm_0, x = p_5_cast_fp16)[name = string("transpose_463")];
            tensor<fp16, [1, 8, 188, 128]> transpose_200 = transpose(perm = transpose_200_perm_0, x = q_v_5_cast_fp16)[name = string("transpose_464")];
            tensor<fp16, [1, 8, 188, 375]> input_73_cast_fp16 = matmul(transpose_x = input_73_transpose_x_0, transpose_y = input_73_transpose_y_0, x = transpose_200, y = transpose_201)[name = string("input_73_cast_fp16")];
            tensor<int32, [8]> var_2927_pad_0 = const()[name = string("op_2927_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_2927_mode_0 = const()[name = string("op_2927_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]> var_2927_cast_fp16 = pad(constant_val = const_14_to_fp16, mode = var_2927_mode_0, pad = var_2927_pad_0, x = input_73_cast_fp16)[name = string("op_2927_cast_fp16")];
            tensor<int32, [4]> var_2928 = const()[name = string("op_2928"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_9_cast_fp16 = reshape(shape = var_2928, x = var_2927_cast_fp16)[name = string("scores_9_cast_fp16")];
            tensor<int32, [4]> var_2932_begin_0 = const()[name = string("op_2932_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_2932_end_0 = const()[name = string("op_2932_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_2932_end_mask_0 = const()[name = string("op_2932_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_2932_cast_fp16 = slice_by_index(begin = var_2932_begin_0, end = var_2932_end_0, end_mask = var_2932_end_mask_0, x = scores_9_cast_fp16)[name = string("op_2932_cast_fp16")];
            tensor<int32, [4]> var_2933 = const()[name = string("op_2933"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_2934_cast_fp16 = reshape(shape = var_2933, x = var_2932_cast_fp16)[name = string("op_2934_cast_fp16")];
            tensor<int32, [4]> bd_5_begin_0 = const()[name = string("bd_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_5_end_0 = const()[name = string("bd_5_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_5_end_mask_0 = const()[name = string("bd_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_5_cast_fp16 = slice_by_index(begin = bd_5_begin_0, end = bd_5_end_0, end_mask = bd_5_end_mask_0, x = var_2934_cast_fp16)[name = string("bd_5_cast_fp16")];
            bool var_2937_transpose_x_0 = const()[name = string("op_2937_transpose_x_0"), val = bool(false)];
            bool var_2937_transpose_y_0 = const()[name = string("op_2937_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_202_perm_0 = const()[name = string("transpose_202_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_203_perm_0 = const()[name = string("transpose_203_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_203 = transpose(perm = transpose_203_perm_0, x = var_2900_cast_fp16)[name = string("transpose_461")];
            tensor<fp16, [1, 8, 188, 128]> transpose_202 = transpose(perm = transpose_202_perm_0, x = q_u_5_cast_fp16)[name = string("transpose_462")];
            tensor<fp16, [1, 8, 188, 188]> var_2937_cast_fp16 = matmul(transpose_x = var_2937_transpose_x_0, transpose_y = var_2937_transpose_y_0, x = transpose_202, y = transpose_203)[name = string("op_2937_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2938_cast_fp16 = add(x = var_2937_cast_fp16, y = bd_5_cast_fp16)[name = string("op_2938_cast_fp16")];
            fp16 var_2939_to_fp16 = const()[name = string("op_2939_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_2940_cast_fp16 = mul(x = var_2938_cast_fp16, y = var_2939_to_fp16)[name = string("op_2940_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_11_cast_fp16 = add(x = var_2940_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_11_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2942_cast_fp16 = softmax(axis = var_2848, x = scores_11_cast_fp16)[name = string("op_2942_cast_fp16")];
            bool out_5_transpose_x_0 = const()[name = string("out_5_transpose_x_0"), val = bool(false)];
            bool out_5_transpose_y_0 = const()[name = string("out_5_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_5_cast_fp16 = transpose(perm = v_5_perm_0, x = var_2908_cast_fp16)[name = string("transpose_460")];
            tensor<fp16, [1, 8, 188, 128]> out_5_cast_fp16 = matmul(transpose_x = out_5_transpose_x_0, transpose_y = out_5_transpose_y_0, x = var_2942_cast_fp16, y = v_5_cast_fp16)[name = string("out_5_cast_fp16")];
            tensor<int32, [4]> var_2944_perm_0 = const()[name = string("op_2944_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2945 = const()[name = string("op_2945"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_2944_cast_fp16 = transpose(perm = var_2944_perm_0, x = out_5_cast_fp16)[name = string("transpose_459")];
            tensor<fp16, [1, 188, 1024]> x_55_cast_fp16 = reshape(shape = var_2945, x = var_2944_cast_fp16)[name = string("x_55_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_2_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24231936))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24494144))))[name = string("blocks_2_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_29_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_2_o_weight_to_fp16_palettized, x = x_55_cast_fp16)[name = string("linear_29_cast_fp16")];
            tensor<fp16, [1024]> blocks_2_o_scale_to_fp16 = const()[name = string("blocks_2_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24494720)))];
            tensor<fp16, [1, 188, 1024]> var_2951_cast_fp16 = mul(x = linear_29_cast_fp16, y = blocks_2_o_scale_to_fp16)[name = string("op_2951_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_75_cast_fp16 = add(x = input_71_cast_fp16, y = var_2951_cast_fp16)[name = string("input_75_cast_fp16")];
            tensor<int32, [1]> x_57_axes_0 = const()[name = string("x_57_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_2_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_2_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24496832)))];
            tensor<fp16, [1024]> enc_layers_2_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_2_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24498944)))];
            tensor<fp16, [1, 188, 1024]> x_57_cast_fp16 = layer_norm(axes = x_57_axes_0, beta = enc_layers_2_norm_conv_bias_to_fp16, epsilon = var_2855_to_fp16, gamma = enc_layers_2_norm_conv_weight_to_fp16, x = input_75_cast_fp16)[name = string("x_57_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_2_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24501056))), lut = tensor<fp16, [128, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25287552))))[name = string("blocks_2_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_30_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_2_pw1_weight_to_fp16_palettized, x = x_57_cast_fp16)[name = string("linear_30_cast_fp16")];
            tensor<fp16, [2048]> blocks_2_pw1_scale_to_fp16 = const()[name = string("blocks_2_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25289664)))];
            tensor<fp16, [1, 188, 2048]> input_77_cast_fp16 = mul(x = linear_30_cast_fp16, y = blocks_2_pw1_scale_to_fp16)[name = string("input_77_cast_fp16")];
            int32 var_2960_split_num_splits_0 = const()[name = string("op_2960_split_num_splits_0"), val = int32(2)];
            int32 var_2960_split_axis_0 = const()[name = string("op_2960_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_2960_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_2960_split_cast_fp16_1 = split(axis = var_2960_split_axis_0, num_splits = var_2960_split_num_splits_0, x = input_77_cast_fp16)[name = string("op_2960_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_2960_split_1_sigmoid_cast_fp16 = sigmoid(x = var_2960_split_cast_fp16_1)[name = string("op_2960_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_2960_cast_fp16 = mul(x = var_2960_split_cast_fp16_0, y = var_2960_split_1_sigmoid_cast_fp16)[name = string("op_2960_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_9_cast_fp16 = mul(x = var_2960_cast_fp16, y = valid_cast_fp16)[name = string("h_9_cast_fp16")];
            tensor<int32, [3]> input_79_perm_0 = const()[name = string("input_79_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_81_pad_type_0 = const()[name = string("input_81_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_81_pad_0 = const()[name = string("input_81_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_81_groups_0 = const()[name = string("input_81_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_81_strides_0 = const()[name = string("input_81_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_81_dilations_0 = const()[name = string("input_81_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_40_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25293824))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25303104))))[name = string("const_40_to_fp16_quantized")];
            tensor<fp16, [1024]> const_41_to_fp16 = const()[name = string("const_41_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25305216)))];
            tensor<fp16, [1, 1024, 188]> input_79_cast_fp16 = transpose(perm = input_79_perm_0, x = h_9_cast_fp16)[name = string("transpose_458")];
            tensor<fp16, [1, 1024, 188]> input_83_cast_fp16 = conv(bias = const_41_to_fp16, dilations = input_81_dilations_0, groups = input_81_groups_0, pad = input_81_pad_0, pad_type = input_81_pad_type_0, strides = input_81_strides_0, weight = const_40_to_fp16_quantized, x = input_79_cast_fp16)[name = string("input_83_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_11_cast_fp16 = silu(x = input_83_cast_fp16)[name = string("h_11_cast_fp16")];
            tensor<int32, [3]> x_59_perm_0 = const()[name = string("x_59_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_2_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25307328))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25700608))))[name = string("blocks_2_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_59_cast_fp16 = transpose(perm = x_59_perm_0, x = h_11_cast_fp16)[name = string("transpose_457")];
            tensor<fp16, [1, 188, 1024]> linear_31_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_2_pw2_weight_to_fp16_palettized, x = x_59_cast_fp16)[name = string("linear_31_cast_fp16")];
            tensor<fp16, [1024]> blocks_2_pw2_scale_to_fp16 = const()[name = string("blocks_2_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25701696)))];
            tensor<fp16, [1, 188, 1024]> var_2975_cast_fp16 = mul(x = linear_31_cast_fp16, y = blocks_2_pw2_scale_to_fp16)[name = string("op_2975_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_85_cast_fp16 = add(x = input_75_cast_fp16, y = var_2975_cast_fp16)[name = string("input_85_cast_fp16")];
            tensor<int32, [1]> x_61_axes_0 = const()[name = string("x_61_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_2_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_2_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25703808)))];
            tensor<fp16, [1024]> enc_layers_2_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_2_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25705920)))];
            tensor<fp16, [1, 188, 1024]> x_61_cast_fp16 = layer_norm(axes = x_61_axes_0, beta = enc_layers_2_norm_feed_forward2_bias_to_fp16, epsilon = var_2855_to_fp16, gamma = enc_layers_2_norm_feed_forward2_weight_to_fp16, x = input_85_cast_fp16)[name = string("x_61_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_2_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25708032))), lut = tensor<fp16, [256, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27280960))))[name = string("blocks_2_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_32_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_2_ff2_1_weight_to_fp16_palettized, x = x_61_cast_fp16)[name = string("linear_32_cast_fp16")];
            tensor<fp16, [4096]> blocks_2_ff2_1_scale_to_fp16 = const()[name = string("blocks_2_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27285120)))];
            tensor<fp16, [1, 188, 4096]> input_87_cast_fp16 = mul(x = linear_32_cast_fp16, y = blocks_2_ff2_1_scale_to_fp16)[name = string("input_87_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_63_cast_fp16 = silu(x = input_87_cast_fp16)[name = string("x_63_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_2_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27293376))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28866304))))[name = string("blocks_2_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_33_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_2_ff2_2_weight_to_fp16_palettized, x = x_63_cast_fp16)[name = string("linear_33_cast_fp16")];
            tensor<fp16, [1024]> blocks_2_ff2_2_scale_to_fp16 = const()[name = string("blocks_2_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28867392)))];
            tensor<fp16, [1, 188, 1024]> var_2989_cast_fp16 = mul(x = linear_33_cast_fp16, y = blocks_2_ff2_2_scale_to_fp16)[name = string("op_2989_cast_fp16")];
            fp16 var_2990_to_fp16 = const()[name = string("op_2990_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2991_cast_fp16 = mul(x = var_2989_cast_fp16, y = var_2990_to_fp16)[name = string("op_2991_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_89_cast_fp16 = add(x = input_85_cast_fp16, y = var_2991_cast_fp16)[name = string("input_89_cast_fp16")];
            tensor<int32, [1]> input_91_axes_0 = const()[name = string("input_91_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_2_norm_out_weight_to_fp16 = const()[name = string("enc_layers_2_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28869504)))];
            tensor<fp16, [1024]> enc_layers_2_norm_out_bias_to_fp16 = const()[name = string("enc_layers_2_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28871616)))];
            tensor<fp16, [1, 188, 1024]> input_91_cast_fp16 = layer_norm(axes = input_91_axes_0, beta = enc_layers_2_norm_out_bias_to_fp16, epsilon = var_2855_to_fp16, gamma = enc_layers_2_norm_out_weight_to_fp16, x = input_89_cast_fp16)[name = string("input_91_cast_fp16")];
            int32 var_3005 = const()[name = string("op_3005"), val = int32(-1)];
            tensor<int32, [1]> x_65_axes_0 = const()[name = string("x_65_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_3_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_3_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28873728)))];
            tensor<fp16, [1024]> enc_layers_3_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_3_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28875840)))];
            fp16 var_3012_to_fp16 = const()[name = string("op_3012_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_65_cast_fp16 = layer_norm(axes = x_65_axes_0, beta = enc_layers_3_norm_feed_forward1_bias_to_fp16, epsilon = var_3012_to_fp16, gamma = enc_layers_3_norm_feed_forward1_weight_to_fp16, x = input_91_cast_fp16)[name = string("x_65_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_3_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28877952))), lut = tensor<fp16, [256, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30450880))))[name = string("blocks_3_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_34_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_3_ff1_1_weight_to_fp16_palettized, x = x_65_cast_fp16)[name = string("linear_34_cast_fp16")];
            tensor<fp16, [4096]> blocks_3_ff1_1_scale_to_fp16 = const()[name = string("blocks_3_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30455040)))];
            tensor<fp16, [1, 188, 4096]> input_93_cast_fp16 = mul(x = linear_34_cast_fp16, y = blocks_3_ff1_1_scale_to_fp16)[name = string("input_93_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_67_cast_fp16 = silu(x = input_93_cast_fp16)[name = string("x_67_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_3_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30463296))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32036224))))[name = string("blocks_3_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_35_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_3_ff1_2_weight_to_fp16_palettized, x = x_67_cast_fp16)[name = string("linear_35_cast_fp16")];
            tensor<fp16, [1024]> blocks_3_ff1_2_scale_to_fp16 = const()[name = string("blocks_3_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32037312)))];
            tensor<fp16, [1, 188, 1024]> var_3037_cast_fp16 = mul(x = linear_35_cast_fp16, y = blocks_3_ff1_2_scale_to_fp16)[name = string("op_3037_cast_fp16")];
            fp16 var_3038_to_fp16 = const()[name = string("op_3038_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3039_cast_fp16 = mul(x = var_3037_cast_fp16, y = var_3038_to_fp16)[name = string("op_3039_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_95_cast_fp16 = add(x = input_91_cast_fp16, y = var_3039_cast_fp16)[name = string("input_95_cast_fp16")];
            tensor<int32, [1]> x_69_axes_0 = const()[name = string("x_69_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_3_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_3_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32039424)))];
            tensor<fp16, [1024]> enc_layers_3_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_3_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32041536)))];
            tensor<fp16, [1, 188, 1024]> x_69_cast_fp16 = layer_norm(axes = x_69_axes_0, beta = enc_layers_3_norm_self_att_bias_to_fp16, epsilon = var_3012_to_fp16, gamma = enc_layers_3_norm_self_att_weight_to_fp16, x = input_95_cast_fp16)[name = string("x_69_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_3_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32043648))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32305856))))[name = string("blocks_3_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_36_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_3_q_weight_to_fp16_palettized, x = x_69_cast_fp16)[name = string("linear_36_cast_fp16")];
            tensor<fp16, [1024]> blocks_3_q_scale_to_fp16 = const()[name = string("blocks_3_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32306432)))];
            tensor<fp16, [1, 188, 1024]> var_3047_cast_fp16 = mul(x = linear_36_cast_fp16, y = blocks_3_q_scale_to_fp16)[name = string("op_3047_cast_fp16")];
            tensor<int32, [4]> var_3048 = const()[name = string("op_3048"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3049_cast_fp16 = reshape(shape = var_3048, x = var_3047_cast_fp16)[name = string("op_3049_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_3_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32308544))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32570752))))[name = string("blocks_3_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_37_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_3_k_weight_to_fp16_palettized, x = x_69_cast_fp16)[name = string("linear_37_cast_fp16")];
            tensor<fp16, [1024]> blocks_3_k_scale_to_fp16 = const()[name = string("blocks_3_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32571328)))];
            tensor<fp16, [1, 188, 1024]> var_3055_cast_fp16 = mul(x = linear_37_cast_fp16, y = blocks_3_k_scale_to_fp16)[name = string("op_3055_cast_fp16")];
            tensor<int32, [4]> var_3056 = const()[name = string("op_3056"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3057_cast_fp16 = reshape(shape = var_3056, x = var_3055_cast_fp16)[name = string("op_3057_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_3_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32573440))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32835648))))[name = string("blocks_3_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_38_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_3_v_weight_to_fp16_palettized, x = x_69_cast_fp16)[name = string("linear_38_cast_fp16")];
            tensor<fp16, [1024]> blocks_3_v_scale_to_fp16 = const()[name = string("blocks_3_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32836224)))];
            tensor<fp16, [1, 188, 1024]> var_3063_cast_fp16 = mul(x = linear_38_cast_fp16, y = blocks_3_v_scale_to_fp16)[name = string("op_3063_cast_fp16")];
            tensor<int32, [4]> var_3064 = const()[name = string("op_3064"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3065_cast_fp16 = reshape(shape = var_3064, x = var_3063_cast_fp16)[name = string("op_3065_cast_fp16")];
            tensor<int32, [4]> v_7_perm_0 = const()[name = string("v_7_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_90_to_fp16 = const()[name = string("const_90_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32838336)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_7_cast_fp16 = add(x = var_3049_cast_fp16, y = const_90_to_fp16)[name = string("q_u_7_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_91_to_fp16 = const()[name = string("const_91_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32840448)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_7_cast_fp16 = add(x = var_3049_cast_fp16, y = const_91_to_fp16)[name = string("q_v_7_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_3_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32842560))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33104768))))[name = string("blocks_3_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_39_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_3_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_39_cast_fp16")];
            tensor<fp16, [1024]> blocks_3_pos_scale_to_fp16 = const()[name = string("blocks_3_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33105344)))];
            tensor<fp16, [1, 375, 1024]> var_3077_cast_fp16 = mul(x = linear_39_cast_fp16, y = blocks_3_pos_scale_to_fp16)[name = string("op_3077_cast_fp16")];
            tensor<int32, [4]> var_3078 = const()[name = string("op_3078"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_7_cast_fp16 = reshape(shape = var_3078, x = var_3077_cast_fp16)[name = string("p_7_cast_fp16")];
            bool input_97_transpose_x_0 = const()[name = string("input_97_transpose_x_0"), val = bool(false)];
            bool input_97_transpose_y_0 = const()[name = string("input_97_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_204_perm_0 = const()[name = string("transpose_204_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_205_perm_0 = const()[name = string("transpose_205_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_205 = transpose(perm = transpose_205_perm_0, x = p_7_cast_fp16)[name = string("transpose_455")];
            tensor<fp16, [1, 8, 188, 128]> transpose_204 = transpose(perm = transpose_204_perm_0, x = q_v_7_cast_fp16)[name = string("transpose_456")];
            tensor<fp16, [1, 8, 188, 375]> input_97_cast_fp16 = matmul(transpose_x = input_97_transpose_x_0, transpose_y = input_97_transpose_y_0, x = transpose_204, y = transpose_205)[name = string("input_97_cast_fp16")];
            tensor<int32, [8]> var_3084_pad_0 = const()[name = string("op_3084_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_3084_mode_0 = const()[name = string("op_3084_mode_0"), val = string("constant")];
            fp16 const_15_to_fp16 = const()[name = string("const_15_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_3084_cast_fp16 = pad(constant_val = const_15_to_fp16, mode = var_3084_mode_0, pad = var_3084_pad_0, x = input_97_cast_fp16)[name = string("op_3084_cast_fp16")];
            tensor<int32, [4]> var_3085 = const()[name = string("op_3085"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_13_cast_fp16 = reshape(shape = var_3085, x = var_3084_cast_fp16)[name = string("scores_13_cast_fp16")];
            tensor<int32, [4]> var_3089_begin_0 = const()[name = string("op_3089_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_3089_end_0 = const()[name = string("op_3089_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_3089_end_mask_0 = const()[name = string("op_3089_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_3089_cast_fp16 = slice_by_index(begin = var_3089_begin_0, end = var_3089_end_0, end_mask = var_3089_end_mask_0, x = scores_13_cast_fp16)[name = string("op_3089_cast_fp16")];
            tensor<int32, [4]> var_3090 = const()[name = string("op_3090"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_3091_cast_fp16 = reshape(shape = var_3090, x = var_3089_cast_fp16)[name = string("op_3091_cast_fp16")];
            tensor<int32, [4]> bd_7_begin_0 = const()[name = string("bd_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_7_end_0 = const()[name = string("bd_7_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_7_end_mask_0 = const()[name = string("bd_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_7_cast_fp16 = slice_by_index(begin = bd_7_begin_0, end = bd_7_end_0, end_mask = bd_7_end_mask_0, x = var_3091_cast_fp16)[name = string("bd_7_cast_fp16")];
            bool var_3094_transpose_x_0 = const()[name = string("op_3094_transpose_x_0"), val = bool(false)];
            bool var_3094_transpose_y_0 = const()[name = string("op_3094_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_206_perm_0 = const()[name = string("transpose_206_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_207_perm_0 = const()[name = string("transpose_207_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_207 = transpose(perm = transpose_207_perm_0, x = var_3057_cast_fp16)[name = string("transpose_453")];
            tensor<fp16, [1, 8, 188, 128]> transpose_206 = transpose(perm = transpose_206_perm_0, x = q_u_7_cast_fp16)[name = string("transpose_454")];
            tensor<fp16, [1, 8, 188, 188]> var_3094_cast_fp16 = matmul(transpose_x = var_3094_transpose_x_0, transpose_y = var_3094_transpose_y_0, x = transpose_206, y = transpose_207)[name = string("op_3094_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3095_cast_fp16 = add(x = var_3094_cast_fp16, y = bd_7_cast_fp16)[name = string("op_3095_cast_fp16")];
            fp16 var_3096_to_fp16 = const()[name = string("op_3096_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_3097_cast_fp16 = mul(x = var_3095_cast_fp16, y = var_3096_to_fp16)[name = string("op_3097_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_15_cast_fp16 = add(x = var_3097_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_15_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3099_cast_fp16 = softmax(axis = var_3005, x = scores_15_cast_fp16)[name = string("op_3099_cast_fp16")];
            bool out_7_transpose_x_0 = const()[name = string("out_7_transpose_x_0"), val = bool(false)];
            bool out_7_transpose_y_0 = const()[name = string("out_7_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_7_cast_fp16 = transpose(perm = v_7_perm_0, x = var_3065_cast_fp16)[name = string("transpose_452")];
            tensor<fp16, [1, 8, 188, 128]> out_7_cast_fp16 = matmul(transpose_x = out_7_transpose_x_0, transpose_y = out_7_transpose_y_0, x = var_3099_cast_fp16, y = v_7_cast_fp16)[name = string("out_7_cast_fp16")];
            tensor<int32, [4]> var_3101_perm_0 = const()[name = string("op_3101_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3102 = const()[name = string("op_3102"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_3101_cast_fp16 = transpose(perm = var_3101_perm_0, x = out_7_cast_fp16)[name = string("transpose_451")];
            tensor<fp16, [1, 188, 1024]> x_71_cast_fp16 = reshape(shape = var_3102, x = var_3101_cast_fp16)[name = string("x_71_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_3_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33107456))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33369664))))[name = string("blocks_3_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_40_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_3_o_weight_to_fp16_palettized, x = x_71_cast_fp16)[name = string("linear_40_cast_fp16")];
            tensor<fp16, [1024]> blocks_3_o_scale_to_fp16 = const()[name = string("blocks_3_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33370240)))];
            tensor<fp16, [1, 188, 1024]> var_3108_cast_fp16 = mul(x = linear_40_cast_fp16, y = blocks_3_o_scale_to_fp16)[name = string("op_3108_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_99_cast_fp16 = add(x = input_95_cast_fp16, y = var_3108_cast_fp16)[name = string("input_99_cast_fp16")];
            tensor<int32, [1]> x_73_axes_0 = const()[name = string("x_73_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_3_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_3_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33372352)))];
            tensor<fp16, [1024]> enc_layers_3_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_3_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33374464)))];
            tensor<fp16, [1, 188, 1024]> x_73_cast_fp16 = layer_norm(axes = x_73_axes_0, beta = enc_layers_3_norm_conv_bias_to_fp16, epsilon = var_3012_to_fp16, gamma = enc_layers_3_norm_conv_weight_to_fp16, x = input_99_cast_fp16)[name = string("x_73_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_3_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33376576))), lut = tensor<fp16, [128, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33900928))))[name = string("blocks_3_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_41_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_3_pw1_weight_to_fp16_palettized, x = x_73_cast_fp16)[name = string("linear_41_cast_fp16")];
            tensor<fp16, [2048]> blocks_3_pw1_scale_to_fp16 = const()[name = string("blocks_3_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33902016)))];
            tensor<fp16, [1, 188, 2048]> input_101_cast_fp16 = mul(x = linear_41_cast_fp16, y = blocks_3_pw1_scale_to_fp16)[name = string("input_101_cast_fp16")];
            int32 var_3117_split_num_splits_0 = const()[name = string("op_3117_split_num_splits_0"), val = int32(2)];
            int32 var_3117_split_axis_0 = const()[name = string("op_3117_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_3117_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_3117_split_cast_fp16_1 = split(axis = var_3117_split_axis_0, num_splits = var_3117_split_num_splits_0, x = input_101_cast_fp16)[name = string("op_3117_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_3117_split_1_sigmoid_cast_fp16 = sigmoid(x = var_3117_split_cast_fp16_1)[name = string("op_3117_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_3117_cast_fp16 = mul(x = var_3117_split_cast_fp16_0, y = var_3117_split_1_sigmoid_cast_fp16)[name = string("op_3117_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_13_cast_fp16 = mul(x = var_3117_cast_fp16, y = valid_cast_fp16)[name = string("h_13_cast_fp16")];
            tensor<int32, [3]> input_103_perm_0 = const()[name = string("input_103_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_105_pad_type_0 = const()[name = string("input_105_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_105_pad_0 = const()[name = string("input_105_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_105_groups_0 = const()[name = string("input_105_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_105_strides_0 = const()[name = string("input_105_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_105_dilations_0 = const()[name = string("input_105_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_42_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33906176))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33915456))))[name = string("const_42_to_fp16_quantized")];
            tensor<fp16, [1024]> const_43_to_fp16 = const()[name = string("const_43_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33917568)))];
            tensor<fp16, [1, 1024, 188]> input_103_cast_fp16 = transpose(perm = input_103_perm_0, x = h_13_cast_fp16)[name = string("transpose_450")];
            tensor<fp16, [1, 1024, 188]> input_107_cast_fp16 = conv(bias = const_43_to_fp16, dilations = input_105_dilations_0, groups = input_105_groups_0, pad = input_105_pad_0, pad_type = input_105_pad_type_0, strides = input_105_strides_0, weight = const_42_to_fp16_quantized, x = input_103_cast_fp16)[name = string("input_107_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_15_cast_fp16 = silu(x = input_107_cast_fp16)[name = string("h_15_cast_fp16")];
            tensor<int32, [3]> x_75_perm_0 = const()[name = string("x_75_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_3_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33919680))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34181888))))[name = string("blocks_3_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_75_cast_fp16 = transpose(perm = x_75_perm_0, x = h_15_cast_fp16)[name = string("transpose_449")];
            tensor<fp16, [1, 188, 1024]> linear_42_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_3_pw2_weight_to_fp16_palettized, x = x_75_cast_fp16)[name = string("linear_42_cast_fp16")];
            tensor<fp16, [1024]> blocks_3_pw2_scale_to_fp16 = const()[name = string("blocks_3_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34182464)))];
            tensor<fp16, [1, 188, 1024]> var_3132_cast_fp16 = mul(x = linear_42_cast_fp16, y = blocks_3_pw2_scale_to_fp16)[name = string("op_3132_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_109_cast_fp16 = add(x = input_99_cast_fp16, y = var_3132_cast_fp16)[name = string("input_109_cast_fp16")];
            tensor<int32, [1]> x_77_axes_0 = const()[name = string("x_77_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_3_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_3_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34184576)))];
            tensor<fp16, [1024]> enc_layers_3_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_3_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34186688)))];
            tensor<fp16, [1, 188, 1024]> x_77_cast_fp16 = layer_norm(axes = x_77_axes_0, beta = enc_layers_3_norm_feed_forward2_bias_to_fp16, epsilon = var_3012_to_fp16, gamma = enc_layers_3_norm_feed_forward2_weight_to_fp16, x = input_109_cast_fp16)[name = string("x_77_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_3_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34188800))), lut = tensor<fp16, [256, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35761728))))[name = string("blocks_3_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_43_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_3_ff2_1_weight_to_fp16_palettized, x = x_77_cast_fp16)[name = string("linear_43_cast_fp16")];
            tensor<fp16, [4096]> blocks_3_ff2_1_scale_to_fp16 = const()[name = string("blocks_3_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35765888)))];
            tensor<fp16, [1, 188, 4096]> input_111_cast_fp16 = mul(x = linear_43_cast_fp16, y = blocks_3_ff2_1_scale_to_fp16)[name = string("input_111_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_79_cast_fp16 = silu(x = input_111_cast_fp16)[name = string("x_79_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_3_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35774144))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37347072))))[name = string("blocks_3_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_44_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_3_ff2_2_weight_to_fp16_palettized, x = x_79_cast_fp16)[name = string("linear_44_cast_fp16")];
            tensor<fp16, [1024]> blocks_3_ff2_2_scale_to_fp16 = const()[name = string("blocks_3_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37348160)))];
            tensor<fp16, [1, 188, 1024]> var_3146_cast_fp16 = mul(x = linear_44_cast_fp16, y = blocks_3_ff2_2_scale_to_fp16)[name = string("op_3146_cast_fp16")];
            fp16 var_3147_to_fp16 = const()[name = string("op_3147_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3148_cast_fp16 = mul(x = var_3146_cast_fp16, y = var_3147_to_fp16)[name = string("op_3148_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_113_cast_fp16 = add(x = input_109_cast_fp16, y = var_3148_cast_fp16)[name = string("input_113_cast_fp16")];
            tensor<int32, [1]> input_115_axes_0 = const()[name = string("input_115_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_3_norm_out_weight_to_fp16 = const()[name = string("enc_layers_3_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37350272)))];
            tensor<fp16, [1024]> enc_layers_3_norm_out_bias_to_fp16 = const()[name = string("enc_layers_3_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37352384)))];
            tensor<fp16, [1, 188, 1024]> input_115_cast_fp16 = layer_norm(axes = input_115_axes_0, beta = enc_layers_3_norm_out_bias_to_fp16, epsilon = var_3012_to_fp16, gamma = enc_layers_3_norm_out_weight_to_fp16, x = input_113_cast_fp16)[name = string("input_115_cast_fp16")];
            int32 var_3162 = const()[name = string("op_3162"), val = int32(-1)];
            tensor<int32, [1]> x_81_axes_0 = const()[name = string("x_81_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_4_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_4_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37354496)))];
            tensor<fp16, [1024]> enc_layers_4_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_4_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37356608)))];
            fp16 var_3169_to_fp16 = const()[name = string("op_3169_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_81_cast_fp16 = layer_norm(axes = x_81_axes_0, beta = enc_layers_4_norm_feed_forward1_bias_to_fp16, epsilon = var_3169_to_fp16, gamma = enc_layers_4_norm_feed_forward1_weight_to_fp16, x = input_115_cast_fp16)[name = string("x_81_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_4_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37358720))), lut = tensor<fp16, [256, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38407360))))[name = string("blocks_4_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_45_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_4_ff1_1_weight_to_fp16_palettized, x = x_81_cast_fp16)[name = string("linear_45_cast_fp16")];
            tensor<fp16, [4096]> blocks_4_ff1_1_scale_to_fp16 = const()[name = string("blocks_4_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38409472)))];
            tensor<fp16, [1, 188, 4096]> input_117_cast_fp16 = mul(x = linear_45_cast_fp16, y = blocks_4_ff1_1_scale_to_fp16)[name = string("input_117_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_83_cast_fp16 = silu(x = input_117_cast_fp16)[name = string("x_83_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_4_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38417728))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39466368))))[name = string("blocks_4_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_46_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_4_ff1_2_weight_to_fp16_palettized, x = x_83_cast_fp16)[name = string("linear_46_cast_fp16")];
            tensor<fp16, [1024]> blocks_4_ff1_2_scale_to_fp16 = const()[name = string("blocks_4_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39466944)))];
            tensor<fp16, [1, 188, 1024]> var_3194_cast_fp16 = mul(x = linear_46_cast_fp16, y = blocks_4_ff1_2_scale_to_fp16)[name = string("op_3194_cast_fp16")];
            fp16 var_3195_to_fp16 = const()[name = string("op_3195_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3196_cast_fp16 = mul(x = var_3194_cast_fp16, y = var_3195_to_fp16)[name = string("op_3196_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_119_cast_fp16 = add(x = input_115_cast_fp16, y = var_3196_cast_fp16)[name = string("input_119_cast_fp16")];
            tensor<int32, [1]> x_85_axes_0 = const()[name = string("x_85_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_4_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_4_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39469056)))];
            tensor<fp16, [1024]> enc_layers_4_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_4_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39471168)))];
            tensor<fp16, [1, 188, 1024]> x_85_cast_fp16 = layer_norm(axes = x_85_axes_0, beta = enc_layers_4_norm_self_att_bias_to_fp16, epsilon = var_3169_to_fp16, gamma = enc_layers_4_norm_self_att_weight_to_fp16, x = input_119_cast_fp16)[name = string("x_85_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_4_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39473280))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39735488))))[name = string("blocks_4_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_47_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_4_q_weight_to_fp16_palettized, x = x_85_cast_fp16)[name = string("linear_47_cast_fp16")];
            tensor<fp16, [1024]> blocks_4_q_scale_to_fp16 = const()[name = string("blocks_4_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39736064)))];
            tensor<fp16, [1, 188, 1024]> var_3204_cast_fp16 = mul(x = linear_47_cast_fp16, y = blocks_4_q_scale_to_fp16)[name = string("op_3204_cast_fp16")];
            tensor<int32, [4]> var_3205 = const()[name = string("op_3205"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3206_cast_fp16 = reshape(shape = var_3205, x = var_3204_cast_fp16)[name = string("op_3206_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_4_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39738176))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40000384))))[name = string("blocks_4_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_48_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_4_k_weight_to_fp16_palettized, x = x_85_cast_fp16)[name = string("linear_48_cast_fp16")];
            tensor<fp16, [1024]> blocks_4_k_scale_to_fp16 = const()[name = string("blocks_4_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40000960)))];
            tensor<fp16, [1, 188, 1024]> var_3212_cast_fp16 = mul(x = linear_48_cast_fp16, y = blocks_4_k_scale_to_fp16)[name = string("op_3212_cast_fp16")];
            tensor<int32, [4]> var_3213 = const()[name = string("op_3213"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3214_cast_fp16 = reshape(shape = var_3213, x = var_3212_cast_fp16)[name = string("op_3214_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_4_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40003072))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40265280))))[name = string("blocks_4_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_49_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_4_v_weight_to_fp16_palettized, x = x_85_cast_fp16)[name = string("linear_49_cast_fp16")];
            tensor<fp16, [1024]> blocks_4_v_scale_to_fp16 = const()[name = string("blocks_4_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40265856)))];
            tensor<fp16, [1, 188, 1024]> var_3220_cast_fp16 = mul(x = linear_49_cast_fp16, y = blocks_4_v_scale_to_fp16)[name = string("op_3220_cast_fp16")];
            tensor<int32, [4]> var_3221 = const()[name = string("op_3221"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3222_cast_fp16 = reshape(shape = var_3221, x = var_3220_cast_fp16)[name = string("op_3222_cast_fp16")];
            tensor<int32, [4]> v_9_perm_0 = const()[name = string("v_9_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_92_to_fp16 = const()[name = string("const_92_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40267968)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_9_cast_fp16 = add(x = var_3206_cast_fp16, y = const_92_to_fp16)[name = string("q_u_9_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_93_to_fp16 = const()[name = string("const_93_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40270080)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_9_cast_fp16 = add(x = var_3206_cast_fp16, y = const_93_to_fp16)[name = string("q_v_9_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_4_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40272192))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40534400))))[name = string("blocks_4_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_50_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_4_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_50_cast_fp16")];
            tensor<fp16, [1024]> blocks_4_pos_scale_to_fp16 = const()[name = string("blocks_4_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40534976)))];
            tensor<fp16, [1, 375, 1024]> var_3234_cast_fp16 = mul(x = linear_50_cast_fp16, y = blocks_4_pos_scale_to_fp16)[name = string("op_3234_cast_fp16")];
            tensor<int32, [4]> var_3235 = const()[name = string("op_3235"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_9_cast_fp16 = reshape(shape = var_3235, x = var_3234_cast_fp16)[name = string("p_9_cast_fp16")];
            bool input_121_transpose_x_0 = const()[name = string("input_121_transpose_x_0"), val = bool(false)];
            bool input_121_transpose_y_0 = const()[name = string("input_121_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_208_perm_0 = const()[name = string("transpose_208_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_209_perm_0 = const()[name = string("transpose_209_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_209 = transpose(perm = transpose_209_perm_0, x = p_9_cast_fp16)[name = string("transpose_447")];
            tensor<fp16, [1, 8, 188, 128]> transpose_208 = transpose(perm = transpose_208_perm_0, x = q_v_9_cast_fp16)[name = string("transpose_448")];
            tensor<fp16, [1, 8, 188, 375]> input_121_cast_fp16 = matmul(transpose_x = input_121_transpose_x_0, transpose_y = input_121_transpose_y_0, x = transpose_208, y = transpose_209)[name = string("input_121_cast_fp16")];
            tensor<int32, [8]> var_3241_pad_0 = const()[name = string("op_3241_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_3241_mode_0 = const()[name = string("op_3241_mode_0"), val = string("constant")];
            fp16 const_16_to_fp16 = const()[name = string("const_16_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_3241_cast_fp16 = pad(constant_val = const_16_to_fp16, mode = var_3241_mode_0, pad = var_3241_pad_0, x = input_121_cast_fp16)[name = string("op_3241_cast_fp16")];
            tensor<int32, [4]> var_3242 = const()[name = string("op_3242"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_17_cast_fp16 = reshape(shape = var_3242, x = var_3241_cast_fp16)[name = string("scores_17_cast_fp16")];
            tensor<int32, [4]> var_3246_begin_0 = const()[name = string("op_3246_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_3246_end_0 = const()[name = string("op_3246_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_3246_end_mask_0 = const()[name = string("op_3246_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_3246_cast_fp16 = slice_by_index(begin = var_3246_begin_0, end = var_3246_end_0, end_mask = var_3246_end_mask_0, x = scores_17_cast_fp16)[name = string("op_3246_cast_fp16")];
            tensor<int32, [4]> var_3247 = const()[name = string("op_3247"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_3248_cast_fp16 = reshape(shape = var_3247, x = var_3246_cast_fp16)[name = string("op_3248_cast_fp16")];
            tensor<int32, [4]> bd_9_begin_0 = const()[name = string("bd_9_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_9_end_0 = const()[name = string("bd_9_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_9_end_mask_0 = const()[name = string("bd_9_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_9_cast_fp16 = slice_by_index(begin = bd_9_begin_0, end = bd_9_end_0, end_mask = bd_9_end_mask_0, x = var_3248_cast_fp16)[name = string("bd_9_cast_fp16")];
            bool var_3251_transpose_x_0 = const()[name = string("op_3251_transpose_x_0"), val = bool(false)];
            bool var_3251_transpose_y_0 = const()[name = string("op_3251_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_210_perm_0 = const()[name = string("transpose_210_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_211_perm_0 = const()[name = string("transpose_211_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_211 = transpose(perm = transpose_211_perm_0, x = var_3214_cast_fp16)[name = string("transpose_445")];
            tensor<fp16, [1, 8, 188, 128]> transpose_210 = transpose(perm = transpose_210_perm_0, x = q_u_9_cast_fp16)[name = string("transpose_446")];
            tensor<fp16, [1, 8, 188, 188]> var_3251_cast_fp16 = matmul(transpose_x = var_3251_transpose_x_0, transpose_y = var_3251_transpose_y_0, x = transpose_210, y = transpose_211)[name = string("op_3251_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3252_cast_fp16 = add(x = var_3251_cast_fp16, y = bd_9_cast_fp16)[name = string("op_3252_cast_fp16")];
            fp16 var_3253_to_fp16 = const()[name = string("op_3253_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_3254_cast_fp16 = mul(x = var_3252_cast_fp16, y = var_3253_to_fp16)[name = string("op_3254_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_19_cast_fp16 = add(x = var_3254_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_19_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3256_cast_fp16 = softmax(axis = var_3162, x = scores_19_cast_fp16)[name = string("op_3256_cast_fp16")];
            bool out_9_transpose_x_0 = const()[name = string("out_9_transpose_x_0"), val = bool(false)];
            bool out_9_transpose_y_0 = const()[name = string("out_9_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_9_cast_fp16 = transpose(perm = v_9_perm_0, x = var_3222_cast_fp16)[name = string("transpose_444")];
            tensor<fp16, [1, 8, 188, 128]> out_9_cast_fp16 = matmul(transpose_x = out_9_transpose_x_0, transpose_y = out_9_transpose_y_0, x = var_3256_cast_fp16, y = v_9_cast_fp16)[name = string("out_9_cast_fp16")];
            tensor<int32, [4]> var_3258_perm_0 = const()[name = string("op_3258_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3259 = const()[name = string("op_3259"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_3258_cast_fp16 = transpose(perm = var_3258_perm_0, x = out_9_cast_fp16)[name = string("transpose_443")];
            tensor<fp16, [1, 188, 1024]> x_87_cast_fp16 = reshape(shape = var_3259, x = var_3258_cast_fp16)[name = string("x_87_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_4_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40537088))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40799296))))[name = string("blocks_4_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_51_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_4_o_weight_to_fp16_palettized, x = x_87_cast_fp16)[name = string("linear_51_cast_fp16")];
            tensor<fp16, [1024]> blocks_4_o_scale_to_fp16 = const()[name = string("blocks_4_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40799872)))];
            tensor<fp16, [1, 188, 1024]> var_3265_cast_fp16 = mul(x = linear_51_cast_fp16, y = blocks_4_o_scale_to_fp16)[name = string("op_3265_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_123_cast_fp16 = add(x = input_119_cast_fp16, y = var_3265_cast_fp16)[name = string("input_123_cast_fp16")];
            tensor<int32, [1]> x_89_axes_0 = const()[name = string("x_89_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_4_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_4_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40801984)))];
            tensor<fp16, [1024]> enc_layers_4_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_4_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40804096)))];
            tensor<fp16, [1, 188, 1024]> x_89_cast_fp16 = layer_norm(axes = x_89_axes_0, beta = enc_layers_4_norm_conv_bias_to_fp16, epsilon = var_3169_to_fp16, gamma = enc_layers_4_norm_conv_weight_to_fp16, x = input_123_cast_fp16)[name = string("x_89_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_4_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40806208))), lut = tensor<fp16, [128, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41592704))))[name = string("blocks_4_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_52_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_4_pw1_weight_to_fp16_palettized, x = x_89_cast_fp16)[name = string("linear_52_cast_fp16")];
            tensor<fp16, [2048]> blocks_4_pw1_scale_to_fp16 = const()[name = string("blocks_4_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41594816)))];
            tensor<fp16, [1, 188, 2048]> input_125_cast_fp16 = mul(x = linear_52_cast_fp16, y = blocks_4_pw1_scale_to_fp16)[name = string("input_125_cast_fp16")];
            int32 var_3274_split_num_splits_0 = const()[name = string("op_3274_split_num_splits_0"), val = int32(2)];
            int32 var_3274_split_axis_0 = const()[name = string("op_3274_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_3274_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_3274_split_cast_fp16_1 = split(axis = var_3274_split_axis_0, num_splits = var_3274_split_num_splits_0, x = input_125_cast_fp16)[name = string("op_3274_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_3274_split_1_sigmoid_cast_fp16 = sigmoid(x = var_3274_split_cast_fp16_1)[name = string("op_3274_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_3274_cast_fp16 = mul(x = var_3274_split_cast_fp16_0, y = var_3274_split_1_sigmoid_cast_fp16)[name = string("op_3274_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_17_cast_fp16 = mul(x = var_3274_cast_fp16, y = valid_cast_fp16)[name = string("h_17_cast_fp16")];
            tensor<int32, [3]> input_127_perm_0 = const()[name = string("input_127_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_129_pad_type_0 = const()[name = string("input_129_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_129_pad_0 = const()[name = string("input_129_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_129_groups_0 = const()[name = string("input_129_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_129_strides_0 = const()[name = string("input_129_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_129_dilations_0 = const()[name = string("input_129_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_44_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41598976))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41608256))))[name = string("const_44_to_fp16_quantized")];
            tensor<fp16, [1024]> const_45_to_fp16 = const()[name = string("const_45_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41610368)))];
            tensor<fp16, [1, 1024, 188]> input_127_cast_fp16 = transpose(perm = input_127_perm_0, x = h_17_cast_fp16)[name = string("transpose_442")];
            tensor<fp16, [1, 1024, 188]> input_131_cast_fp16 = conv(bias = const_45_to_fp16, dilations = input_129_dilations_0, groups = input_129_groups_0, pad = input_129_pad_0, pad_type = input_129_pad_type_0, strides = input_129_strides_0, weight = const_44_to_fp16_quantized, x = input_127_cast_fp16)[name = string("input_131_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_19_cast_fp16 = silu(x = input_131_cast_fp16)[name = string("h_19_cast_fp16")];
            tensor<int32, [3]> x_91_perm_0 = const()[name = string("x_91_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_4_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41612480))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42005760))))[name = string("blocks_4_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_91_cast_fp16 = transpose(perm = x_91_perm_0, x = h_19_cast_fp16)[name = string("transpose_441")];
            tensor<fp16, [1, 188, 1024]> linear_53_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_4_pw2_weight_to_fp16_palettized, x = x_91_cast_fp16)[name = string("linear_53_cast_fp16")];
            tensor<fp16, [1024]> blocks_4_pw2_scale_to_fp16 = const()[name = string("blocks_4_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42006848)))];
            tensor<fp16, [1, 188, 1024]> var_3289_cast_fp16 = mul(x = linear_53_cast_fp16, y = blocks_4_pw2_scale_to_fp16)[name = string("op_3289_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_133_cast_fp16 = add(x = input_123_cast_fp16, y = var_3289_cast_fp16)[name = string("input_133_cast_fp16")];
            tensor<int32, [1]> x_93_axes_0 = const()[name = string("x_93_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_4_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_4_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42008960)))];
            tensor<fp16, [1024]> enc_layers_4_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_4_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42011072)))];
            tensor<fp16, [1, 188, 1024]> x_93_cast_fp16 = layer_norm(axes = x_93_axes_0, beta = enc_layers_4_norm_feed_forward2_bias_to_fp16, epsilon = var_3169_to_fp16, gamma = enc_layers_4_norm_feed_forward2_weight_to_fp16, x = input_133_cast_fp16)[name = string("x_93_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_4_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42013184))), lut = tensor<fp16, [256, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43061824))))[name = string("blocks_4_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_54_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_4_ff2_1_weight_to_fp16_palettized, x = x_93_cast_fp16)[name = string("linear_54_cast_fp16")];
            tensor<fp16, [4096]> blocks_4_ff2_1_scale_to_fp16 = const()[name = string("blocks_4_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43063936)))];
            tensor<fp16, [1, 188, 4096]> input_135_cast_fp16 = mul(x = linear_54_cast_fp16, y = blocks_4_ff2_1_scale_to_fp16)[name = string("input_135_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_95_cast_fp16 = silu(x = input_135_cast_fp16)[name = string("x_95_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_4_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43072192))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44120832))))[name = string("blocks_4_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_55_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_4_ff2_2_weight_to_fp16_palettized, x = x_95_cast_fp16)[name = string("linear_55_cast_fp16")];
            tensor<fp16, [1024]> blocks_4_ff2_2_scale_to_fp16 = const()[name = string("blocks_4_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44121408)))];
            tensor<fp16, [1, 188, 1024]> var_3303_cast_fp16 = mul(x = linear_55_cast_fp16, y = blocks_4_ff2_2_scale_to_fp16)[name = string("op_3303_cast_fp16")];
            fp16 var_3304_to_fp16 = const()[name = string("op_3304_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3305_cast_fp16 = mul(x = var_3303_cast_fp16, y = var_3304_to_fp16)[name = string("op_3305_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_137_cast_fp16 = add(x = input_133_cast_fp16, y = var_3305_cast_fp16)[name = string("input_137_cast_fp16")];
            tensor<int32, [1]> input_139_axes_0 = const()[name = string("input_139_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_4_norm_out_weight_to_fp16 = const()[name = string("enc_layers_4_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44123520)))];
            tensor<fp16, [1024]> enc_layers_4_norm_out_bias_to_fp16 = const()[name = string("enc_layers_4_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44125632)))];
            tensor<fp16, [1, 188, 1024]> input_139_cast_fp16 = layer_norm(axes = input_139_axes_0, beta = enc_layers_4_norm_out_bias_to_fp16, epsilon = var_3169_to_fp16, gamma = enc_layers_4_norm_out_weight_to_fp16, x = input_137_cast_fp16)[name = string("input_139_cast_fp16")];
            int32 var_3319 = const()[name = string("op_3319"), val = int32(-1)];
            tensor<int32, [1]> x_97_axes_0 = const()[name = string("x_97_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_5_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_5_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44127744)))];
            tensor<fp16, [1024]> enc_layers_5_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_5_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44129856)))];
            fp16 var_3326_to_fp16 = const()[name = string("op_3326_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_97_cast_fp16 = layer_norm(axes = x_97_axes_0, beta = enc_layers_5_norm_feed_forward1_bias_to_fp16, epsilon = var_3326_to_fp16, gamma = enc_layers_5_norm_feed_forward1_weight_to_fp16, x = input_139_cast_fp16)[name = string("x_97_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_5_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44131968))), lut = tensor<fp16, [256, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45180608))))[name = string("blocks_5_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_56_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_5_ff1_1_weight_to_fp16_palettized, x = x_97_cast_fp16)[name = string("linear_56_cast_fp16")];
            tensor<fp16, [4096]> blocks_5_ff1_1_scale_to_fp16 = const()[name = string("blocks_5_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45182720)))];
            tensor<fp16, [1, 188, 4096]> input_141_cast_fp16 = mul(x = linear_56_cast_fp16, y = blocks_5_ff1_1_scale_to_fp16)[name = string("input_141_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_99_cast_fp16 = silu(x = input_141_cast_fp16)[name = string("x_99_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_5_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45190976))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46239616))))[name = string("blocks_5_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_57_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_5_ff1_2_weight_to_fp16_palettized, x = x_99_cast_fp16)[name = string("linear_57_cast_fp16")];
            tensor<fp16, [1024]> blocks_5_ff1_2_scale_to_fp16 = const()[name = string("blocks_5_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46240192)))];
            tensor<fp16, [1, 188, 1024]> var_3351_cast_fp16 = mul(x = linear_57_cast_fp16, y = blocks_5_ff1_2_scale_to_fp16)[name = string("op_3351_cast_fp16")];
            fp16 var_3352_to_fp16 = const()[name = string("op_3352_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3353_cast_fp16 = mul(x = var_3351_cast_fp16, y = var_3352_to_fp16)[name = string("op_3353_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_143_cast_fp16 = add(x = input_139_cast_fp16, y = var_3353_cast_fp16)[name = string("input_143_cast_fp16")];
            tensor<int32, [1]> x_101_axes_0 = const()[name = string("x_101_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_5_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_5_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46242304)))];
            tensor<fp16, [1024]> enc_layers_5_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_5_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46244416)))];
            tensor<fp16, [1, 188, 1024]> x_101_cast_fp16 = layer_norm(axes = x_101_axes_0, beta = enc_layers_5_norm_self_att_bias_to_fp16, epsilon = var_3326_to_fp16, gamma = enc_layers_5_norm_self_att_weight_to_fp16, x = input_143_cast_fp16)[name = string("x_101_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_5_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46246528))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46508736))))[name = string("blocks_5_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_58_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_5_q_weight_to_fp16_palettized, x = x_101_cast_fp16)[name = string("linear_58_cast_fp16")];
            tensor<fp16, [1024]> blocks_5_q_scale_to_fp16 = const()[name = string("blocks_5_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46509312)))];
            tensor<fp16, [1, 188, 1024]> var_3361_cast_fp16 = mul(x = linear_58_cast_fp16, y = blocks_5_q_scale_to_fp16)[name = string("op_3361_cast_fp16")];
            tensor<int32, [4]> var_3362 = const()[name = string("op_3362"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3363_cast_fp16 = reshape(shape = var_3362, x = var_3361_cast_fp16)[name = string("op_3363_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_5_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46511424))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46773632))))[name = string("blocks_5_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_59_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_5_k_weight_to_fp16_palettized, x = x_101_cast_fp16)[name = string("linear_59_cast_fp16")];
            tensor<fp16, [1024]> blocks_5_k_scale_to_fp16 = const()[name = string("blocks_5_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46774208)))];
            tensor<fp16, [1, 188, 1024]> var_3369_cast_fp16 = mul(x = linear_59_cast_fp16, y = blocks_5_k_scale_to_fp16)[name = string("op_3369_cast_fp16")];
            tensor<int32, [4]> var_3370 = const()[name = string("op_3370"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3371_cast_fp16 = reshape(shape = var_3370, x = var_3369_cast_fp16)[name = string("op_3371_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_5_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46776320))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47038528))))[name = string("blocks_5_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_60_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_5_v_weight_to_fp16_palettized, x = x_101_cast_fp16)[name = string("linear_60_cast_fp16")];
            tensor<fp16, [1024]> blocks_5_v_scale_to_fp16 = const()[name = string("blocks_5_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47039104)))];
            tensor<fp16, [1, 188, 1024]> var_3377_cast_fp16 = mul(x = linear_60_cast_fp16, y = blocks_5_v_scale_to_fp16)[name = string("op_3377_cast_fp16")];
            tensor<int32, [4]> var_3378 = const()[name = string("op_3378"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3379_cast_fp16 = reshape(shape = var_3378, x = var_3377_cast_fp16)[name = string("op_3379_cast_fp16")];
            tensor<int32, [4]> v_11_perm_0 = const()[name = string("v_11_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_94_to_fp16 = const()[name = string("const_94_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47041216)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_11_cast_fp16 = add(x = var_3363_cast_fp16, y = const_94_to_fp16)[name = string("q_u_11_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_95_to_fp16 = const()[name = string("const_95_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47043328)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_11_cast_fp16 = add(x = var_3363_cast_fp16, y = const_95_to_fp16)[name = string("q_v_11_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_5_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47045440))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47307648))))[name = string("blocks_5_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_61_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_5_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_61_cast_fp16")];
            tensor<fp16, [1024]> blocks_5_pos_scale_to_fp16 = const()[name = string("blocks_5_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47308224)))];
            tensor<fp16, [1, 375, 1024]> var_3391_cast_fp16 = mul(x = linear_61_cast_fp16, y = blocks_5_pos_scale_to_fp16)[name = string("op_3391_cast_fp16")];
            tensor<int32, [4]> var_3392 = const()[name = string("op_3392"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_11_cast_fp16 = reshape(shape = var_3392, x = var_3391_cast_fp16)[name = string("p_11_cast_fp16")];
            bool input_145_transpose_x_0 = const()[name = string("input_145_transpose_x_0"), val = bool(false)];
            bool input_145_transpose_y_0 = const()[name = string("input_145_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_212_perm_0 = const()[name = string("transpose_212_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_213_perm_0 = const()[name = string("transpose_213_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_213 = transpose(perm = transpose_213_perm_0, x = p_11_cast_fp16)[name = string("transpose_439")];
            tensor<fp16, [1, 8, 188, 128]> transpose_212 = transpose(perm = transpose_212_perm_0, x = q_v_11_cast_fp16)[name = string("transpose_440")];
            tensor<fp16, [1, 8, 188, 375]> input_145_cast_fp16 = matmul(transpose_x = input_145_transpose_x_0, transpose_y = input_145_transpose_y_0, x = transpose_212, y = transpose_213)[name = string("input_145_cast_fp16")];
            tensor<int32, [8]> var_3398_pad_0 = const()[name = string("op_3398_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_3398_mode_0 = const()[name = string("op_3398_mode_0"), val = string("constant")];
            fp16 const_17_to_fp16 = const()[name = string("const_17_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_3398_cast_fp16 = pad(constant_val = const_17_to_fp16, mode = var_3398_mode_0, pad = var_3398_pad_0, x = input_145_cast_fp16)[name = string("op_3398_cast_fp16")];
            tensor<int32, [4]> var_3399 = const()[name = string("op_3399"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_21_cast_fp16 = reshape(shape = var_3399, x = var_3398_cast_fp16)[name = string("scores_21_cast_fp16")];
            tensor<int32, [4]> var_3403_begin_0 = const()[name = string("op_3403_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_3403_end_0 = const()[name = string("op_3403_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_3403_end_mask_0 = const()[name = string("op_3403_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_3403_cast_fp16 = slice_by_index(begin = var_3403_begin_0, end = var_3403_end_0, end_mask = var_3403_end_mask_0, x = scores_21_cast_fp16)[name = string("op_3403_cast_fp16")];
            tensor<int32, [4]> var_3404 = const()[name = string("op_3404"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_3405_cast_fp16 = reshape(shape = var_3404, x = var_3403_cast_fp16)[name = string("op_3405_cast_fp16")];
            tensor<int32, [4]> bd_11_begin_0 = const()[name = string("bd_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_11_end_0 = const()[name = string("bd_11_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_11_end_mask_0 = const()[name = string("bd_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_11_cast_fp16 = slice_by_index(begin = bd_11_begin_0, end = bd_11_end_0, end_mask = bd_11_end_mask_0, x = var_3405_cast_fp16)[name = string("bd_11_cast_fp16")];
            bool var_3408_transpose_x_0 = const()[name = string("op_3408_transpose_x_0"), val = bool(false)];
            bool var_3408_transpose_y_0 = const()[name = string("op_3408_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_214_perm_0 = const()[name = string("transpose_214_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_215_perm_0 = const()[name = string("transpose_215_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_215 = transpose(perm = transpose_215_perm_0, x = var_3371_cast_fp16)[name = string("transpose_437")];
            tensor<fp16, [1, 8, 188, 128]> transpose_214 = transpose(perm = transpose_214_perm_0, x = q_u_11_cast_fp16)[name = string("transpose_438")];
            tensor<fp16, [1, 8, 188, 188]> var_3408_cast_fp16 = matmul(transpose_x = var_3408_transpose_x_0, transpose_y = var_3408_transpose_y_0, x = transpose_214, y = transpose_215)[name = string("op_3408_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3409_cast_fp16 = add(x = var_3408_cast_fp16, y = bd_11_cast_fp16)[name = string("op_3409_cast_fp16")];
            fp16 var_3410_to_fp16 = const()[name = string("op_3410_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_3411_cast_fp16 = mul(x = var_3409_cast_fp16, y = var_3410_to_fp16)[name = string("op_3411_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_23_cast_fp16 = add(x = var_3411_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_23_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3413_cast_fp16 = softmax(axis = var_3319, x = scores_23_cast_fp16)[name = string("op_3413_cast_fp16")];
            bool out_11_transpose_x_0 = const()[name = string("out_11_transpose_x_0"), val = bool(false)];
            bool out_11_transpose_y_0 = const()[name = string("out_11_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_11_cast_fp16 = transpose(perm = v_11_perm_0, x = var_3379_cast_fp16)[name = string("transpose_436")];
            tensor<fp16, [1, 8, 188, 128]> out_11_cast_fp16 = matmul(transpose_x = out_11_transpose_x_0, transpose_y = out_11_transpose_y_0, x = var_3413_cast_fp16, y = v_11_cast_fp16)[name = string("out_11_cast_fp16")];
            tensor<int32, [4]> var_3415_perm_0 = const()[name = string("op_3415_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3416 = const()[name = string("op_3416"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_3415_cast_fp16 = transpose(perm = var_3415_perm_0, x = out_11_cast_fp16)[name = string("transpose_435")];
            tensor<fp16, [1, 188, 1024]> x_103_cast_fp16 = reshape(shape = var_3416, x = var_3415_cast_fp16)[name = string("x_103_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_5_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47310336))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47572544))))[name = string("blocks_5_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_62_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_5_o_weight_to_fp16_palettized, x = x_103_cast_fp16)[name = string("linear_62_cast_fp16")];
            tensor<fp16, [1024]> blocks_5_o_scale_to_fp16 = const()[name = string("blocks_5_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47573120)))];
            tensor<fp16, [1, 188, 1024]> var_3422_cast_fp16 = mul(x = linear_62_cast_fp16, y = blocks_5_o_scale_to_fp16)[name = string("op_3422_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_147_cast_fp16 = add(x = input_143_cast_fp16, y = var_3422_cast_fp16)[name = string("input_147_cast_fp16")];
            tensor<int32, [1]> x_105_axes_0 = const()[name = string("x_105_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_5_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_5_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47575232)))];
            tensor<fp16, [1024]> enc_layers_5_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_5_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47577344)))];
            tensor<fp16, [1, 188, 1024]> x_105_cast_fp16 = layer_norm(axes = x_105_axes_0, beta = enc_layers_5_norm_conv_bias_to_fp16, epsilon = var_3326_to_fp16, gamma = enc_layers_5_norm_conv_weight_to_fp16, x = input_147_cast_fp16)[name = string("x_105_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_5_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47579456))), lut = tensor<fp16, [128, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48103808))))[name = string("blocks_5_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_63_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_5_pw1_weight_to_fp16_palettized, x = x_105_cast_fp16)[name = string("linear_63_cast_fp16")];
            tensor<fp16, [2048]> blocks_5_pw1_scale_to_fp16 = const()[name = string("blocks_5_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48104896)))];
            tensor<fp16, [1, 188, 2048]> input_149_cast_fp16 = mul(x = linear_63_cast_fp16, y = blocks_5_pw1_scale_to_fp16)[name = string("input_149_cast_fp16")];
            int32 var_3431_split_num_splits_0 = const()[name = string("op_3431_split_num_splits_0"), val = int32(2)];
            int32 var_3431_split_axis_0 = const()[name = string("op_3431_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_3431_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_3431_split_cast_fp16_1 = split(axis = var_3431_split_axis_0, num_splits = var_3431_split_num_splits_0, x = input_149_cast_fp16)[name = string("op_3431_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_3431_split_1_sigmoid_cast_fp16 = sigmoid(x = var_3431_split_cast_fp16_1)[name = string("op_3431_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_3431_cast_fp16 = mul(x = var_3431_split_cast_fp16_0, y = var_3431_split_1_sigmoid_cast_fp16)[name = string("op_3431_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_21_cast_fp16 = mul(x = var_3431_cast_fp16, y = valid_cast_fp16)[name = string("h_21_cast_fp16")];
            tensor<int32, [3]> input_151_perm_0 = const()[name = string("input_151_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_153_pad_type_0 = const()[name = string("input_153_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_153_pad_0 = const()[name = string("input_153_pad_0"), val = tensor<int32, [2]>([4, 4])];
            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, [1]> input_153_dilations_0 = const()[name = string("input_153_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_46_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48109056))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48118336))))[name = string("const_46_to_fp16_quantized")];
            tensor<fp16, [1024]> const_47_to_fp16 = const()[name = string("const_47_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48120448)))];
            tensor<fp16, [1, 1024, 188]> input_151_cast_fp16 = transpose(perm = input_151_perm_0, x = h_21_cast_fp16)[name = string("transpose_434")];
            tensor<fp16, [1, 1024, 188]> input_155_cast_fp16 = conv(bias = const_47_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_46_to_fp16_quantized, x = input_151_cast_fp16)[name = string("input_155_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_23_cast_fp16 = silu(x = input_155_cast_fp16)[name = string("h_23_cast_fp16")];
            tensor<int32, [3]> x_107_perm_0 = const()[name = string("x_107_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_5_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48122560))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48384768))))[name = string("blocks_5_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_107_cast_fp16 = transpose(perm = x_107_perm_0, x = h_23_cast_fp16)[name = string("transpose_433")];
            tensor<fp16, [1, 188, 1024]> linear_64_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_5_pw2_weight_to_fp16_palettized, x = x_107_cast_fp16)[name = string("linear_64_cast_fp16")];
            tensor<fp16, [1024]> blocks_5_pw2_scale_to_fp16 = const()[name = string("blocks_5_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48385344)))];
            tensor<fp16, [1, 188, 1024]> var_3446_cast_fp16 = mul(x = linear_64_cast_fp16, y = blocks_5_pw2_scale_to_fp16)[name = string("op_3446_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_157_cast_fp16 = add(x = input_147_cast_fp16, y = var_3446_cast_fp16)[name = string("input_157_cast_fp16")];
            tensor<int32, [1]> x_109_axes_0 = const()[name = string("x_109_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_5_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_5_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48387456)))];
            tensor<fp16, [1024]> enc_layers_5_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_5_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48389568)))];
            tensor<fp16, [1, 188, 1024]> x_109_cast_fp16 = layer_norm(axes = x_109_axes_0, beta = enc_layers_5_norm_feed_forward2_bias_to_fp16, epsilon = var_3326_to_fp16, gamma = enc_layers_5_norm_feed_forward2_weight_to_fp16, x = input_157_cast_fp16)[name = string("x_109_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_5_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48391680))), lut = tensor<fp16, [256, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49440320))))[name = string("blocks_5_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_65_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_5_ff2_1_weight_to_fp16_palettized, x = x_109_cast_fp16)[name = string("linear_65_cast_fp16")];
            tensor<fp16, [4096]> blocks_5_ff2_1_scale_to_fp16 = const()[name = string("blocks_5_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49442432)))];
            tensor<fp16, [1, 188, 4096]> input_159_cast_fp16 = mul(x = linear_65_cast_fp16, y = blocks_5_ff2_1_scale_to_fp16)[name = string("input_159_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_111_cast_fp16 = silu(x = input_159_cast_fp16)[name = string("x_111_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_5_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49450688))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50499328))))[name = string("blocks_5_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_66_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_5_ff2_2_weight_to_fp16_palettized, x = x_111_cast_fp16)[name = string("linear_66_cast_fp16")];
            tensor<fp16, [1024]> blocks_5_ff2_2_scale_to_fp16 = const()[name = string("blocks_5_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50499904)))];
            tensor<fp16, [1, 188, 1024]> var_3460_cast_fp16 = mul(x = linear_66_cast_fp16, y = blocks_5_ff2_2_scale_to_fp16)[name = string("op_3460_cast_fp16")];
            fp16 var_3461_to_fp16 = const()[name = string("op_3461_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3462_cast_fp16 = mul(x = var_3460_cast_fp16, y = var_3461_to_fp16)[name = string("op_3462_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_161_cast_fp16 = add(x = input_157_cast_fp16, y = var_3462_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]> enc_layers_5_norm_out_weight_to_fp16 = const()[name = string("enc_layers_5_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50502016)))];
            tensor<fp16, [1024]> enc_layers_5_norm_out_bias_to_fp16 = const()[name = string("enc_layers_5_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50504128)))];
            tensor<fp16, [1, 188, 1024]> input_163_cast_fp16 = layer_norm(axes = input_163_axes_0, beta = enc_layers_5_norm_out_bias_to_fp16, epsilon = var_3326_to_fp16, gamma = enc_layers_5_norm_out_weight_to_fp16, x = input_161_cast_fp16)[name = string("input_163_cast_fp16")];
            int32 var_3476 = const()[name = string("op_3476"), val = int32(-1)];
            tensor<int32, [1]> x_113_axes_0 = const()[name = string("x_113_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_6_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_6_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50506240)))];
            tensor<fp16, [1024]> enc_layers_6_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_6_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50508352)))];
            fp16 var_3483_to_fp16 = const()[name = string("op_3483_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_113_cast_fp16 = layer_norm(axes = x_113_axes_0, beta = enc_layers_6_norm_feed_forward1_bias_to_fp16, epsilon = var_3483_to_fp16, gamma = enc_layers_6_norm_feed_forward1_weight_to_fp16, x = input_163_cast_fp16)[name = string("x_113_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_6_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50510464))), lut = tensor<fp16, [256, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51559104))))[name = string("blocks_6_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_67_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_6_ff1_1_weight_to_fp16_palettized, x = x_113_cast_fp16)[name = string("linear_67_cast_fp16")];
            tensor<fp16, [4096]> blocks_6_ff1_1_scale_to_fp16 = const()[name = string("blocks_6_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51561216)))];
            tensor<fp16, [1, 188, 4096]> input_165_cast_fp16 = mul(x = linear_67_cast_fp16, y = blocks_6_ff1_1_scale_to_fp16)[name = string("input_165_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_115_cast_fp16 = silu(x = input_165_cast_fp16)[name = string("x_115_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_6_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51569472))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52618112))))[name = string("blocks_6_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_68_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_6_ff1_2_weight_to_fp16_palettized, x = x_115_cast_fp16)[name = string("linear_68_cast_fp16")];
            tensor<fp16, [1024]> blocks_6_ff1_2_scale_to_fp16 = const()[name = string("blocks_6_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52618688)))];
            tensor<fp16, [1, 188, 1024]> var_3508_cast_fp16 = mul(x = linear_68_cast_fp16, y = blocks_6_ff1_2_scale_to_fp16)[name = string("op_3508_cast_fp16")];
            fp16 var_3509_to_fp16 = const()[name = string("op_3509_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3510_cast_fp16 = mul(x = var_3508_cast_fp16, y = var_3509_to_fp16)[name = string("op_3510_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_167_cast_fp16 = add(x = input_163_cast_fp16, y = var_3510_cast_fp16)[name = string("input_167_cast_fp16")];
            tensor<int32, [1]> x_117_axes_0 = const()[name = string("x_117_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_6_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_6_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52620800)))];
            tensor<fp16, [1024]> enc_layers_6_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_6_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52622912)))];
            tensor<fp16, [1, 188, 1024]> x_117_cast_fp16 = layer_norm(axes = x_117_axes_0, beta = enc_layers_6_norm_self_att_bias_to_fp16, epsilon = var_3483_to_fp16, gamma = enc_layers_6_norm_self_att_weight_to_fp16, x = input_167_cast_fp16)[name = string("x_117_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_6_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52625024))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52887232))))[name = string("blocks_6_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_69_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_6_q_weight_to_fp16_palettized, x = x_117_cast_fp16)[name = string("linear_69_cast_fp16")];
            tensor<fp16, [1024]> blocks_6_q_scale_to_fp16 = const()[name = string("blocks_6_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52887808)))];
            tensor<fp16, [1, 188, 1024]> var_3518_cast_fp16 = mul(x = linear_69_cast_fp16, y = blocks_6_q_scale_to_fp16)[name = string("op_3518_cast_fp16")];
            tensor<int32, [4]> var_3519 = const()[name = string("op_3519"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3520_cast_fp16 = reshape(shape = var_3519, x = var_3518_cast_fp16)[name = string("op_3520_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_6_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52889920))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53152128))))[name = string("blocks_6_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_70_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_6_k_weight_to_fp16_palettized, x = x_117_cast_fp16)[name = string("linear_70_cast_fp16")];
            tensor<fp16, [1024]> blocks_6_k_scale_to_fp16 = const()[name = string("blocks_6_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53152704)))];
            tensor<fp16, [1, 188, 1024]> var_3526_cast_fp16 = mul(x = linear_70_cast_fp16, y = blocks_6_k_scale_to_fp16)[name = string("op_3526_cast_fp16")];
            tensor<int32, [4]> var_3527 = const()[name = string("op_3527"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3528_cast_fp16 = reshape(shape = var_3527, x = var_3526_cast_fp16)[name = string("op_3528_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_6_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53154816))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53417024))))[name = string("blocks_6_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_71_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_6_v_weight_to_fp16_palettized, x = x_117_cast_fp16)[name = string("linear_71_cast_fp16")];
            tensor<fp16, [1024]> blocks_6_v_scale_to_fp16 = const()[name = string("blocks_6_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53417600)))];
            tensor<fp16, [1, 188, 1024]> var_3534_cast_fp16 = mul(x = linear_71_cast_fp16, y = blocks_6_v_scale_to_fp16)[name = string("op_3534_cast_fp16")];
            tensor<int32, [4]> var_3535 = const()[name = string("op_3535"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3536_cast_fp16 = reshape(shape = var_3535, x = var_3534_cast_fp16)[name = string("op_3536_cast_fp16")];
            tensor<int32, [4]> v_13_perm_0 = const()[name = string("v_13_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_96_to_fp16 = const()[name = string("const_96_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53419712)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_13_cast_fp16 = add(x = var_3520_cast_fp16, y = const_96_to_fp16)[name = string("q_u_13_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_97_to_fp16 = const()[name = string("const_97_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53421824)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_13_cast_fp16 = add(x = var_3520_cast_fp16, y = const_97_to_fp16)[name = string("q_v_13_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_6_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53423936))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53686144))))[name = string("blocks_6_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_72_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_6_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_72_cast_fp16")];
            tensor<fp16, [1024]> blocks_6_pos_scale_to_fp16 = const()[name = string("blocks_6_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53686720)))];
            tensor<fp16, [1, 375, 1024]> var_3548_cast_fp16 = mul(x = linear_72_cast_fp16, y = blocks_6_pos_scale_to_fp16)[name = string("op_3548_cast_fp16")];
            tensor<int32, [4]> var_3549 = const()[name = string("op_3549"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_13_cast_fp16 = reshape(shape = var_3549, x = var_3548_cast_fp16)[name = string("p_13_cast_fp16")];
            bool input_169_transpose_x_0 = const()[name = string("input_169_transpose_x_0"), val = bool(false)];
            bool input_169_transpose_y_0 = const()[name = string("input_169_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_216_perm_0 = const()[name = string("transpose_216_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_217_perm_0 = const()[name = string("transpose_217_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_217 = transpose(perm = transpose_217_perm_0, x = p_13_cast_fp16)[name = string("transpose_431")];
            tensor<fp16, [1, 8, 188, 128]> transpose_216 = transpose(perm = transpose_216_perm_0, x = q_v_13_cast_fp16)[name = string("transpose_432")];
            tensor<fp16, [1, 8, 188, 375]> input_169_cast_fp16 = matmul(transpose_x = input_169_transpose_x_0, transpose_y = input_169_transpose_y_0, x = transpose_216, y = transpose_217)[name = string("input_169_cast_fp16")];
            tensor<int32, [8]> var_3555_pad_0 = const()[name = string("op_3555_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_3555_mode_0 = const()[name = string("op_3555_mode_0"), val = string("constant")];
            fp16 const_18_to_fp16 = const()[name = string("const_18_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_3555_cast_fp16 = pad(constant_val = const_18_to_fp16, mode = var_3555_mode_0, pad = var_3555_pad_0, x = input_169_cast_fp16)[name = string("op_3555_cast_fp16")];
            tensor<int32, [4]> var_3556 = const()[name = string("op_3556"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_25_cast_fp16 = reshape(shape = var_3556, x = var_3555_cast_fp16)[name = string("scores_25_cast_fp16")];
            tensor<int32, [4]> var_3560_begin_0 = const()[name = string("op_3560_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_3560_end_0 = const()[name = string("op_3560_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_3560_end_mask_0 = const()[name = string("op_3560_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_3560_cast_fp16 = slice_by_index(begin = var_3560_begin_0, end = var_3560_end_0, end_mask = var_3560_end_mask_0, x = scores_25_cast_fp16)[name = string("op_3560_cast_fp16")];
            tensor<int32, [4]> var_3561 = const()[name = string("op_3561"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_3562_cast_fp16 = reshape(shape = var_3561, x = var_3560_cast_fp16)[name = string("op_3562_cast_fp16")];
            tensor<int32, [4]> bd_13_begin_0 = const()[name = string("bd_13_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_13_end_0 = const()[name = string("bd_13_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_13_end_mask_0 = const()[name = string("bd_13_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_13_cast_fp16 = slice_by_index(begin = bd_13_begin_0, end = bd_13_end_0, end_mask = bd_13_end_mask_0, x = var_3562_cast_fp16)[name = string("bd_13_cast_fp16")];
            bool var_3565_transpose_x_0 = const()[name = string("op_3565_transpose_x_0"), val = bool(false)];
            bool var_3565_transpose_y_0 = const()[name = string("op_3565_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_218_perm_0 = const()[name = string("transpose_218_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_219_perm_0 = const()[name = string("transpose_219_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_219 = transpose(perm = transpose_219_perm_0, x = var_3528_cast_fp16)[name = string("transpose_429")];
            tensor<fp16, [1, 8, 188, 128]> transpose_218 = transpose(perm = transpose_218_perm_0, x = q_u_13_cast_fp16)[name = string("transpose_430")];
            tensor<fp16, [1, 8, 188, 188]> var_3565_cast_fp16 = matmul(transpose_x = var_3565_transpose_x_0, transpose_y = var_3565_transpose_y_0, x = transpose_218, y = transpose_219)[name = string("op_3565_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3566_cast_fp16 = add(x = var_3565_cast_fp16, y = bd_13_cast_fp16)[name = string("op_3566_cast_fp16")];
            fp16 var_3567_to_fp16 = const()[name = string("op_3567_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_3568_cast_fp16 = mul(x = var_3566_cast_fp16, y = var_3567_to_fp16)[name = string("op_3568_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_27_cast_fp16 = add(x = var_3568_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_27_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3570_cast_fp16 = softmax(axis = var_3476, x = scores_27_cast_fp16)[name = string("op_3570_cast_fp16")];
            bool out_13_transpose_x_0 = const()[name = string("out_13_transpose_x_0"), val = bool(false)];
            bool out_13_transpose_y_0 = const()[name = string("out_13_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_13_cast_fp16 = transpose(perm = v_13_perm_0, x = var_3536_cast_fp16)[name = string("transpose_428")];
            tensor<fp16, [1, 8, 188, 128]> out_13_cast_fp16 = matmul(transpose_x = out_13_transpose_x_0, transpose_y = out_13_transpose_y_0, x = var_3570_cast_fp16, y = v_13_cast_fp16)[name = string("out_13_cast_fp16")];
            tensor<int32, [4]> var_3572_perm_0 = const()[name = string("op_3572_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3573 = const()[name = string("op_3573"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_3572_cast_fp16 = transpose(perm = var_3572_perm_0, x = out_13_cast_fp16)[name = string("transpose_427")];
            tensor<fp16, [1, 188, 1024]> x_119_cast_fp16 = reshape(shape = var_3573, x = var_3572_cast_fp16)[name = string("x_119_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_6_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53688832))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53951040))))[name = string("blocks_6_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_73_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_6_o_weight_to_fp16_palettized, x = x_119_cast_fp16)[name = string("linear_73_cast_fp16")];
            tensor<fp16, [1024]> blocks_6_o_scale_to_fp16 = const()[name = string("blocks_6_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53951616)))];
            tensor<fp16, [1, 188, 1024]> var_3579_cast_fp16 = mul(x = linear_73_cast_fp16, y = blocks_6_o_scale_to_fp16)[name = string("op_3579_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_171_cast_fp16 = add(x = input_167_cast_fp16, y = var_3579_cast_fp16)[name = string("input_171_cast_fp16")];
            tensor<int32, [1]> x_121_axes_0 = const()[name = string("x_121_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_6_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_6_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53953728)))];
            tensor<fp16, [1024]> enc_layers_6_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_6_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53955840)))];
            tensor<fp16, [1, 188, 1024]> x_121_cast_fp16 = layer_norm(axes = x_121_axes_0, beta = enc_layers_6_norm_conv_bias_to_fp16, epsilon = var_3483_to_fp16, gamma = enc_layers_6_norm_conv_weight_to_fp16, x = input_171_cast_fp16)[name = string("x_121_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_6_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53957952))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55006592))))[name = string("blocks_6_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_74_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_6_pw1_weight_to_fp16_palettized, x = x_121_cast_fp16)[name = string("linear_74_cast_fp16")];
            tensor<fp16, [2048]> blocks_6_pw1_scale_to_fp16 = const()[name = string("blocks_6_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55010752)))];
            tensor<fp16, [1, 188, 2048]> input_173_cast_fp16 = mul(x = linear_74_cast_fp16, y = blocks_6_pw1_scale_to_fp16)[name = string("input_173_cast_fp16")];
            int32 var_3588_split_num_splits_0 = const()[name = string("op_3588_split_num_splits_0"), val = int32(2)];
            int32 var_3588_split_axis_0 = const()[name = string("op_3588_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_3588_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_3588_split_cast_fp16_1 = split(axis = var_3588_split_axis_0, num_splits = var_3588_split_num_splits_0, x = input_173_cast_fp16)[name = string("op_3588_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_3588_split_1_sigmoid_cast_fp16 = sigmoid(x = var_3588_split_cast_fp16_1)[name = string("op_3588_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_3588_cast_fp16 = mul(x = var_3588_split_cast_fp16_0, y = var_3588_split_1_sigmoid_cast_fp16)[name = string("op_3588_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_25_cast_fp16 = mul(x = var_3588_cast_fp16, y = valid_cast_fp16)[name = string("h_25_cast_fp16")];
            tensor<int32, [3]> input_175_perm_0 = const()[name = string("input_175_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_177_pad_type_0 = const()[name = string("input_177_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_177_pad_0 = const()[name = string("input_177_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_177_groups_0 = const()[name = string("input_177_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_177_strides_0 = const()[name = string("input_177_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_177_dilations_0 = const()[name = string("input_177_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_48_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55014912))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55024192))))[name = string("const_48_to_fp16_quantized")];
            tensor<fp16, [1024]> const_49_to_fp16 = const()[name = string("const_49_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55026304)))];
            tensor<fp16, [1, 1024, 188]> input_175_cast_fp16 = transpose(perm = input_175_perm_0, x = h_25_cast_fp16)[name = string("transpose_426")];
            tensor<fp16, [1, 1024, 188]> input_179_cast_fp16 = conv(bias = const_49_to_fp16, dilations = input_177_dilations_0, groups = input_177_groups_0, pad = input_177_pad_0, pad_type = input_177_pad_type_0, strides = input_177_strides_0, weight = const_48_to_fp16_quantized, x = input_175_cast_fp16)[name = string("input_179_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_27_cast_fp16 = silu(x = input_179_cast_fp16)[name = string("h_27_cast_fp16")];
            tensor<int32, [3]> x_123_perm_0 = const()[name = string("x_123_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_6_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55028416))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55552768))))[name = string("blocks_6_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_123_cast_fp16 = transpose(perm = x_123_perm_0, x = h_27_cast_fp16)[name = string("transpose_425")];
            tensor<fp16, [1, 188, 1024]> linear_75_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_6_pw2_weight_to_fp16_palettized, x = x_123_cast_fp16)[name = string("linear_75_cast_fp16")];
            tensor<fp16, [1024]> blocks_6_pw2_scale_to_fp16 = const()[name = string("blocks_6_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55554880)))];
            tensor<fp16, [1, 188, 1024]> var_3603_cast_fp16 = mul(x = linear_75_cast_fp16, y = blocks_6_pw2_scale_to_fp16)[name = string("op_3603_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_181_cast_fp16 = add(x = input_171_cast_fp16, y = var_3603_cast_fp16)[name = string("input_181_cast_fp16")];
            tensor<int32, [1]> x_125_axes_0 = const()[name = string("x_125_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_6_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_6_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55556992)))];
            tensor<fp16, [1024]> enc_layers_6_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_6_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55559104)))];
            tensor<fp16, [1, 188, 1024]> x_125_cast_fp16 = layer_norm(axes = x_125_axes_0, beta = enc_layers_6_norm_feed_forward2_bias_to_fp16, epsilon = var_3483_to_fp16, gamma = enc_layers_6_norm_feed_forward2_weight_to_fp16, x = input_181_cast_fp16)[name = string("x_125_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_6_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55561216))), lut = tensor<fp16, [256, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56609856))))[name = string("blocks_6_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_76_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_6_ff2_1_weight_to_fp16_palettized, x = x_125_cast_fp16)[name = string("linear_76_cast_fp16")];
            tensor<fp16, [4096]> blocks_6_ff2_1_scale_to_fp16 = const()[name = string("blocks_6_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56611968)))];
            tensor<fp16, [1, 188, 4096]> input_183_cast_fp16 = mul(x = linear_76_cast_fp16, y = blocks_6_ff2_1_scale_to_fp16)[name = string("input_183_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_127_cast_fp16 = silu(x = input_183_cast_fp16)[name = string("x_127_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_6_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56620224))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57668864))))[name = string("blocks_6_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_77_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_6_ff2_2_weight_to_fp16_palettized, x = x_127_cast_fp16)[name = string("linear_77_cast_fp16")];
            tensor<fp16, [1024]> blocks_6_ff2_2_scale_to_fp16 = const()[name = string("blocks_6_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57669440)))];
            tensor<fp16, [1, 188, 1024]> var_3617_cast_fp16 = mul(x = linear_77_cast_fp16, y = blocks_6_ff2_2_scale_to_fp16)[name = string("op_3617_cast_fp16")];
            fp16 var_3618_to_fp16 = const()[name = string("op_3618_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3619_cast_fp16 = mul(x = var_3617_cast_fp16, y = var_3618_to_fp16)[name = string("op_3619_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_185_cast_fp16 = add(x = input_181_cast_fp16, y = var_3619_cast_fp16)[name = string("input_185_cast_fp16")];
            tensor<int32, [1]> input_187_axes_0 = const()[name = string("input_187_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_6_norm_out_weight_to_fp16 = const()[name = string("enc_layers_6_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57671552)))];
            tensor<fp16, [1024]> enc_layers_6_norm_out_bias_to_fp16 = const()[name = string("enc_layers_6_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57673664)))];
            tensor<fp16, [1, 188, 1024]> input_187_cast_fp16 = layer_norm(axes = input_187_axes_0, beta = enc_layers_6_norm_out_bias_to_fp16, epsilon = var_3483_to_fp16, gamma = enc_layers_6_norm_out_weight_to_fp16, x = input_185_cast_fp16)[name = string("input_187_cast_fp16")];
            int32 var_3633 = const()[name = string("op_3633"), val = int32(-1)];
            tensor<int32, [1]> x_129_axes_0 = const()[name = string("x_129_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_7_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_7_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57675776)))];
            tensor<fp16, [1024]> enc_layers_7_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_7_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57677888)))];
            fp16 var_3640_to_fp16 = const()[name = string("op_3640_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_129_cast_fp16 = layer_norm(axes = x_129_axes_0, beta = enc_layers_7_norm_feed_forward1_bias_to_fp16, epsilon = var_3640_to_fp16, gamma = enc_layers_7_norm_feed_forward1_weight_to_fp16, x = input_187_cast_fp16)[name = string("x_129_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_7_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57680000))), lut = tensor<fp16, [256, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58728640))))[name = string("blocks_7_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_78_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_7_ff1_1_weight_to_fp16_palettized, x = x_129_cast_fp16)[name = string("linear_78_cast_fp16")];
            tensor<fp16, [4096]> blocks_7_ff1_1_scale_to_fp16 = const()[name = string("blocks_7_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58730752)))];
            tensor<fp16, [1, 188, 4096]> input_189_cast_fp16 = mul(x = linear_78_cast_fp16, y = blocks_7_ff1_1_scale_to_fp16)[name = string("input_189_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_131_cast_fp16 = silu(x = input_189_cast_fp16)[name = string("x_131_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_7_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58739008))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59787648))))[name = string("blocks_7_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_79_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_7_ff1_2_weight_to_fp16_palettized, x = x_131_cast_fp16)[name = string("linear_79_cast_fp16")];
            tensor<fp16, [1024]> blocks_7_ff1_2_scale_to_fp16 = const()[name = string("blocks_7_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59788224)))];
            tensor<fp16, [1, 188, 1024]> var_3665_cast_fp16 = mul(x = linear_79_cast_fp16, y = blocks_7_ff1_2_scale_to_fp16)[name = string("op_3665_cast_fp16")];
            fp16 var_3666_to_fp16 = const()[name = string("op_3666_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3667_cast_fp16 = mul(x = var_3665_cast_fp16, y = var_3666_to_fp16)[name = string("op_3667_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_191_cast_fp16 = add(x = input_187_cast_fp16, y = var_3667_cast_fp16)[name = string("input_191_cast_fp16")];
            tensor<int32, [1]> x_133_axes_0 = const()[name = string("x_133_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_7_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_7_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59790336)))];
            tensor<fp16, [1024]> enc_layers_7_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_7_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59792448)))];
            tensor<fp16, [1, 188, 1024]> x_133_cast_fp16 = layer_norm(axes = x_133_axes_0, beta = enc_layers_7_norm_self_att_bias_to_fp16, epsilon = var_3640_to_fp16, gamma = enc_layers_7_norm_self_att_weight_to_fp16, x = input_191_cast_fp16)[name = string("x_133_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_7_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59794560))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60056768))))[name = string("blocks_7_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_80_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_7_q_weight_to_fp16_palettized, x = x_133_cast_fp16)[name = string("linear_80_cast_fp16")];
            tensor<fp16, [1024]> blocks_7_q_scale_to_fp16 = const()[name = string("blocks_7_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60057344)))];
            tensor<fp16, [1, 188, 1024]> var_3675_cast_fp16 = mul(x = linear_80_cast_fp16, y = blocks_7_q_scale_to_fp16)[name = string("op_3675_cast_fp16")];
            tensor<int32, [4]> var_3676 = const()[name = string("op_3676"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3677_cast_fp16 = reshape(shape = var_3676, x = var_3675_cast_fp16)[name = string("op_3677_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_7_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60059456))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60321664))))[name = string("blocks_7_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_81_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_7_k_weight_to_fp16_palettized, x = x_133_cast_fp16)[name = string("linear_81_cast_fp16")];
            tensor<fp16, [1024]> blocks_7_k_scale_to_fp16 = const()[name = string("blocks_7_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60322240)))];
            tensor<fp16, [1, 188, 1024]> var_3683_cast_fp16 = mul(x = linear_81_cast_fp16, y = blocks_7_k_scale_to_fp16)[name = string("op_3683_cast_fp16")];
            tensor<int32, [4]> var_3684 = const()[name = string("op_3684"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3685_cast_fp16 = reshape(shape = var_3684, x = var_3683_cast_fp16)[name = string("op_3685_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_7_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60324352))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60586560))))[name = string("blocks_7_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_82_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_7_v_weight_to_fp16_palettized, x = x_133_cast_fp16)[name = string("linear_82_cast_fp16")];
            tensor<fp16, [1024]> blocks_7_v_scale_to_fp16 = const()[name = string("blocks_7_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60587136)))];
            tensor<fp16, [1, 188, 1024]> var_3691_cast_fp16 = mul(x = linear_82_cast_fp16, y = blocks_7_v_scale_to_fp16)[name = string("op_3691_cast_fp16")];
            tensor<int32, [4]> var_3692 = const()[name = string("op_3692"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3693_cast_fp16 = reshape(shape = var_3692, x = var_3691_cast_fp16)[name = string("op_3693_cast_fp16")];
            tensor<int32, [4]> v_15_perm_0 = const()[name = string("v_15_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_98_to_fp16 = const()[name = string("const_98_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60589248)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_15_cast_fp16 = add(x = var_3677_cast_fp16, y = const_98_to_fp16)[name = string("q_u_15_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_99_to_fp16 = const()[name = string("const_99_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60591360)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_15_cast_fp16 = add(x = var_3677_cast_fp16, y = const_99_to_fp16)[name = string("q_v_15_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_7_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60593472))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60855680))))[name = string("blocks_7_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_83_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_7_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_83_cast_fp16")];
            tensor<fp16, [1024]> blocks_7_pos_scale_to_fp16 = const()[name = string("blocks_7_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60856256)))];
            tensor<fp16, [1, 375, 1024]> var_3705_cast_fp16 = mul(x = linear_83_cast_fp16, y = blocks_7_pos_scale_to_fp16)[name = string("op_3705_cast_fp16")];
            tensor<int32, [4]> var_3706 = const()[name = string("op_3706"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_15_cast_fp16 = reshape(shape = var_3706, x = var_3705_cast_fp16)[name = string("p_15_cast_fp16")];
            bool input_193_transpose_x_0 = const()[name = string("input_193_transpose_x_0"), val = bool(false)];
            bool input_193_transpose_y_0 = const()[name = string("input_193_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_220_perm_0 = const()[name = string("transpose_220_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_221_perm_0 = const()[name = string("transpose_221_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_221 = transpose(perm = transpose_221_perm_0, x = p_15_cast_fp16)[name = string("transpose_423")];
            tensor<fp16, [1, 8, 188, 128]> transpose_220 = transpose(perm = transpose_220_perm_0, x = q_v_15_cast_fp16)[name = string("transpose_424")];
            tensor<fp16, [1, 8, 188, 375]> input_193_cast_fp16 = matmul(transpose_x = input_193_transpose_x_0, transpose_y = input_193_transpose_y_0, x = transpose_220, y = transpose_221)[name = string("input_193_cast_fp16")];
            tensor<int32, [8]> var_3712_pad_0 = const()[name = string("op_3712_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_3712_mode_0 = const()[name = string("op_3712_mode_0"), val = string("constant")];
            fp16 const_19_to_fp16 = const()[name = string("const_19_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_3712_cast_fp16 = pad(constant_val = const_19_to_fp16, mode = var_3712_mode_0, pad = var_3712_pad_0, x = input_193_cast_fp16)[name = string("op_3712_cast_fp16")];
            tensor<int32, [4]> var_3713 = const()[name = string("op_3713"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_29_cast_fp16 = reshape(shape = var_3713, x = var_3712_cast_fp16)[name = string("scores_29_cast_fp16")];
            tensor<int32, [4]> var_3717_begin_0 = const()[name = string("op_3717_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_3717_end_0 = const()[name = string("op_3717_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_3717_end_mask_0 = const()[name = string("op_3717_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_3717_cast_fp16 = slice_by_index(begin = var_3717_begin_0, end = var_3717_end_0, end_mask = var_3717_end_mask_0, x = scores_29_cast_fp16)[name = string("op_3717_cast_fp16")];
            tensor<int32, [4]> var_3718 = const()[name = string("op_3718"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_3719_cast_fp16 = reshape(shape = var_3718, x = var_3717_cast_fp16)[name = string("op_3719_cast_fp16")];
            tensor<int32, [4]> bd_15_begin_0 = const()[name = string("bd_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_15_end_0 = const()[name = string("bd_15_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_15_end_mask_0 = const()[name = string("bd_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_15_cast_fp16 = slice_by_index(begin = bd_15_begin_0, end = bd_15_end_0, end_mask = bd_15_end_mask_0, x = var_3719_cast_fp16)[name = string("bd_15_cast_fp16")];
            bool var_3722_transpose_x_0 = const()[name = string("op_3722_transpose_x_0"), val = bool(false)];
            bool var_3722_transpose_y_0 = const()[name = string("op_3722_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_222_perm_0 = const()[name = string("transpose_222_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_223_perm_0 = const()[name = string("transpose_223_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_223 = transpose(perm = transpose_223_perm_0, x = var_3685_cast_fp16)[name = string("transpose_421")];
            tensor<fp16, [1, 8, 188, 128]> transpose_222 = transpose(perm = transpose_222_perm_0, x = q_u_15_cast_fp16)[name = string("transpose_422")];
            tensor<fp16, [1, 8, 188, 188]> var_3722_cast_fp16 = matmul(transpose_x = var_3722_transpose_x_0, transpose_y = var_3722_transpose_y_0, x = transpose_222, y = transpose_223)[name = string("op_3722_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3723_cast_fp16 = add(x = var_3722_cast_fp16, y = bd_15_cast_fp16)[name = string("op_3723_cast_fp16")];
            fp16 var_3724_to_fp16 = const()[name = string("op_3724_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_3725_cast_fp16 = mul(x = var_3723_cast_fp16, y = var_3724_to_fp16)[name = string("op_3725_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_31_cast_fp16 = add(x = var_3725_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_31_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3727_cast_fp16 = softmax(axis = var_3633, x = scores_31_cast_fp16)[name = string("op_3727_cast_fp16")];
            bool out_15_transpose_x_0 = const()[name = string("out_15_transpose_x_0"), val = bool(false)];
            bool out_15_transpose_y_0 = const()[name = string("out_15_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_15_cast_fp16 = transpose(perm = v_15_perm_0, x = var_3693_cast_fp16)[name = string("transpose_420")];
            tensor<fp16, [1, 8, 188, 128]> out_15_cast_fp16 = matmul(transpose_x = out_15_transpose_x_0, transpose_y = out_15_transpose_y_0, x = var_3727_cast_fp16, y = v_15_cast_fp16)[name = string("out_15_cast_fp16")];
            tensor<int32, [4]> var_3729_perm_0 = const()[name = string("op_3729_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3730 = const()[name = string("op_3730"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_3729_cast_fp16 = transpose(perm = var_3729_perm_0, x = out_15_cast_fp16)[name = string("transpose_419")];
            tensor<fp16, [1, 188, 1024]> x_135_cast_fp16 = reshape(shape = var_3730, x = var_3729_cast_fp16)[name = string("x_135_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_7_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60858368))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61120576))))[name = string("blocks_7_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_84_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_7_o_weight_to_fp16_palettized, x = x_135_cast_fp16)[name = string("linear_84_cast_fp16")];
            tensor<fp16, [1024]> blocks_7_o_scale_to_fp16 = const()[name = string("blocks_7_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61121152)))];
            tensor<fp16, [1, 188, 1024]> var_3736_cast_fp16 = mul(x = linear_84_cast_fp16, y = blocks_7_o_scale_to_fp16)[name = string("op_3736_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_195_cast_fp16 = add(x = input_191_cast_fp16, y = var_3736_cast_fp16)[name = string("input_195_cast_fp16")];
            tensor<int32, [1]> x_137_axes_0 = const()[name = string("x_137_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_7_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_7_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61123264)))];
            tensor<fp16, [1024]> enc_layers_7_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_7_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61125376)))];
            tensor<fp16, [1, 188, 1024]> x_137_cast_fp16 = layer_norm(axes = x_137_axes_0, beta = enc_layers_7_norm_conv_bias_to_fp16, epsilon = var_3640_to_fp16, gamma = enc_layers_7_norm_conv_weight_to_fp16, x = input_195_cast_fp16)[name = string("x_137_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_7_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61127488))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62176128))))[name = string("blocks_7_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_85_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_7_pw1_weight_to_fp16_palettized, x = x_137_cast_fp16)[name = string("linear_85_cast_fp16")];
            tensor<fp16, [2048]> blocks_7_pw1_scale_to_fp16 = const()[name = string("blocks_7_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62180288)))];
            tensor<fp16, [1, 188, 2048]> input_197_cast_fp16 = mul(x = linear_85_cast_fp16, y = blocks_7_pw1_scale_to_fp16)[name = string("input_197_cast_fp16")];
            int32 var_3745_split_num_splits_0 = const()[name = string("op_3745_split_num_splits_0"), val = int32(2)];
            int32 var_3745_split_axis_0 = const()[name = string("op_3745_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_3745_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_3745_split_cast_fp16_1 = split(axis = var_3745_split_axis_0, num_splits = var_3745_split_num_splits_0, x = input_197_cast_fp16)[name = string("op_3745_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_3745_split_1_sigmoid_cast_fp16 = sigmoid(x = var_3745_split_cast_fp16_1)[name = string("op_3745_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_3745_cast_fp16 = mul(x = var_3745_split_cast_fp16_0, y = var_3745_split_1_sigmoid_cast_fp16)[name = string("op_3745_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_29_cast_fp16 = mul(x = var_3745_cast_fp16, y = valid_cast_fp16)[name = string("h_29_cast_fp16")];
            tensor<int32, [3]> input_199_perm_0 = const()[name = string("input_199_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_201_pad_type_0 = const()[name = string("input_201_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_201_pad_0 = const()[name = string("input_201_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_201_groups_0 = const()[name = string("input_201_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_201_strides_0 = const()[name = string("input_201_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_201_dilations_0 = const()[name = string("input_201_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_50_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62184448))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62193728))))[name = string("const_50_to_fp16_quantized")];
            tensor<fp16, [1024]> const_51_to_fp16 = const()[name = string("const_51_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62195840)))];
            tensor<fp16, [1, 1024, 188]> input_199_cast_fp16 = transpose(perm = input_199_perm_0, x = h_29_cast_fp16)[name = string("transpose_418")];
            tensor<fp16, [1, 1024, 188]> input_203_cast_fp16 = conv(bias = const_51_to_fp16, dilations = input_201_dilations_0, groups = input_201_groups_0, pad = input_201_pad_0, pad_type = input_201_pad_type_0, strides = input_201_strides_0, weight = const_50_to_fp16_quantized, x = input_199_cast_fp16)[name = string("input_203_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_31_cast_fp16 = silu(x = input_203_cast_fp16)[name = string("h_31_cast_fp16")];
            tensor<int32, [3]> x_139_perm_0 = const()[name = string("x_139_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_7_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62197952))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62722304))))[name = string("blocks_7_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_139_cast_fp16 = transpose(perm = x_139_perm_0, x = h_31_cast_fp16)[name = string("transpose_417")];
            tensor<fp16, [1, 188, 1024]> linear_86_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_7_pw2_weight_to_fp16_palettized, x = x_139_cast_fp16)[name = string("linear_86_cast_fp16")];
            tensor<fp16, [1024]> blocks_7_pw2_scale_to_fp16 = const()[name = string("blocks_7_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62724416)))];
            tensor<fp16, [1, 188, 1024]> var_3760_cast_fp16 = mul(x = linear_86_cast_fp16, y = blocks_7_pw2_scale_to_fp16)[name = string("op_3760_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_205_cast_fp16 = add(x = input_195_cast_fp16, y = var_3760_cast_fp16)[name = string("input_205_cast_fp16")];
            tensor<int32, [1]> x_141_axes_0 = const()[name = string("x_141_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_7_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_7_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62726528)))];
            tensor<fp16, [1024]> enc_layers_7_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_7_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62728640)))];
            tensor<fp16, [1, 188, 1024]> x_141_cast_fp16 = layer_norm(axes = x_141_axes_0, beta = enc_layers_7_norm_feed_forward2_bias_to_fp16, epsilon = var_3640_to_fp16, gamma = enc_layers_7_norm_feed_forward2_weight_to_fp16, x = input_205_cast_fp16)[name = string("x_141_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_7_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62730752))), lut = tensor<fp16, [256, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63779392))))[name = string("blocks_7_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_87_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_7_ff2_1_weight_to_fp16_palettized, x = x_141_cast_fp16)[name = string("linear_87_cast_fp16")];
            tensor<fp16, [4096]> blocks_7_ff2_1_scale_to_fp16 = const()[name = string("blocks_7_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63781504)))];
            tensor<fp16, [1, 188, 4096]> input_207_cast_fp16 = mul(x = linear_87_cast_fp16, y = blocks_7_ff2_1_scale_to_fp16)[name = string("input_207_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_143_cast_fp16 = silu(x = input_207_cast_fp16)[name = string("x_143_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_7_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63789760))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64838400))))[name = string("blocks_7_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_88_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_7_ff2_2_weight_to_fp16_palettized, x = x_143_cast_fp16)[name = string("linear_88_cast_fp16")];
            tensor<fp16, [1024]> blocks_7_ff2_2_scale_to_fp16 = const()[name = string("blocks_7_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64838976)))];
            tensor<fp16, [1, 188, 1024]> var_3774_cast_fp16 = mul(x = linear_88_cast_fp16, y = blocks_7_ff2_2_scale_to_fp16)[name = string("op_3774_cast_fp16")];
            fp16 var_3775_to_fp16 = const()[name = string("op_3775_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3776_cast_fp16 = mul(x = var_3774_cast_fp16, y = var_3775_to_fp16)[name = string("op_3776_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_209_cast_fp16 = add(x = input_205_cast_fp16, y = var_3776_cast_fp16)[name = string("input_209_cast_fp16")];
            tensor<int32, [1]> input_211_axes_0 = const()[name = string("input_211_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_7_norm_out_weight_to_fp16 = const()[name = string("enc_layers_7_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64841088)))];
            tensor<fp16, [1024]> enc_layers_7_norm_out_bias_to_fp16 = const()[name = string("enc_layers_7_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64843200)))];
            tensor<fp16, [1, 188, 1024]> input_211_cast_fp16 = layer_norm(axes = input_211_axes_0, beta = enc_layers_7_norm_out_bias_to_fp16, epsilon = var_3640_to_fp16, gamma = enc_layers_7_norm_out_weight_to_fp16, x = input_209_cast_fp16)[name = string("input_211_cast_fp16")];
            int32 var_3790 = const()[name = string("op_3790"), val = int32(-1)];
            tensor<int32, [1]> x_145_axes_0 = const()[name = string("x_145_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_8_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_8_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64845312)))];
            tensor<fp16, [1024]> enc_layers_8_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_8_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64847424)))];
            fp16 var_3797_to_fp16 = const()[name = string("op_3797_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_145_cast_fp16 = layer_norm(axes = x_145_axes_0, beta = enc_layers_8_norm_feed_forward1_bias_to_fp16, epsilon = var_3797_to_fp16, gamma = enc_layers_8_norm_feed_forward1_weight_to_fp16, x = input_211_cast_fp16)[name = string("x_145_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_8_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64849536))), lut = tensor<fp16, [256, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66422464))))[name = string("blocks_8_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_89_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_8_ff1_1_weight_to_fp16_palettized, x = x_145_cast_fp16)[name = string("linear_89_cast_fp16")];
            tensor<fp16, [4096]> blocks_8_ff1_1_scale_to_fp16 = const()[name = string("blocks_8_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66426624)))];
            tensor<fp16, [1, 188, 4096]> input_213_cast_fp16 = mul(x = linear_89_cast_fp16, y = blocks_8_ff1_1_scale_to_fp16)[name = string("input_213_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_147_cast_fp16 = silu(x = input_213_cast_fp16)[name = string("x_147_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_8_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66434880))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68007808))))[name = string("blocks_8_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_90_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_8_ff1_2_weight_to_fp16_palettized, x = x_147_cast_fp16)[name = string("linear_90_cast_fp16")];
            tensor<fp16, [1024]> blocks_8_ff1_2_scale_to_fp16 = const()[name = string("blocks_8_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68008896)))];
            tensor<fp16, [1, 188, 1024]> var_3822_cast_fp16 = mul(x = linear_90_cast_fp16, y = blocks_8_ff1_2_scale_to_fp16)[name = string("op_3822_cast_fp16")];
            fp16 var_3823_to_fp16 = const()[name = string("op_3823_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3824_cast_fp16 = mul(x = var_3822_cast_fp16, y = var_3823_to_fp16)[name = string("op_3824_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_215_cast_fp16 = add(x = input_211_cast_fp16, y = var_3824_cast_fp16)[name = string("input_215_cast_fp16")];
            tensor<int32, [1]> x_149_axes_0 = const()[name = string("x_149_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_8_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_8_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68011008)))];
            tensor<fp16, [1024]> enc_layers_8_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_8_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68013120)))];
            tensor<fp16, [1, 188, 1024]> x_149_cast_fp16 = layer_norm(axes = x_149_axes_0, beta = enc_layers_8_norm_self_att_bias_to_fp16, epsilon = var_3797_to_fp16, gamma = enc_layers_8_norm_self_att_weight_to_fp16, x = input_215_cast_fp16)[name = string("x_149_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_8_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68015232))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68277440))))[name = string("blocks_8_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_91_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_8_q_weight_to_fp16_palettized, x = x_149_cast_fp16)[name = string("linear_91_cast_fp16")];
            tensor<fp16, [1024]> blocks_8_q_scale_to_fp16 = const()[name = string("blocks_8_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68278016)))];
            tensor<fp16, [1, 188, 1024]> var_3832_cast_fp16 = mul(x = linear_91_cast_fp16, y = blocks_8_q_scale_to_fp16)[name = string("op_3832_cast_fp16")];
            tensor<int32, [4]> var_3833 = const()[name = string("op_3833"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3834_cast_fp16 = reshape(shape = var_3833, x = var_3832_cast_fp16)[name = string("op_3834_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_8_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68280128))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68542336))))[name = string("blocks_8_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_92_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_8_k_weight_to_fp16_palettized, x = x_149_cast_fp16)[name = string("linear_92_cast_fp16")];
            tensor<fp16, [1024]> blocks_8_k_scale_to_fp16 = const()[name = string("blocks_8_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68542912)))];
            tensor<fp16, [1, 188, 1024]> var_3840_cast_fp16 = mul(x = linear_92_cast_fp16, y = blocks_8_k_scale_to_fp16)[name = string("op_3840_cast_fp16")];
            tensor<int32, [4]> var_3841 = const()[name = string("op_3841"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3842_cast_fp16 = reshape(shape = var_3841, x = var_3840_cast_fp16)[name = string("op_3842_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_8_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68545024))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68807232))))[name = string("blocks_8_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_93_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_8_v_weight_to_fp16_palettized, x = x_149_cast_fp16)[name = string("linear_93_cast_fp16")];
            tensor<fp16, [1024]> blocks_8_v_scale_to_fp16 = const()[name = string("blocks_8_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68807808)))];
            tensor<fp16, [1, 188, 1024]> var_3848_cast_fp16 = mul(x = linear_93_cast_fp16, y = blocks_8_v_scale_to_fp16)[name = string("op_3848_cast_fp16")];
            tensor<int32, [4]> var_3849 = const()[name = string("op_3849"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3850_cast_fp16 = reshape(shape = var_3849, x = var_3848_cast_fp16)[name = string("op_3850_cast_fp16")];
            tensor<int32, [4]> v_17_perm_0 = const()[name = string("v_17_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_100_to_fp16 = const()[name = string("const_100_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68809920)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_17_cast_fp16 = add(x = var_3834_cast_fp16, y = const_100_to_fp16)[name = string("q_u_17_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_101_to_fp16 = const()[name = string("const_101_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68812032)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_17_cast_fp16 = add(x = var_3834_cast_fp16, y = const_101_to_fp16)[name = string("q_v_17_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_8_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68814144))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69076352))))[name = string("blocks_8_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_94_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_8_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_94_cast_fp16")];
            tensor<fp16, [1024]> blocks_8_pos_scale_to_fp16 = const()[name = string("blocks_8_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69076928)))];
            tensor<fp16, [1, 375, 1024]> var_3862_cast_fp16 = mul(x = linear_94_cast_fp16, y = blocks_8_pos_scale_to_fp16)[name = string("op_3862_cast_fp16")];
            tensor<int32, [4]> var_3863 = const()[name = string("op_3863"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_17_cast_fp16 = reshape(shape = var_3863, x = var_3862_cast_fp16)[name = string("p_17_cast_fp16")];
            bool input_217_transpose_x_0 = const()[name = string("input_217_transpose_x_0"), val = bool(false)];
            bool input_217_transpose_y_0 = const()[name = string("input_217_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_224_perm_0 = const()[name = string("transpose_224_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_225_perm_0 = const()[name = string("transpose_225_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_225 = transpose(perm = transpose_225_perm_0, x = p_17_cast_fp16)[name = string("transpose_415")];
            tensor<fp16, [1, 8, 188, 128]> transpose_224 = transpose(perm = transpose_224_perm_0, x = q_v_17_cast_fp16)[name = string("transpose_416")];
            tensor<fp16, [1, 8, 188, 375]> input_217_cast_fp16 = matmul(transpose_x = input_217_transpose_x_0, transpose_y = input_217_transpose_y_0, x = transpose_224, y = transpose_225)[name = string("input_217_cast_fp16")];
            tensor<int32, [8]> var_3869_pad_0 = const()[name = string("op_3869_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_3869_mode_0 = const()[name = string("op_3869_mode_0"), val = string("constant")];
            fp16 const_20_to_fp16 = const()[name = string("const_20_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_3869_cast_fp16 = pad(constant_val = const_20_to_fp16, mode = var_3869_mode_0, pad = var_3869_pad_0, x = input_217_cast_fp16)[name = string("op_3869_cast_fp16")];
            tensor<int32, [4]> var_3870 = const()[name = string("op_3870"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_33_cast_fp16 = reshape(shape = var_3870, x = var_3869_cast_fp16)[name = string("scores_33_cast_fp16")];
            tensor<int32, [4]> var_3874_begin_0 = const()[name = string("op_3874_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_3874_end_0 = const()[name = string("op_3874_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_3874_end_mask_0 = const()[name = string("op_3874_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_3874_cast_fp16 = slice_by_index(begin = var_3874_begin_0, end = var_3874_end_0, end_mask = var_3874_end_mask_0, x = scores_33_cast_fp16)[name = string("op_3874_cast_fp16")];
            tensor<int32, [4]> var_3875 = const()[name = string("op_3875"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_3876_cast_fp16 = reshape(shape = var_3875, x = var_3874_cast_fp16)[name = string("op_3876_cast_fp16")];
            tensor<int32, [4]> bd_17_begin_0 = const()[name = string("bd_17_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_17_end_0 = const()[name = string("bd_17_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_17_end_mask_0 = const()[name = string("bd_17_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_17_cast_fp16 = slice_by_index(begin = bd_17_begin_0, end = bd_17_end_0, end_mask = bd_17_end_mask_0, x = var_3876_cast_fp16)[name = string("bd_17_cast_fp16")];
            bool var_3879_transpose_x_0 = const()[name = string("op_3879_transpose_x_0"), val = bool(false)];
            bool var_3879_transpose_y_0 = const()[name = string("op_3879_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_226_perm_0 = const()[name = string("transpose_226_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_227_perm_0 = const()[name = string("transpose_227_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_227 = transpose(perm = transpose_227_perm_0, x = var_3842_cast_fp16)[name = string("transpose_413")];
            tensor<fp16, [1, 8, 188, 128]> transpose_226 = transpose(perm = transpose_226_perm_0, x = q_u_17_cast_fp16)[name = string("transpose_414")];
            tensor<fp16, [1, 8, 188, 188]> var_3879_cast_fp16 = matmul(transpose_x = var_3879_transpose_x_0, transpose_y = var_3879_transpose_y_0, x = transpose_226, y = transpose_227)[name = string("op_3879_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3880_cast_fp16 = add(x = var_3879_cast_fp16, y = bd_17_cast_fp16)[name = string("op_3880_cast_fp16")];
            fp16 var_3881_to_fp16 = const()[name = string("op_3881_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_3882_cast_fp16 = mul(x = var_3880_cast_fp16, y = var_3881_to_fp16)[name = string("op_3882_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_35_cast_fp16 = add(x = var_3882_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_35_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3884_cast_fp16 = softmax(axis = var_3790, x = scores_35_cast_fp16)[name = string("op_3884_cast_fp16")];
            bool out_17_transpose_x_0 = const()[name = string("out_17_transpose_x_0"), val = bool(false)];
            bool out_17_transpose_y_0 = const()[name = string("out_17_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_17_cast_fp16 = transpose(perm = v_17_perm_0, x = var_3850_cast_fp16)[name = string("transpose_412")];
            tensor<fp16, [1, 8, 188, 128]> out_17_cast_fp16 = matmul(transpose_x = out_17_transpose_x_0, transpose_y = out_17_transpose_y_0, x = var_3884_cast_fp16, y = v_17_cast_fp16)[name = string("out_17_cast_fp16")];
            tensor<int32, [4]> var_3886_perm_0 = const()[name = string("op_3886_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3887 = const()[name = string("op_3887"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_3886_cast_fp16 = transpose(perm = var_3886_perm_0, x = out_17_cast_fp16)[name = string("transpose_411")];
            tensor<fp16, [1, 188, 1024]> x_151_cast_fp16 = reshape(shape = var_3887, x = var_3886_cast_fp16)[name = string("x_151_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_8_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69079040))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69341248))))[name = string("blocks_8_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_95_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_8_o_weight_to_fp16_palettized, x = x_151_cast_fp16)[name = string("linear_95_cast_fp16")];
            tensor<fp16, [1024]> blocks_8_o_scale_to_fp16 = const()[name = string("blocks_8_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69341824)))];
            tensor<fp16, [1, 188, 1024]> var_3893_cast_fp16 = mul(x = linear_95_cast_fp16, y = blocks_8_o_scale_to_fp16)[name = string("op_3893_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_219_cast_fp16 = add(x = input_215_cast_fp16, y = var_3893_cast_fp16)[name = string("input_219_cast_fp16")];
            tensor<int32, [1]> x_153_axes_0 = const()[name = string("x_153_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_8_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_8_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69343936)))];
            tensor<fp16, [1024]> enc_layers_8_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_8_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69346048)))];
            tensor<fp16, [1, 188, 1024]> x_153_cast_fp16 = layer_norm(axes = x_153_axes_0, beta = enc_layers_8_norm_conv_bias_to_fp16, epsilon = var_3797_to_fp16, gamma = enc_layers_8_norm_conv_weight_to_fp16, x = input_219_cast_fp16)[name = string("x_153_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_8_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69348160))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70396800))))[name = string("blocks_8_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_96_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_8_pw1_weight_to_fp16_palettized, x = x_153_cast_fp16)[name = string("linear_96_cast_fp16")];
            tensor<fp16, [2048]> blocks_8_pw1_scale_to_fp16 = const()[name = string("blocks_8_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70400960)))];
            tensor<fp16, [1, 188, 2048]> input_221_cast_fp16 = mul(x = linear_96_cast_fp16, y = blocks_8_pw1_scale_to_fp16)[name = string("input_221_cast_fp16")];
            int32 var_3902_split_num_splits_0 = const()[name = string("op_3902_split_num_splits_0"), val = int32(2)];
            int32 var_3902_split_axis_0 = const()[name = string("op_3902_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_3902_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_3902_split_cast_fp16_1 = split(axis = var_3902_split_axis_0, num_splits = var_3902_split_num_splits_0, x = input_221_cast_fp16)[name = string("op_3902_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_3902_split_1_sigmoid_cast_fp16 = sigmoid(x = var_3902_split_cast_fp16_1)[name = string("op_3902_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_3902_cast_fp16 = mul(x = var_3902_split_cast_fp16_0, y = var_3902_split_1_sigmoid_cast_fp16)[name = string("op_3902_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_33_cast_fp16 = mul(x = var_3902_cast_fp16, y = valid_cast_fp16)[name = string("h_33_cast_fp16")];
            tensor<int32, [3]> input_223_perm_0 = const()[name = string("input_223_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_225_pad_type_0 = const()[name = string("input_225_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_225_pad_0 = const()[name = string("input_225_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_225_groups_0 = const()[name = string("input_225_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_225_strides_0 = const()[name = string("input_225_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_225_dilations_0 = const()[name = string("input_225_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_52_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70405120))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70414400))))[name = string("const_52_to_fp16_quantized")];
            tensor<fp16, [1024]> const_53_to_fp16 = const()[name = string("const_53_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70416512)))];
            tensor<fp16, [1, 1024, 188]> input_223_cast_fp16 = transpose(perm = input_223_perm_0, x = h_33_cast_fp16)[name = string("transpose_410")];
            tensor<fp16, [1, 1024, 188]> input_227_cast_fp16 = conv(bias = const_53_to_fp16, dilations = input_225_dilations_0, groups = input_225_groups_0, pad = input_225_pad_0, pad_type = input_225_pad_type_0, strides = input_225_strides_0, weight = const_52_to_fp16_quantized, x = input_223_cast_fp16)[name = string("input_227_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_35_cast_fp16 = silu(x = input_227_cast_fp16)[name = string("h_35_cast_fp16")];
            tensor<int32, [3]> x_155_perm_0 = const()[name = string("x_155_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_8_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70418624))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70942976))))[name = string("blocks_8_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_155_cast_fp16 = transpose(perm = x_155_perm_0, x = h_35_cast_fp16)[name = string("transpose_409")];
            tensor<fp16, [1, 188, 1024]> linear_97_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_8_pw2_weight_to_fp16_palettized, x = x_155_cast_fp16)[name = string("linear_97_cast_fp16")];
            tensor<fp16, [1024]> blocks_8_pw2_scale_to_fp16 = const()[name = string("blocks_8_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70945088)))];
            tensor<fp16, [1, 188, 1024]> var_3917_cast_fp16 = mul(x = linear_97_cast_fp16, y = blocks_8_pw2_scale_to_fp16)[name = string("op_3917_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_229_cast_fp16 = add(x = input_219_cast_fp16, y = var_3917_cast_fp16)[name = string("input_229_cast_fp16")];
            tensor<int32, [1]> x_157_axes_0 = const()[name = string("x_157_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_8_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_8_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70947200)))];
            tensor<fp16, [1024]> enc_layers_8_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_8_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70949312)))];
            tensor<fp16, [1, 188, 1024]> x_157_cast_fp16 = layer_norm(axes = x_157_axes_0, beta = enc_layers_8_norm_feed_forward2_bias_to_fp16, epsilon = var_3797_to_fp16, gamma = enc_layers_8_norm_feed_forward2_weight_to_fp16, x = input_229_cast_fp16)[name = string("x_157_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_8_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70951424))), lut = tensor<fp16, [256, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72524352))))[name = string("blocks_8_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_98_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_8_ff2_1_weight_to_fp16_palettized, x = x_157_cast_fp16)[name = string("linear_98_cast_fp16")];
            tensor<fp16, [4096]> blocks_8_ff2_1_scale_to_fp16 = const()[name = string("blocks_8_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72528512)))];
            tensor<fp16, [1, 188, 4096]> input_231_cast_fp16 = mul(x = linear_98_cast_fp16, y = blocks_8_ff2_1_scale_to_fp16)[name = string("input_231_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_159_cast_fp16 = silu(x = input_231_cast_fp16)[name = string("x_159_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_8_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72536768))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74109696))))[name = string("blocks_8_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_99_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_8_ff2_2_weight_to_fp16_palettized, x = x_159_cast_fp16)[name = string("linear_99_cast_fp16")];
            tensor<fp16, [1024]> blocks_8_ff2_2_scale_to_fp16 = const()[name = string("blocks_8_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74110784)))];
            tensor<fp16, [1, 188, 1024]> var_3931_cast_fp16 = mul(x = linear_99_cast_fp16, y = blocks_8_ff2_2_scale_to_fp16)[name = string("op_3931_cast_fp16")];
            fp16 var_3932_to_fp16 = const()[name = string("op_3932_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3933_cast_fp16 = mul(x = var_3931_cast_fp16, y = var_3932_to_fp16)[name = string("op_3933_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_233_cast_fp16 = add(x = input_229_cast_fp16, y = var_3933_cast_fp16)[name = string("input_233_cast_fp16")];
            tensor<int32, [1]> input_235_axes_0 = const()[name = string("input_235_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_8_norm_out_weight_to_fp16 = const()[name = string("enc_layers_8_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74112896)))];
            tensor<fp16, [1024]> enc_layers_8_norm_out_bias_to_fp16 = const()[name = string("enc_layers_8_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74115008)))];
            tensor<fp16, [1, 188, 1024]> input_235_cast_fp16 = layer_norm(axes = input_235_axes_0, beta = enc_layers_8_norm_out_bias_to_fp16, epsilon = var_3797_to_fp16, gamma = enc_layers_8_norm_out_weight_to_fp16, x = input_233_cast_fp16)[name = string("input_235_cast_fp16")];
            int32 var_3947 = const()[name = string("op_3947"), val = int32(-1)];
            tensor<int32, [1]> x_161_axes_0 = const()[name = string("x_161_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_9_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_9_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74117120)))];
            tensor<fp16, [1024]> enc_layers_9_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_9_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74119232)))];
            fp16 var_3954_to_fp16 = const()[name = string("op_3954_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_161_cast_fp16 = layer_norm(axes = x_161_axes_0, beta = enc_layers_9_norm_feed_forward1_bias_to_fp16, epsilon = var_3954_to_fp16, gamma = enc_layers_9_norm_feed_forward1_weight_to_fp16, x = input_235_cast_fp16)[name = string("x_161_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_9_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74121344))), lut = tensor<fp16, [256, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75169984))))[name = string("blocks_9_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_100_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_9_ff1_1_weight_to_fp16_palettized, x = x_161_cast_fp16)[name = string("linear_100_cast_fp16")];
            tensor<fp16, [4096]> blocks_9_ff1_1_scale_to_fp16 = const()[name = string("blocks_9_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75172096)))];
            tensor<fp16, [1, 188, 4096]> input_237_cast_fp16 = mul(x = linear_100_cast_fp16, y = blocks_9_ff1_1_scale_to_fp16)[name = string("input_237_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_163_cast_fp16 = silu(x = input_237_cast_fp16)[name = string("x_163_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_9_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75180352))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76228992))))[name = string("blocks_9_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_101_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_9_ff1_2_weight_to_fp16_palettized, x = x_163_cast_fp16)[name = string("linear_101_cast_fp16")];
            tensor<fp16, [1024]> blocks_9_ff1_2_scale_to_fp16 = const()[name = string("blocks_9_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76229568)))];
            tensor<fp16, [1, 188, 1024]> var_3979_cast_fp16 = mul(x = linear_101_cast_fp16, y = blocks_9_ff1_2_scale_to_fp16)[name = string("op_3979_cast_fp16")];
            fp16 var_3980_to_fp16 = const()[name = string("op_3980_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3981_cast_fp16 = mul(x = var_3979_cast_fp16, y = var_3980_to_fp16)[name = string("op_3981_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_239_cast_fp16 = add(x = input_235_cast_fp16, y = var_3981_cast_fp16)[name = string("input_239_cast_fp16")];
            tensor<int32, [1]> x_165_axes_0 = const()[name = string("x_165_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_9_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_9_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76231680)))];
            tensor<fp16, [1024]> enc_layers_9_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_9_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76233792)))];
            tensor<fp16, [1, 188, 1024]> x_165_cast_fp16 = layer_norm(axes = x_165_axes_0, beta = enc_layers_9_norm_self_att_bias_to_fp16, epsilon = var_3954_to_fp16, gamma = enc_layers_9_norm_self_att_weight_to_fp16, x = input_239_cast_fp16)[name = string("x_165_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_9_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76235904))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76498112))))[name = string("blocks_9_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_102_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_9_q_weight_to_fp16_palettized, x = x_165_cast_fp16)[name = string("linear_102_cast_fp16")];
            tensor<fp16, [1024]> blocks_9_q_scale_to_fp16 = const()[name = string("blocks_9_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76498688)))];
            tensor<fp16, [1, 188, 1024]> var_3989_cast_fp16 = mul(x = linear_102_cast_fp16, y = blocks_9_q_scale_to_fp16)[name = string("op_3989_cast_fp16")];
            tensor<int32, [4]> var_3990 = const()[name = string("op_3990"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3991_cast_fp16 = reshape(shape = var_3990, x = var_3989_cast_fp16)[name = string("op_3991_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_9_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76500800))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76763008))))[name = string("blocks_9_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_103_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_9_k_weight_to_fp16_palettized, x = x_165_cast_fp16)[name = string("linear_103_cast_fp16")];
            tensor<fp16, [1024]> blocks_9_k_scale_to_fp16 = const()[name = string("blocks_9_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76763584)))];
            tensor<fp16, [1, 188, 1024]> var_3997_cast_fp16 = mul(x = linear_103_cast_fp16, y = blocks_9_k_scale_to_fp16)[name = string("op_3997_cast_fp16")];
            tensor<int32, [4]> var_3998 = const()[name = string("op_3998"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3999_cast_fp16 = reshape(shape = var_3998, x = var_3997_cast_fp16)[name = string("op_3999_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_9_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76765696))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77027904))))[name = string("blocks_9_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_104_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_9_v_weight_to_fp16_palettized, x = x_165_cast_fp16)[name = string("linear_104_cast_fp16")];
            tensor<fp16, [1024]> blocks_9_v_scale_to_fp16 = const()[name = string("blocks_9_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77028480)))];
            tensor<fp16, [1, 188, 1024]> var_4005_cast_fp16 = mul(x = linear_104_cast_fp16, y = blocks_9_v_scale_to_fp16)[name = string("op_4005_cast_fp16")];
            tensor<int32, [4]> var_4006 = const()[name = string("op_4006"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4007_cast_fp16 = reshape(shape = var_4006, x = var_4005_cast_fp16)[name = string("op_4007_cast_fp16")];
            tensor<int32, [4]> v_19_perm_0 = const()[name = string("v_19_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_102_to_fp16 = const()[name = string("const_102_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77030592)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_19_cast_fp16 = add(x = var_3991_cast_fp16, y = const_102_to_fp16)[name = string("q_u_19_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_103_to_fp16 = const()[name = string("const_103_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77032704)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_19_cast_fp16 = add(x = var_3991_cast_fp16, y = const_103_to_fp16)[name = string("q_v_19_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_9_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77034816))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77297024))))[name = string("blocks_9_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_105_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_9_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_105_cast_fp16")];
            tensor<fp16, [1024]> blocks_9_pos_scale_to_fp16 = const()[name = string("blocks_9_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77297600)))];
            tensor<fp16, [1, 375, 1024]> var_4019_cast_fp16 = mul(x = linear_105_cast_fp16, y = blocks_9_pos_scale_to_fp16)[name = string("op_4019_cast_fp16")];
            tensor<int32, [4]> var_4020 = const()[name = string("op_4020"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_19_cast_fp16 = reshape(shape = var_4020, x = var_4019_cast_fp16)[name = string("p_19_cast_fp16")];
            bool input_241_transpose_x_0 = const()[name = string("input_241_transpose_x_0"), val = bool(false)];
            bool input_241_transpose_y_0 = const()[name = string("input_241_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_228_perm_0 = const()[name = string("transpose_228_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_229_perm_0 = const()[name = string("transpose_229_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_229 = transpose(perm = transpose_229_perm_0, x = p_19_cast_fp16)[name = string("transpose_407")];
            tensor<fp16, [1, 8, 188, 128]> transpose_228 = transpose(perm = transpose_228_perm_0, x = q_v_19_cast_fp16)[name = string("transpose_408")];
            tensor<fp16, [1, 8, 188, 375]> input_241_cast_fp16 = matmul(transpose_x = input_241_transpose_x_0, transpose_y = input_241_transpose_y_0, x = transpose_228, y = transpose_229)[name = string("input_241_cast_fp16")];
            tensor<int32, [8]> var_4026_pad_0 = const()[name = string("op_4026_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_4026_mode_0 = const()[name = string("op_4026_mode_0"), val = string("constant")];
            fp16 const_21_to_fp16 = const()[name = string("const_21_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_4026_cast_fp16 = pad(constant_val = const_21_to_fp16, mode = var_4026_mode_0, pad = var_4026_pad_0, x = input_241_cast_fp16)[name = string("op_4026_cast_fp16")];
            tensor<int32, [4]> var_4027 = const()[name = string("op_4027"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_37_cast_fp16 = reshape(shape = var_4027, x = var_4026_cast_fp16)[name = string("scores_37_cast_fp16")];
            tensor<int32, [4]> var_4031_begin_0 = const()[name = string("op_4031_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_4031_end_0 = const()[name = string("op_4031_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_4031_end_mask_0 = const()[name = string("op_4031_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_4031_cast_fp16 = slice_by_index(begin = var_4031_begin_0, end = var_4031_end_0, end_mask = var_4031_end_mask_0, x = scores_37_cast_fp16)[name = string("op_4031_cast_fp16")];
            tensor<int32, [4]> var_4032 = const()[name = string("op_4032"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_4033_cast_fp16 = reshape(shape = var_4032, x = var_4031_cast_fp16)[name = string("op_4033_cast_fp16")];
            tensor<int32, [4]> bd_19_begin_0 = const()[name = string("bd_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_19_end_0 = const()[name = string("bd_19_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_19_end_mask_0 = const()[name = string("bd_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_19_cast_fp16 = slice_by_index(begin = bd_19_begin_0, end = bd_19_end_0, end_mask = bd_19_end_mask_0, x = var_4033_cast_fp16)[name = string("bd_19_cast_fp16")];
            bool var_4036_transpose_x_0 = const()[name = string("op_4036_transpose_x_0"), val = bool(false)];
            bool var_4036_transpose_y_0 = const()[name = string("op_4036_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_230_perm_0 = const()[name = string("transpose_230_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_231_perm_0 = const()[name = string("transpose_231_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_231 = transpose(perm = transpose_231_perm_0, x = var_3999_cast_fp16)[name = string("transpose_405")];
            tensor<fp16, [1, 8, 188, 128]> transpose_230 = transpose(perm = transpose_230_perm_0, x = q_u_19_cast_fp16)[name = string("transpose_406")];
            tensor<fp16, [1, 8, 188, 188]> var_4036_cast_fp16 = matmul(transpose_x = var_4036_transpose_x_0, transpose_y = var_4036_transpose_y_0, x = transpose_230, y = transpose_231)[name = string("op_4036_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4037_cast_fp16 = add(x = var_4036_cast_fp16, y = bd_19_cast_fp16)[name = string("op_4037_cast_fp16")];
            fp16 var_4038_to_fp16 = const()[name = string("op_4038_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_4039_cast_fp16 = mul(x = var_4037_cast_fp16, y = var_4038_to_fp16)[name = string("op_4039_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_39_cast_fp16 = add(x = var_4039_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_39_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4041_cast_fp16 = softmax(axis = var_3947, x = scores_39_cast_fp16)[name = string("op_4041_cast_fp16")];
            bool out_19_transpose_x_0 = const()[name = string("out_19_transpose_x_0"), val = bool(false)];
            bool out_19_transpose_y_0 = const()[name = string("out_19_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_19_cast_fp16 = transpose(perm = v_19_perm_0, x = var_4007_cast_fp16)[name = string("transpose_404")];
            tensor<fp16, [1, 8, 188, 128]> out_19_cast_fp16 = matmul(transpose_x = out_19_transpose_x_0, transpose_y = out_19_transpose_y_0, x = var_4041_cast_fp16, y = v_19_cast_fp16)[name = string("out_19_cast_fp16")];
            tensor<int32, [4]> var_4043_perm_0 = const()[name = string("op_4043_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4044 = const()[name = string("op_4044"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_4043_cast_fp16 = transpose(perm = var_4043_perm_0, x = out_19_cast_fp16)[name = string("transpose_403")];
            tensor<fp16, [1, 188, 1024]> x_167_cast_fp16 = reshape(shape = var_4044, x = var_4043_cast_fp16)[name = string("x_167_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_9_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77299712))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77561920))))[name = string("blocks_9_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_106_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_9_o_weight_to_fp16_palettized, x = x_167_cast_fp16)[name = string("linear_106_cast_fp16")];
            tensor<fp16, [1024]> blocks_9_o_scale_to_fp16 = const()[name = string("blocks_9_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77562496)))];
            tensor<fp16, [1, 188, 1024]> var_4050_cast_fp16 = mul(x = linear_106_cast_fp16, y = blocks_9_o_scale_to_fp16)[name = string("op_4050_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_243_cast_fp16 = add(x = input_239_cast_fp16, y = var_4050_cast_fp16)[name = string("input_243_cast_fp16")];
            tensor<int32, [1]> x_169_axes_0 = const()[name = string("x_169_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_9_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_9_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77564608)))];
            tensor<fp16, [1024]> enc_layers_9_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_9_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77566720)))];
            tensor<fp16, [1, 188, 1024]> x_169_cast_fp16 = layer_norm(axes = x_169_axes_0, beta = enc_layers_9_norm_conv_bias_to_fp16, epsilon = var_3954_to_fp16, gamma = enc_layers_9_norm_conv_weight_to_fp16, x = input_243_cast_fp16)[name = string("x_169_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_9_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77568832))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78617472))))[name = string("blocks_9_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_107_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_9_pw1_weight_to_fp16_palettized, x = x_169_cast_fp16)[name = string("linear_107_cast_fp16")];
            tensor<fp16, [2048]> blocks_9_pw1_scale_to_fp16 = const()[name = string("blocks_9_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78621632)))];
            tensor<fp16, [1, 188, 2048]> input_245_cast_fp16 = mul(x = linear_107_cast_fp16, y = blocks_9_pw1_scale_to_fp16)[name = string("input_245_cast_fp16")];
            int32 var_4059_split_num_splits_0 = const()[name = string("op_4059_split_num_splits_0"), val = int32(2)];
            int32 var_4059_split_axis_0 = const()[name = string("op_4059_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_4059_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_4059_split_cast_fp16_1 = split(axis = var_4059_split_axis_0, num_splits = var_4059_split_num_splits_0, x = input_245_cast_fp16)[name = string("op_4059_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_4059_split_1_sigmoid_cast_fp16 = sigmoid(x = var_4059_split_cast_fp16_1)[name = string("op_4059_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_4059_cast_fp16 = mul(x = var_4059_split_cast_fp16_0, y = var_4059_split_1_sigmoid_cast_fp16)[name = string("op_4059_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_37_cast_fp16 = mul(x = var_4059_cast_fp16, y = valid_cast_fp16)[name = string("h_37_cast_fp16")];
            tensor<int32, [3]> input_247_perm_0 = const()[name = string("input_247_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_249_pad_type_0 = const()[name = string("input_249_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_249_pad_0 = const()[name = string("input_249_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_249_groups_0 = const()[name = string("input_249_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_249_strides_0 = const()[name = string("input_249_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_249_dilations_0 = const()[name = string("input_249_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_54_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78625792))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78635072))))[name = string("const_54_to_fp16_quantized")];
            tensor<fp16, [1024]> const_55_to_fp16 = const()[name = string("const_55_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78637184)))];
            tensor<fp16, [1, 1024, 188]> input_247_cast_fp16 = transpose(perm = input_247_perm_0, x = h_37_cast_fp16)[name = string("transpose_402")];
            tensor<fp16, [1, 1024, 188]> input_251_cast_fp16 = conv(bias = const_55_to_fp16, dilations = input_249_dilations_0, groups = input_249_groups_0, pad = input_249_pad_0, pad_type = input_249_pad_type_0, strides = input_249_strides_0, weight = const_54_to_fp16_quantized, x = input_247_cast_fp16)[name = string("input_251_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_39_cast_fp16 = silu(x = input_251_cast_fp16)[name = string("h_39_cast_fp16")];
            tensor<int32, [3]> x_171_perm_0 = const()[name = string("x_171_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_9_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78639296))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79163648))))[name = string("blocks_9_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_171_cast_fp16 = transpose(perm = x_171_perm_0, x = h_39_cast_fp16)[name = string("transpose_401")];
            tensor<fp16, [1, 188, 1024]> linear_108_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_9_pw2_weight_to_fp16_palettized, x = x_171_cast_fp16)[name = string("linear_108_cast_fp16")];
            tensor<fp16, [1024]> blocks_9_pw2_scale_to_fp16 = const()[name = string("blocks_9_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79165760)))];
            tensor<fp16, [1, 188, 1024]> var_4074_cast_fp16 = mul(x = linear_108_cast_fp16, y = blocks_9_pw2_scale_to_fp16)[name = string("op_4074_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_253_cast_fp16 = add(x = input_243_cast_fp16, y = var_4074_cast_fp16)[name = string("input_253_cast_fp16")];
            tensor<int32, [1]> x_173_axes_0 = const()[name = string("x_173_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_9_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_9_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79167872)))];
            tensor<fp16, [1024]> enc_layers_9_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_9_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79169984)))];
            tensor<fp16, [1, 188, 1024]> x_173_cast_fp16 = layer_norm(axes = x_173_axes_0, beta = enc_layers_9_norm_feed_forward2_bias_to_fp16, epsilon = var_3954_to_fp16, gamma = enc_layers_9_norm_feed_forward2_weight_to_fp16, x = input_253_cast_fp16)[name = string("x_173_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_9_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79172096))), lut = tensor<fp16, [256, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80220736))))[name = string("blocks_9_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_109_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_9_ff2_1_weight_to_fp16_palettized, x = x_173_cast_fp16)[name = string("linear_109_cast_fp16")];
            tensor<fp16, [4096]> blocks_9_ff2_1_scale_to_fp16 = const()[name = string("blocks_9_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80222848)))];
            tensor<fp16, [1, 188, 4096]> input_255_cast_fp16 = mul(x = linear_109_cast_fp16, y = blocks_9_ff2_1_scale_to_fp16)[name = string("input_255_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_175_cast_fp16 = silu(x = input_255_cast_fp16)[name = string("x_175_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_9_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80231104))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81279744))))[name = string("blocks_9_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_110_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_9_ff2_2_weight_to_fp16_palettized, x = x_175_cast_fp16)[name = string("linear_110_cast_fp16")];
            tensor<fp16, [1024]> blocks_9_ff2_2_scale_to_fp16 = const()[name = string("blocks_9_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81280320)))];
            tensor<fp16, [1, 188, 1024]> var_4088_cast_fp16 = mul(x = linear_110_cast_fp16, y = blocks_9_ff2_2_scale_to_fp16)[name = string("op_4088_cast_fp16")];
            fp16 var_4089_to_fp16 = const()[name = string("op_4089_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4090_cast_fp16 = mul(x = var_4088_cast_fp16, y = var_4089_to_fp16)[name = string("op_4090_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_257_cast_fp16 = add(x = input_253_cast_fp16, y = var_4090_cast_fp16)[name = string("input_257_cast_fp16")];
            tensor<int32, [1]> input_259_axes_0 = const()[name = string("input_259_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_9_norm_out_weight_to_fp16 = const()[name = string("enc_layers_9_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81282432)))];
            tensor<fp16, [1024]> enc_layers_9_norm_out_bias_to_fp16 = const()[name = string("enc_layers_9_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81284544)))];
            tensor<fp16, [1, 188, 1024]> input_259_cast_fp16 = layer_norm(axes = input_259_axes_0, beta = enc_layers_9_norm_out_bias_to_fp16, epsilon = var_3954_to_fp16, gamma = enc_layers_9_norm_out_weight_to_fp16, x = input_257_cast_fp16)[name = string("input_259_cast_fp16")];
            int32 var_4104 = const()[name = string("op_4104"), val = int32(-1)];
            tensor<int32, [1]> x_177_axes_0 = const()[name = string("x_177_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_10_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_10_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81286656)))];
            tensor<fp16, [1024]> enc_layers_10_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_10_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81288768)))];
            fp16 var_4111_to_fp16 = const()[name = string("op_4111_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_177_cast_fp16 = layer_norm(axes = x_177_axes_0, beta = enc_layers_10_norm_feed_forward1_bias_to_fp16, epsilon = var_4111_to_fp16, gamma = enc_layers_10_norm_feed_forward1_weight_to_fp16, x = input_259_cast_fp16)[name = string("x_177_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_10_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81290880))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83388096))))[name = string("blocks_10_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_111_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_10_ff1_1_weight_to_fp16_palettized, x = x_177_cast_fp16)[name = string("linear_111_cast_fp16")];
            tensor<fp16, [4096]> blocks_10_ff1_1_scale_to_fp16 = const()[name = string("blocks_10_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83396352)))];
            tensor<fp16, [1, 188, 4096]> input_261_cast_fp16 = mul(x = linear_111_cast_fp16, y = blocks_10_ff1_1_scale_to_fp16)[name = string("input_261_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_179_cast_fp16 = silu(x = input_261_cast_fp16)[name = string("x_179_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_10_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83404608))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85501824))))[name = string("blocks_10_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_112_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_10_ff1_2_weight_to_fp16_palettized, x = x_179_cast_fp16)[name = string("linear_112_cast_fp16")];
            tensor<fp16, [1024]> blocks_10_ff1_2_scale_to_fp16 = const()[name = string("blocks_10_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85503936)))];
            tensor<fp16, [1, 188, 1024]> var_4136_cast_fp16 = mul(x = linear_112_cast_fp16, y = blocks_10_ff1_2_scale_to_fp16)[name = string("op_4136_cast_fp16")];
            fp16 var_4137_to_fp16 = const()[name = string("op_4137_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4138_cast_fp16 = mul(x = var_4136_cast_fp16, y = var_4137_to_fp16)[name = string("op_4138_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_263_cast_fp16 = add(x = input_259_cast_fp16, y = var_4138_cast_fp16)[name = string("input_263_cast_fp16")];
            tensor<int32, [1]> x_181_axes_0 = const()[name = string("x_181_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_10_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_10_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85506048)))];
            tensor<fp16, [1024]> enc_layers_10_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_10_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85508160)))];
            tensor<fp16, [1, 188, 1024]> x_181_cast_fp16 = layer_norm(axes = x_181_axes_0, beta = enc_layers_10_norm_self_att_bias_to_fp16, epsilon = var_4111_to_fp16, gamma = enc_layers_10_norm_self_att_weight_to_fp16, x = input_263_cast_fp16)[name = string("x_181_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_10_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85510272))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85772480))))[name = string("blocks_10_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_113_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_10_q_weight_to_fp16_palettized, x = x_181_cast_fp16)[name = string("linear_113_cast_fp16")];
            tensor<fp16, [1024]> blocks_10_q_scale_to_fp16 = const()[name = string("blocks_10_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85773056)))];
            tensor<fp16, [1, 188, 1024]> var_4146_cast_fp16 = mul(x = linear_113_cast_fp16, y = blocks_10_q_scale_to_fp16)[name = string("op_4146_cast_fp16")];
            tensor<int32, [4]> var_4147 = const()[name = string("op_4147"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4148_cast_fp16 = reshape(shape = var_4147, x = var_4146_cast_fp16)[name = string("op_4148_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_10_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85775168))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86037376))))[name = string("blocks_10_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_114_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_10_k_weight_to_fp16_palettized, x = x_181_cast_fp16)[name = string("linear_114_cast_fp16")];
            tensor<fp16, [1024]> blocks_10_k_scale_to_fp16 = const()[name = string("blocks_10_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86037952)))];
            tensor<fp16, [1, 188, 1024]> var_4154_cast_fp16 = mul(x = linear_114_cast_fp16, y = blocks_10_k_scale_to_fp16)[name = string("op_4154_cast_fp16")];
            tensor<int32, [4]> var_4155 = const()[name = string("op_4155"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4156_cast_fp16 = reshape(shape = var_4155, x = var_4154_cast_fp16)[name = string("op_4156_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_10_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86040064))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86302272))))[name = string("blocks_10_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_115_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_10_v_weight_to_fp16_palettized, x = x_181_cast_fp16)[name = string("linear_115_cast_fp16")];
            tensor<fp16, [1024]> blocks_10_v_scale_to_fp16 = const()[name = string("blocks_10_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86302848)))];
            tensor<fp16, [1, 188, 1024]> var_4162_cast_fp16 = mul(x = linear_115_cast_fp16, y = blocks_10_v_scale_to_fp16)[name = string("op_4162_cast_fp16")];
            tensor<int32, [4]> var_4163 = const()[name = string("op_4163"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4164_cast_fp16 = reshape(shape = var_4163, x = var_4162_cast_fp16)[name = string("op_4164_cast_fp16")];
            tensor<int32, [4]> v_21_perm_0 = const()[name = string("v_21_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_104_to_fp16 = const()[name = string("const_104_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86304960)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_21_cast_fp16 = add(x = var_4148_cast_fp16, y = const_104_to_fp16)[name = string("q_u_21_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_105_to_fp16 = const()[name = string("const_105_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86307072)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_21_cast_fp16 = add(x = var_4148_cast_fp16, y = const_105_to_fp16)[name = string("q_v_21_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_10_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86309184))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86571392))))[name = string("blocks_10_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_116_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_10_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_116_cast_fp16")];
            tensor<fp16, [1024]> blocks_10_pos_scale_to_fp16 = const()[name = string("blocks_10_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86571968)))];
            tensor<fp16, [1, 375, 1024]> var_4176_cast_fp16 = mul(x = linear_116_cast_fp16, y = blocks_10_pos_scale_to_fp16)[name = string("op_4176_cast_fp16")];
            tensor<int32, [4]> var_4177 = const()[name = string("op_4177"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_21_cast_fp16 = reshape(shape = var_4177, x = var_4176_cast_fp16)[name = string("p_21_cast_fp16")];
            bool input_265_transpose_x_0 = const()[name = string("input_265_transpose_x_0"), val = bool(false)];
            bool input_265_transpose_y_0 = const()[name = string("input_265_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_232_perm_0 = const()[name = string("transpose_232_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_233_perm_0 = const()[name = string("transpose_233_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_233 = transpose(perm = transpose_233_perm_0, x = p_21_cast_fp16)[name = string("transpose_399")];
            tensor<fp16, [1, 8, 188, 128]> transpose_232 = transpose(perm = transpose_232_perm_0, x = q_v_21_cast_fp16)[name = string("transpose_400")];
            tensor<fp16, [1, 8, 188, 375]> input_265_cast_fp16 = matmul(transpose_x = input_265_transpose_x_0, transpose_y = input_265_transpose_y_0, x = transpose_232, y = transpose_233)[name = string("input_265_cast_fp16")];
            tensor<int32, [8]> var_4183_pad_0 = const()[name = string("op_4183_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_4183_mode_0 = const()[name = string("op_4183_mode_0"), val = string("constant")];
            fp16 const_22_to_fp16 = const()[name = string("const_22_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_4183_cast_fp16 = pad(constant_val = const_22_to_fp16, mode = var_4183_mode_0, pad = var_4183_pad_0, x = input_265_cast_fp16)[name = string("op_4183_cast_fp16")];
            tensor<int32, [4]> var_4184 = const()[name = string("op_4184"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_41_cast_fp16 = reshape(shape = var_4184, x = var_4183_cast_fp16)[name = string("scores_41_cast_fp16")];
            tensor<int32, [4]> var_4188_begin_0 = const()[name = string("op_4188_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_4188_end_0 = const()[name = string("op_4188_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_4188_end_mask_0 = const()[name = string("op_4188_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_4188_cast_fp16 = slice_by_index(begin = var_4188_begin_0, end = var_4188_end_0, end_mask = var_4188_end_mask_0, x = scores_41_cast_fp16)[name = string("op_4188_cast_fp16")];
            tensor<int32, [4]> var_4189 = const()[name = string("op_4189"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_4190_cast_fp16 = reshape(shape = var_4189, x = var_4188_cast_fp16)[name = string("op_4190_cast_fp16")];
            tensor<int32, [4]> bd_21_begin_0 = const()[name = string("bd_21_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_21_end_0 = const()[name = string("bd_21_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_21_end_mask_0 = const()[name = string("bd_21_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_21_cast_fp16 = slice_by_index(begin = bd_21_begin_0, end = bd_21_end_0, end_mask = bd_21_end_mask_0, x = var_4190_cast_fp16)[name = string("bd_21_cast_fp16")];
            bool var_4193_transpose_x_0 = const()[name = string("op_4193_transpose_x_0"), val = bool(false)];
            bool var_4193_transpose_y_0 = const()[name = string("op_4193_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_234_perm_0 = const()[name = string("transpose_234_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_235_perm_0 = const()[name = string("transpose_235_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_235 = transpose(perm = transpose_235_perm_0, x = var_4156_cast_fp16)[name = string("transpose_397")];
            tensor<fp16, [1, 8, 188, 128]> transpose_234 = transpose(perm = transpose_234_perm_0, x = q_u_21_cast_fp16)[name = string("transpose_398")];
            tensor<fp16, [1, 8, 188, 188]> var_4193_cast_fp16 = matmul(transpose_x = var_4193_transpose_x_0, transpose_y = var_4193_transpose_y_0, x = transpose_234, y = transpose_235)[name = string("op_4193_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4194_cast_fp16 = add(x = var_4193_cast_fp16, y = bd_21_cast_fp16)[name = string("op_4194_cast_fp16")];
            fp16 var_4195_to_fp16 = const()[name = string("op_4195_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_4196_cast_fp16 = mul(x = var_4194_cast_fp16, y = var_4195_to_fp16)[name = string("op_4196_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_43_cast_fp16 = add(x = var_4196_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_43_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4198_cast_fp16 = softmax(axis = var_4104, x = scores_43_cast_fp16)[name = string("op_4198_cast_fp16")];
            bool out_21_transpose_x_0 = const()[name = string("out_21_transpose_x_0"), val = bool(false)];
            bool out_21_transpose_y_0 = const()[name = string("out_21_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_21_cast_fp16 = transpose(perm = v_21_perm_0, x = var_4164_cast_fp16)[name = string("transpose_396")];
            tensor<fp16, [1, 8, 188, 128]> out_21_cast_fp16 = matmul(transpose_x = out_21_transpose_x_0, transpose_y = out_21_transpose_y_0, x = var_4198_cast_fp16, y = v_21_cast_fp16)[name = string("out_21_cast_fp16")];
            tensor<int32, [4]> var_4200_perm_0 = const()[name = string("op_4200_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4201 = const()[name = string("op_4201"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_4200_cast_fp16 = transpose(perm = var_4200_perm_0, x = out_21_cast_fp16)[name = string("transpose_395")];
            tensor<fp16, [1, 188, 1024]> x_183_cast_fp16 = reshape(shape = var_4201, x = var_4200_cast_fp16)[name = string("x_183_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_10_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86574080))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86836288))))[name = string("blocks_10_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_117_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_10_o_weight_to_fp16_palettized, x = x_183_cast_fp16)[name = string("linear_117_cast_fp16")];
            tensor<fp16, [1024]> blocks_10_o_scale_to_fp16 = const()[name = string("blocks_10_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86836864)))];
            tensor<fp16, [1, 188, 1024]> var_4207_cast_fp16 = mul(x = linear_117_cast_fp16, y = blocks_10_o_scale_to_fp16)[name = string("op_4207_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_267_cast_fp16 = add(x = input_263_cast_fp16, y = var_4207_cast_fp16)[name = string("input_267_cast_fp16")];
            tensor<int32, [1]> x_185_axes_0 = const()[name = string("x_185_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_10_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_10_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86838976)))];
            tensor<fp16, [1024]> enc_layers_10_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_10_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86841088)))];
            tensor<fp16, [1, 188, 1024]> x_185_cast_fp16 = layer_norm(axes = x_185_axes_0, beta = enc_layers_10_norm_conv_bias_to_fp16, epsilon = var_4111_to_fp16, gamma = enc_layers_10_norm_conv_weight_to_fp16, x = input_267_cast_fp16)[name = string("x_185_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_10_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86843200))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87891840))))[name = string("blocks_10_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_118_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_10_pw1_weight_to_fp16_palettized, x = x_185_cast_fp16)[name = string("linear_118_cast_fp16")];
            tensor<fp16, [2048]> blocks_10_pw1_scale_to_fp16 = const()[name = string("blocks_10_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87896000)))];
            tensor<fp16, [1, 188, 2048]> input_269_cast_fp16 = mul(x = linear_118_cast_fp16, y = blocks_10_pw1_scale_to_fp16)[name = string("input_269_cast_fp16")];
            int32 var_4216_split_num_splits_0 = const()[name = string("op_4216_split_num_splits_0"), val = int32(2)];
            int32 var_4216_split_axis_0 = const()[name = string("op_4216_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_4216_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_4216_split_cast_fp16_1 = split(axis = var_4216_split_axis_0, num_splits = var_4216_split_num_splits_0, x = input_269_cast_fp16)[name = string("op_4216_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_4216_split_1_sigmoid_cast_fp16 = sigmoid(x = var_4216_split_cast_fp16_1)[name = string("op_4216_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_4216_cast_fp16 = mul(x = var_4216_split_cast_fp16_0, y = var_4216_split_1_sigmoid_cast_fp16)[name = string("op_4216_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_41_cast_fp16 = mul(x = var_4216_cast_fp16, y = valid_cast_fp16)[name = string("h_41_cast_fp16")];
            tensor<int32, [3]> input_271_perm_0 = const()[name = string("input_271_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_273_pad_type_0 = const()[name = string("input_273_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_273_pad_0 = const()[name = string("input_273_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_273_groups_0 = const()[name = string("input_273_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_273_strides_0 = const()[name = string("input_273_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_273_dilations_0 = const()[name = string("input_273_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_56_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87900160))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87909440))))[name = string("const_56_to_fp16_quantized")];
            tensor<fp16, [1024]> const_57_to_fp16 = const()[name = string("const_57_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87911552)))];
            tensor<fp16, [1, 1024, 188]> input_271_cast_fp16 = transpose(perm = input_271_perm_0, x = h_41_cast_fp16)[name = string("transpose_394")];
            tensor<fp16, [1, 1024, 188]> input_275_cast_fp16 = conv(bias = const_57_to_fp16, dilations = input_273_dilations_0, groups = input_273_groups_0, pad = input_273_pad_0, pad_type = input_273_pad_type_0, strides = input_273_strides_0, weight = const_56_to_fp16_quantized, x = input_271_cast_fp16)[name = string("input_275_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_43_cast_fp16 = silu(x = input_275_cast_fp16)[name = string("h_43_cast_fp16")];
            tensor<int32, [3]> x_187_perm_0 = const()[name = string("x_187_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_10_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87913664))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88438016))))[name = string("blocks_10_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_187_cast_fp16 = transpose(perm = x_187_perm_0, x = h_43_cast_fp16)[name = string("transpose_393")];
            tensor<fp16, [1, 188, 1024]> linear_119_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_10_pw2_weight_to_fp16_palettized, x = x_187_cast_fp16)[name = string("linear_119_cast_fp16")];
            tensor<fp16, [1024]> blocks_10_pw2_scale_to_fp16 = const()[name = string("blocks_10_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88440128)))];
            tensor<fp16, [1, 188, 1024]> var_4231_cast_fp16 = mul(x = linear_119_cast_fp16, y = blocks_10_pw2_scale_to_fp16)[name = string("op_4231_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_277_cast_fp16 = add(x = input_267_cast_fp16, y = var_4231_cast_fp16)[name = string("input_277_cast_fp16")];
            tensor<int32, [1]> x_189_axes_0 = const()[name = string("x_189_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_10_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_10_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88442240)))];
            tensor<fp16, [1024]> enc_layers_10_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_10_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88444352)))];
            tensor<fp16, [1, 188, 1024]> x_189_cast_fp16 = layer_norm(axes = x_189_axes_0, beta = enc_layers_10_norm_feed_forward2_bias_to_fp16, epsilon = var_4111_to_fp16, gamma = enc_layers_10_norm_feed_forward2_weight_to_fp16, x = input_277_cast_fp16)[name = string("x_189_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_10_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88446464))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90543680))))[name = string("blocks_10_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_120_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_10_ff2_1_weight_to_fp16_palettized, x = x_189_cast_fp16)[name = string("linear_120_cast_fp16")];
            tensor<fp16, [4096]> blocks_10_ff2_1_scale_to_fp16 = const()[name = string("blocks_10_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90551936)))];
            tensor<fp16, [1, 188, 4096]> input_279_cast_fp16 = mul(x = linear_120_cast_fp16, y = blocks_10_ff2_1_scale_to_fp16)[name = string("input_279_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_191_cast_fp16 = silu(x = input_279_cast_fp16)[name = string("x_191_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_10_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90560192))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92657408))))[name = string("blocks_10_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_121_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_10_ff2_2_weight_to_fp16_palettized, x = x_191_cast_fp16)[name = string("linear_121_cast_fp16")];
            tensor<fp16, [1024]> blocks_10_ff2_2_scale_to_fp16 = const()[name = string("blocks_10_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92659520)))];
            tensor<fp16, [1, 188, 1024]> var_4245_cast_fp16 = mul(x = linear_121_cast_fp16, y = blocks_10_ff2_2_scale_to_fp16)[name = string("op_4245_cast_fp16")];
            fp16 var_4246_to_fp16 = const()[name = string("op_4246_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4247_cast_fp16 = mul(x = var_4245_cast_fp16, y = var_4246_to_fp16)[name = string("op_4247_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_281_cast_fp16 = add(x = input_277_cast_fp16, y = var_4247_cast_fp16)[name = string("input_281_cast_fp16")];
            tensor<int32, [1]> input_283_axes_0 = const()[name = string("input_283_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_10_norm_out_weight_to_fp16 = const()[name = string("enc_layers_10_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92661632)))];
            tensor<fp16, [1024]> enc_layers_10_norm_out_bias_to_fp16 = const()[name = string("enc_layers_10_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92663744)))];
            tensor<fp16, [1, 188, 1024]> input_283_cast_fp16 = layer_norm(axes = input_283_axes_0, beta = enc_layers_10_norm_out_bias_to_fp16, epsilon = var_4111_to_fp16, gamma = enc_layers_10_norm_out_weight_to_fp16, x = input_281_cast_fp16)[name = string("input_283_cast_fp16")];
            int32 var_4261 = const()[name = string("op_4261"), val = int32(-1)];
            tensor<int32, [1]> x_193_axes_0 = const()[name = string("x_193_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_11_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_11_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92665856)))];
            tensor<fp16, [1024]> enc_layers_11_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_11_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92667968)))];
            fp16 var_4268_to_fp16 = const()[name = string("op_4268_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_193_cast_fp16 = layer_norm(axes = x_193_axes_0, beta = enc_layers_11_norm_feed_forward1_bias_to_fp16, epsilon = var_4268_to_fp16, gamma = enc_layers_11_norm_feed_forward1_weight_to_fp16, x = input_283_cast_fp16)[name = string("x_193_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_11_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92670080))), lut = tensor<fp16, [256, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93718720))))[name = string("blocks_11_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_122_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_11_ff1_1_weight_to_fp16_palettized, x = x_193_cast_fp16)[name = string("linear_122_cast_fp16")];
            tensor<fp16, [4096]> blocks_11_ff1_1_scale_to_fp16 = const()[name = string("blocks_11_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93720832)))];
            tensor<fp16, [1, 188, 4096]> input_285_cast_fp16 = mul(x = linear_122_cast_fp16, y = blocks_11_ff1_1_scale_to_fp16)[name = string("input_285_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_195_cast_fp16 = silu(x = input_285_cast_fp16)[name = string("x_195_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_11_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93729088))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94777728))))[name = string("blocks_11_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_123_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_11_ff1_2_weight_to_fp16_palettized, x = x_195_cast_fp16)[name = string("linear_123_cast_fp16")];
            tensor<fp16, [1024]> blocks_11_ff1_2_scale_to_fp16 = const()[name = string("blocks_11_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94778304)))];
            tensor<fp16, [1, 188, 1024]> var_4293_cast_fp16 = mul(x = linear_123_cast_fp16, y = blocks_11_ff1_2_scale_to_fp16)[name = string("op_4293_cast_fp16")];
            fp16 var_4294_to_fp16 = const()[name = string("op_4294_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4295_cast_fp16 = mul(x = var_4293_cast_fp16, y = var_4294_to_fp16)[name = string("op_4295_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_287_cast_fp16 = add(x = input_283_cast_fp16, y = var_4295_cast_fp16)[name = string("input_287_cast_fp16")];
            tensor<int32, [1]> x_197_axes_0 = const()[name = string("x_197_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_11_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_11_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94780416)))];
            tensor<fp16, [1024]> enc_layers_11_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_11_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94782528)))];
            tensor<fp16, [1, 188, 1024]> x_197_cast_fp16 = layer_norm(axes = x_197_axes_0, beta = enc_layers_11_norm_self_att_bias_to_fp16, epsilon = var_4268_to_fp16, gamma = enc_layers_11_norm_self_att_weight_to_fp16, x = input_287_cast_fp16)[name = string("x_197_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_11_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94784640))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95046848))))[name = string("blocks_11_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_124_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_11_q_weight_to_fp16_palettized, x = x_197_cast_fp16)[name = string("linear_124_cast_fp16")];
            tensor<fp16, [1024]> blocks_11_q_scale_to_fp16 = const()[name = string("blocks_11_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95047424)))];
            tensor<fp16, [1, 188, 1024]> var_4303_cast_fp16 = mul(x = linear_124_cast_fp16, y = blocks_11_q_scale_to_fp16)[name = string("op_4303_cast_fp16")];
            tensor<int32, [4]> var_4304 = const()[name = string("op_4304"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4305_cast_fp16 = reshape(shape = var_4304, x = var_4303_cast_fp16)[name = string("op_4305_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_11_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95049536))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95311744))))[name = string("blocks_11_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_125_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_11_k_weight_to_fp16_palettized, x = x_197_cast_fp16)[name = string("linear_125_cast_fp16")];
            tensor<fp16, [1024]> blocks_11_k_scale_to_fp16 = const()[name = string("blocks_11_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95312320)))];
            tensor<fp16, [1, 188, 1024]> var_4311_cast_fp16 = mul(x = linear_125_cast_fp16, y = blocks_11_k_scale_to_fp16)[name = string("op_4311_cast_fp16")];
            tensor<int32, [4]> var_4312 = const()[name = string("op_4312"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4313_cast_fp16 = reshape(shape = var_4312, x = var_4311_cast_fp16)[name = string("op_4313_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_11_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95314432))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95576640))))[name = string("blocks_11_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_126_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_11_v_weight_to_fp16_palettized, x = x_197_cast_fp16)[name = string("linear_126_cast_fp16")];
            tensor<fp16, [1024]> blocks_11_v_scale_to_fp16 = const()[name = string("blocks_11_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95577216)))];
            tensor<fp16, [1, 188, 1024]> var_4319_cast_fp16 = mul(x = linear_126_cast_fp16, y = blocks_11_v_scale_to_fp16)[name = string("op_4319_cast_fp16")];
            tensor<int32, [4]> var_4320 = const()[name = string("op_4320"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4321_cast_fp16 = reshape(shape = var_4320, x = var_4319_cast_fp16)[name = string("op_4321_cast_fp16")];
            tensor<int32, [4]> v_23_perm_0 = const()[name = string("v_23_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_106_to_fp16 = const()[name = string("const_106_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95579328)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_23_cast_fp16 = add(x = var_4305_cast_fp16, y = const_106_to_fp16)[name = string("q_u_23_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_107_to_fp16 = const()[name = string("const_107_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95581440)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_23_cast_fp16 = add(x = var_4305_cast_fp16, y = const_107_to_fp16)[name = string("q_v_23_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_11_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95583552))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95845760))))[name = string("blocks_11_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_127_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_11_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_127_cast_fp16")];
            tensor<fp16, [1024]> blocks_11_pos_scale_to_fp16 = const()[name = string("blocks_11_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95846336)))];
            tensor<fp16, [1, 375, 1024]> var_4333_cast_fp16 = mul(x = linear_127_cast_fp16, y = blocks_11_pos_scale_to_fp16)[name = string("op_4333_cast_fp16")];
            tensor<int32, [4]> var_4334 = const()[name = string("op_4334"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_23_cast_fp16 = reshape(shape = var_4334, x = var_4333_cast_fp16)[name = string("p_23_cast_fp16")];
            bool input_289_transpose_x_0 = const()[name = string("input_289_transpose_x_0"), val = bool(false)];
            bool input_289_transpose_y_0 = const()[name = string("input_289_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_236_perm_0 = const()[name = string("transpose_236_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_237_perm_0 = const()[name = string("transpose_237_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_237 = transpose(perm = transpose_237_perm_0, x = p_23_cast_fp16)[name = string("transpose_391")];
            tensor<fp16, [1, 8, 188, 128]> transpose_236 = transpose(perm = transpose_236_perm_0, x = q_v_23_cast_fp16)[name = string("transpose_392")];
            tensor<fp16, [1, 8, 188, 375]> input_289_cast_fp16 = matmul(transpose_x = input_289_transpose_x_0, transpose_y = input_289_transpose_y_0, x = transpose_236, y = transpose_237)[name = string("input_289_cast_fp16")];
            tensor<int32, [8]> var_4340_pad_0 = const()[name = string("op_4340_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_4340_mode_0 = const()[name = string("op_4340_mode_0"), val = string("constant")];
            fp16 const_23_to_fp16 = const()[name = string("const_23_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_4340_cast_fp16 = pad(constant_val = const_23_to_fp16, mode = var_4340_mode_0, pad = var_4340_pad_0, x = input_289_cast_fp16)[name = string("op_4340_cast_fp16")];
            tensor<int32, [4]> var_4341 = const()[name = string("op_4341"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_45_cast_fp16 = reshape(shape = var_4341, x = var_4340_cast_fp16)[name = string("scores_45_cast_fp16")];
            tensor<int32, [4]> var_4345_begin_0 = const()[name = string("op_4345_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_4345_end_0 = const()[name = string("op_4345_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_4345_end_mask_0 = const()[name = string("op_4345_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_4345_cast_fp16 = slice_by_index(begin = var_4345_begin_0, end = var_4345_end_0, end_mask = var_4345_end_mask_0, x = scores_45_cast_fp16)[name = string("op_4345_cast_fp16")];
            tensor<int32, [4]> var_4346 = const()[name = string("op_4346"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_4347_cast_fp16 = reshape(shape = var_4346, x = var_4345_cast_fp16)[name = string("op_4347_cast_fp16")];
            tensor<int32, [4]> bd_23_begin_0 = const()[name = string("bd_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_23_end_0 = const()[name = string("bd_23_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_23_end_mask_0 = const()[name = string("bd_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_23_cast_fp16 = slice_by_index(begin = bd_23_begin_0, end = bd_23_end_0, end_mask = bd_23_end_mask_0, x = var_4347_cast_fp16)[name = string("bd_23_cast_fp16")];
            bool var_4350_transpose_x_0 = const()[name = string("op_4350_transpose_x_0"), val = bool(false)];
            bool var_4350_transpose_y_0 = const()[name = string("op_4350_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_238_perm_0 = const()[name = string("transpose_238_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_239_perm_0 = const()[name = string("transpose_239_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_239 = transpose(perm = transpose_239_perm_0, x = var_4313_cast_fp16)[name = string("transpose_389")];
            tensor<fp16, [1, 8, 188, 128]> transpose_238 = transpose(perm = transpose_238_perm_0, x = q_u_23_cast_fp16)[name = string("transpose_390")];
            tensor<fp16, [1, 8, 188, 188]> var_4350_cast_fp16 = matmul(transpose_x = var_4350_transpose_x_0, transpose_y = var_4350_transpose_y_0, x = transpose_238, y = transpose_239)[name = string("op_4350_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4351_cast_fp16 = add(x = var_4350_cast_fp16, y = bd_23_cast_fp16)[name = string("op_4351_cast_fp16")];
            fp16 var_4352_to_fp16 = const()[name = string("op_4352_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_4353_cast_fp16 = mul(x = var_4351_cast_fp16, y = var_4352_to_fp16)[name = string("op_4353_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_47_cast_fp16 = add(x = var_4353_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_47_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4355_cast_fp16 = softmax(axis = var_4261, x = scores_47_cast_fp16)[name = string("op_4355_cast_fp16")];
            bool out_23_transpose_x_0 = const()[name = string("out_23_transpose_x_0"), val = bool(false)];
            bool out_23_transpose_y_0 = const()[name = string("out_23_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_23_cast_fp16 = transpose(perm = v_23_perm_0, x = var_4321_cast_fp16)[name = string("transpose_388")];
            tensor<fp16, [1, 8, 188, 128]> out_23_cast_fp16 = matmul(transpose_x = out_23_transpose_x_0, transpose_y = out_23_transpose_y_0, x = var_4355_cast_fp16, y = v_23_cast_fp16)[name = string("out_23_cast_fp16")];
            tensor<int32, [4]> var_4357_perm_0 = const()[name = string("op_4357_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4358 = const()[name = string("op_4358"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_4357_cast_fp16 = transpose(perm = var_4357_perm_0, x = out_23_cast_fp16)[name = string("transpose_387")];
            tensor<fp16, [1, 188, 1024]> x_199_cast_fp16 = reshape(shape = var_4358, x = var_4357_cast_fp16)[name = string("x_199_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_11_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95848448))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96110656))))[name = string("blocks_11_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_128_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_11_o_weight_to_fp16_palettized, x = x_199_cast_fp16)[name = string("linear_128_cast_fp16")];
            tensor<fp16, [1024]> blocks_11_o_scale_to_fp16 = const()[name = string("blocks_11_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96111232)))];
            tensor<fp16, [1, 188, 1024]> var_4364_cast_fp16 = mul(x = linear_128_cast_fp16, y = blocks_11_o_scale_to_fp16)[name = string("op_4364_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_291_cast_fp16 = add(x = input_287_cast_fp16, y = var_4364_cast_fp16)[name = string("input_291_cast_fp16")];
            tensor<int32, [1]> x_201_axes_0 = const()[name = string("x_201_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_11_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_11_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96113344)))];
            tensor<fp16, [1024]> enc_layers_11_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_11_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96115456)))];
            tensor<fp16, [1, 188, 1024]> x_201_cast_fp16 = layer_norm(axes = x_201_axes_0, beta = enc_layers_11_norm_conv_bias_to_fp16, epsilon = var_4268_to_fp16, gamma = enc_layers_11_norm_conv_weight_to_fp16, x = input_291_cast_fp16)[name = string("x_201_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_11_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96117568))), lut = tensor<fp16, [128, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96641920))))[name = string("blocks_11_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_129_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_11_pw1_weight_to_fp16_palettized, x = x_201_cast_fp16)[name = string("linear_129_cast_fp16")];
            tensor<fp16, [2048]> blocks_11_pw1_scale_to_fp16 = const()[name = string("blocks_11_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96643008)))];
            tensor<fp16, [1, 188, 2048]> input_293_cast_fp16 = mul(x = linear_129_cast_fp16, y = blocks_11_pw1_scale_to_fp16)[name = string("input_293_cast_fp16")];
            int32 var_4373_split_num_splits_0 = const()[name = string("op_4373_split_num_splits_0"), val = int32(2)];
            int32 var_4373_split_axis_0 = const()[name = string("op_4373_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_4373_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_4373_split_cast_fp16_1 = split(axis = var_4373_split_axis_0, num_splits = var_4373_split_num_splits_0, x = input_293_cast_fp16)[name = string("op_4373_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_4373_split_1_sigmoid_cast_fp16 = sigmoid(x = var_4373_split_cast_fp16_1)[name = string("op_4373_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_4373_cast_fp16 = mul(x = var_4373_split_cast_fp16_0, y = var_4373_split_1_sigmoid_cast_fp16)[name = string("op_4373_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_45_cast_fp16 = mul(x = var_4373_cast_fp16, y = valid_cast_fp16)[name = string("h_45_cast_fp16")];
            tensor<int32, [3]> input_295_perm_0 = const()[name = string("input_295_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_297_pad_type_0 = const()[name = string("input_297_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_297_pad_0 = const()[name = string("input_297_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_297_groups_0 = const()[name = string("input_297_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_297_strides_0 = const()[name = string("input_297_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_297_dilations_0 = const()[name = string("input_297_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_58_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96647168))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96656448))))[name = string("const_58_to_fp16_quantized")];
            tensor<fp16, [1024]> const_59_to_fp16 = const()[name = string("const_59_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96658560)))];
            tensor<fp16, [1, 1024, 188]> input_295_cast_fp16 = transpose(perm = input_295_perm_0, x = h_45_cast_fp16)[name = string("transpose_386")];
            tensor<fp16, [1, 1024, 188]> input_299_cast_fp16 = conv(bias = const_59_to_fp16, dilations = input_297_dilations_0, groups = input_297_groups_0, pad = input_297_pad_0, pad_type = input_297_pad_type_0, strides = input_297_strides_0, weight = const_58_to_fp16_quantized, x = input_295_cast_fp16)[name = string("input_299_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_47_cast_fp16 = silu(x = input_299_cast_fp16)[name = string("h_47_cast_fp16")];
            tensor<int32, [3]> x_203_perm_0 = const()[name = string("x_203_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_11_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96660672))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96922880))))[name = string("blocks_11_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_203_cast_fp16 = transpose(perm = x_203_perm_0, x = h_47_cast_fp16)[name = string("transpose_385")];
            tensor<fp16, [1, 188, 1024]> linear_130_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_11_pw2_weight_to_fp16_palettized, x = x_203_cast_fp16)[name = string("linear_130_cast_fp16")];
            tensor<fp16, [1024]> blocks_11_pw2_scale_to_fp16 = const()[name = string("blocks_11_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96923456)))];
            tensor<fp16, [1, 188, 1024]> var_4388_cast_fp16 = mul(x = linear_130_cast_fp16, y = blocks_11_pw2_scale_to_fp16)[name = string("op_4388_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_301_cast_fp16 = add(x = input_291_cast_fp16, y = var_4388_cast_fp16)[name = string("input_301_cast_fp16")];
            tensor<int32, [1]> x_205_axes_0 = const()[name = string("x_205_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_11_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_11_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96925568)))];
            tensor<fp16, [1024]> enc_layers_11_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_11_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96927680)))];
            tensor<fp16, [1, 188, 1024]> x_205_cast_fp16 = layer_norm(axes = x_205_axes_0, beta = enc_layers_11_norm_feed_forward2_bias_to_fp16, epsilon = var_4268_to_fp16, gamma = enc_layers_11_norm_feed_forward2_weight_to_fp16, x = input_301_cast_fp16)[name = string("x_205_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_11_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96929792))), lut = tensor<fp16, [256, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97978432))))[name = string("blocks_11_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_131_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_11_ff2_1_weight_to_fp16_palettized, x = x_205_cast_fp16)[name = string("linear_131_cast_fp16")];
            tensor<fp16, [4096]> blocks_11_ff2_1_scale_to_fp16 = const()[name = string("blocks_11_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97980544)))];
            tensor<fp16, [1, 188, 4096]> input_303_cast_fp16 = mul(x = linear_131_cast_fp16, y = blocks_11_ff2_1_scale_to_fp16)[name = string("input_303_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_207_cast_fp16 = silu(x = input_303_cast_fp16)[name = string("x_207_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_11_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97988800))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99037440))))[name = string("blocks_11_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_132_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_11_ff2_2_weight_to_fp16_palettized, x = x_207_cast_fp16)[name = string("linear_132_cast_fp16")];
            tensor<fp16, [1024]> blocks_11_ff2_2_scale_to_fp16 = const()[name = string("blocks_11_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99038016)))];
            tensor<fp16, [1, 188, 1024]> var_4402_cast_fp16 = mul(x = linear_132_cast_fp16, y = blocks_11_ff2_2_scale_to_fp16)[name = string("op_4402_cast_fp16")];
            fp16 var_4403_to_fp16 = const()[name = string("op_4403_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4404_cast_fp16 = mul(x = var_4402_cast_fp16, y = var_4403_to_fp16)[name = string("op_4404_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_305_cast_fp16 = add(x = input_301_cast_fp16, y = var_4404_cast_fp16)[name = string("input_305_cast_fp16")];
            tensor<int32, [1]> input_307_axes_0 = const()[name = string("input_307_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_11_norm_out_weight_to_fp16 = const()[name = string("enc_layers_11_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99040128)))];
            tensor<fp16, [1024]> enc_layers_11_norm_out_bias_to_fp16 = const()[name = string("enc_layers_11_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99042240)))];
            tensor<fp16, [1, 188, 1024]> input_307_cast_fp16 = layer_norm(axes = input_307_axes_0, beta = enc_layers_11_norm_out_bias_to_fp16, epsilon = var_4268_to_fp16, gamma = enc_layers_11_norm_out_weight_to_fp16, x = input_305_cast_fp16)[name = string("input_307_cast_fp16")];
            int32 var_4418 = const()[name = string("op_4418"), val = int32(-1)];
            tensor<int32, [1]> x_209_axes_0 = const()[name = string("x_209_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_12_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_12_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99044352)))];
            tensor<fp16, [1024]> enc_layers_12_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_12_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99046464)))];
            fp16 var_4425_to_fp16 = const()[name = string("op_4425_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_209_cast_fp16 = layer_norm(axes = x_209_axes_0, beta = enc_layers_12_norm_feed_forward1_bias_to_fp16, epsilon = var_4425_to_fp16, gamma = enc_layers_12_norm_feed_forward1_weight_to_fp16, x = input_307_cast_fp16)[name = string("x_209_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_12_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99048576))), lut = tensor<fp16, [256, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100621504))))[name = string("blocks_12_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_133_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_12_ff1_1_weight_to_fp16_palettized, x = x_209_cast_fp16)[name = string("linear_133_cast_fp16")];
            tensor<fp16, [4096]> blocks_12_ff1_1_scale_to_fp16 = const()[name = string("blocks_12_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100625664)))];
            tensor<fp16, [1, 188, 4096]> input_309_cast_fp16 = mul(x = linear_133_cast_fp16, y = blocks_12_ff1_1_scale_to_fp16)[name = string("input_309_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_211_cast_fp16 = silu(x = input_309_cast_fp16)[name = string("x_211_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_12_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100633920))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102206848))))[name = string("blocks_12_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_134_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_12_ff1_2_weight_to_fp16_palettized, x = x_211_cast_fp16)[name = string("linear_134_cast_fp16")];
            tensor<fp16, [1024]> blocks_12_ff1_2_scale_to_fp16 = const()[name = string("blocks_12_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102207936)))];
            tensor<fp16, [1, 188, 1024]> var_4450_cast_fp16 = mul(x = linear_134_cast_fp16, y = blocks_12_ff1_2_scale_to_fp16)[name = string("op_4450_cast_fp16")];
            fp16 var_4451_to_fp16 = const()[name = string("op_4451_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4452_cast_fp16 = mul(x = var_4450_cast_fp16, y = var_4451_to_fp16)[name = string("op_4452_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_311_cast_fp16 = add(x = input_307_cast_fp16, y = var_4452_cast_fp16)[name = string("input_311_cast_fp16")];
            tensor<int32, [1]> x_213_axes_0 = const()[name = string("x_213_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_12_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_12_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102210048)))];
            tensor<fp16, [1024]> enc_layers_12_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_12_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102212160)))];
            tensor<fp16, [1, 188, 1024]> x_213_cast_fp16 = layer_norm(axes = x_213_axes_0, beta = enc_layers_12_norm_self_att_bias_to_fp16, epsilon = var_4425_to_fp16, gamma = enc_layers_12_norm_self_att_weight_to_fp16, x = input_311_cast_fp16)[name = string("x_213_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_12_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102214272))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102476480))))[name = string("blocks_12_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_135_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_12_q_weight_to_fp16_palettized, x = x_213_cast_fp16)[name = string("linear_135_cast_fp16")];
            tensor<fp16, [1024]> blocks_12_q_scale_to_fp16 = const()[name = string("blocks_12_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102477056)))];
            tensor<fp16, [1, 188, 1024]> var_4460_cast_fp16 = mul(x = linear_135_cast_fp16, y = blocks_12_q_scale_to_fp16)[name = string("op_4460_cast_fp16")];
            tensor<int32, [4]> var_4461 = const()[name = string("op_4461"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4462_cast_fp16 = reshape(shape = var_4461, x = var_4460_cast_fp16)[name = string("op_4462_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_12_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102479168))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102741376))))[name = string("blocks_12_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_136_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_12_k_weight_to_fp16_palettized, x = x_213_cast_fp16)[name = string("linear_136_cast_fp16")];
            tensor<fp16, [1024]> blocks_12_k_scale_to_fp16 = const()[name = string("blocks_12_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102741952)))];
            tensor<fp16, [1, 188, 1024]> var_4468_cast_fp16 = mul(x = linear_136_cast_fp16, y = blocks_12_k_scale_to_fp16)[name = string("op_4468_cast_fp16")];
            tensor<int32, [4]> var_4469 = const()[name = string("op_4469"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4470_cast_fp16 = reshape(shape = var_4469, x = var_4468_cast_fp16)[name = string("op_4470_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_12_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102744064))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103006272))))[name = string("blocks_12_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_137_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_12_v_weight_to_fp16_palettized, x = x_213_cast_fp16)[name = string("linear_137_cast_fp16")];
            tensor<fp16, [1024]> blocks_12_v_scale_to_fp16 = const()[name = string("blocks_12_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103006848)))];
            tensor<fp16, [1, 188, 1024]> var_4476_cast_fp16 = mul(x = linear_137_cast_fp16, y = blocks_12_v_scale_to_fp16)[name = string("op_4476_cast_fp16")];
            tensor<int32, [4]> var_4477 = const()[name = string("op_4477"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4478_cast_fp16 = reshape(shape = var_4477, x = var_4476_cast_fp16)[name = string("op_4478_cast_fp16")];
            tensor<int32, [4]> v_25_perm_0 = const()[name = string("v_25_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_108_to_fp16 = const()[name = string("const_108_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103008960)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_25_cast_fp16 = add(x = var_4462_cast_fp16, y = const_108_to_fp16)[name = string("q_u_25_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_109_to_fp16 = const()[name = string("const_109_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103011072)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_25_cast_fp16 = add(x = var_4462_cast_fp16, y = const_109_to_fp16)[name = string("q_v_25_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_12_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103013184))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103275392))))[name = string("blocks_12_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_138_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_12_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_138_cast_fp16")];
            tensor<fp16, [1024]> blocks_12_pos_scale_to_fp16 = const()[name = string("blocks_12_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103275968)))];
            tensor<fp16, [1, 375, 1024]> var_4490_cast_fp16 = mul(x = linear_138_cast_fp16, y = blocks_12_pos_scale_to_fp16)[name = string("op_4490_cast_fp16")];
            tensor<int32, [4]> var_4491 = const()[name = string("op_4491"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_25_cast_fp16 = reshape(shape = var_4491, x = var_4490_cast_fp16)[name = string("p_25_cast_fp16")];
            bool input_313_transpose_x_0 = const()[name = string("input_313_transpose_x_0"), val = bool(false)];
            bool input_313_transpose_y_0 = const()[name = string("input_313_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_240_perm_0 = const()[name = string("transpose_240_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_241_perm_0 = const()[name = string("transpose_241_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_241 = transpose(perm = transpose_241_perm_0, x = p_25_cast_fp16)[name = string("transpose_383")];
            tensor<fp16, [1, 8, 188, 128]> transpose_240 = transpose(perm = transpose_240_perm_0, x = q_v_25_cast_fp16)[name = string("transpose_384")];
            tensor<fp16, [1, 8, 188, 375]> input_313_cast_fp16 = matmul(transpose_x = input_313_transpose_x_0, transpose_y = input_313_transpose_y_0, x = transpose_240, y = transpose_241)[name = string("input_313_cast_fp16")];
            tensor<int32, [8]> var_4497_pad_0 = const()[name = string("op_4497_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_4497_mode_0 = const()[name = string("op_4497_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]> var_4497_cast_fp16 = pad(constant_val = const_24_to_fp16, mode = var_4497_mode_0, pad = var_4497_pad_0, x = input_313_cast_fp16)[name = string("op_4497_cast_fp16")];
            tensor<int32, [4]> var_4498 = const()[name = string("op_4498"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_49_cast_fp16 = reshape(shape = var_4498, x = var_4497_cast_fp16)[name = string("scores_49_cast_fp16")];
            tensor<int32, [4]> var_4502_begin_0 = const()[name = string("op_4502_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_4502_end_0 = const()[name = string("op_4502_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_4502_end_mask_0 = const()[name = string("op_4502_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_4502_cast_fp16 = slice_by_index(begin = var_4502_begin_0, end = var_4502_end_0, end_mask = var_4502_end_mask_0, x = scores_49_cast_fp16)[name = string("op_4502_cast_fp16")];
            tensor<int32, [4]> var_4503 = const()[name = string("op_4503"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_4504_cast_fp16 = reshape(shape = var_4503, x = var_4502_cast_fp16)[name = string("op_4504_cast_fp16")];
            tensor<int32, [4]> bd_25_begin_0 = const()[name = string("bd_25_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_25_end_0 = const()[name = string("bd_25_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_25_end_mask_0 = const()[name = string("bd_25_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_25_cast_fp16 = slice_by_index(begin = bd_25_begin_0, end = bd_25_end_0, end_mask = bd_25_end_mask_0, x = var_4504_cast_fp16)[name = string("bd_25_cast_fp16")];
            bool var_4507_transpose_x_0 = const()[name = string("op_4507_transpose_x_0"), val = bool(false)];
            bool var_4507_transpose_y_0 = const()[name = string("op_4507_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_242_perm_0 = const()[name = string("transpose_242_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_243_perm_0 = const()[name = string("transpose_243_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_243 = transpose(perm = transpose_243_perm_0, x = var_4470_cast_fp16)[name = string("transpose_381")];
            tensor<fp16, [1, 8, 188, 128]> transpose_242 = transpose(perm = transpose_242_perm_0, x = q_u_25_cast_fp16)[name = string("transpose_382")];
            tensor<fp16, [1, 8, 188, 188]> var_4507_cast_fp16 = matmul(transpose_x = var_4507_transpose_x_0, transpose_y = var_4507_transpose_y_0, x = transpose_242, y = transpose_243)[name = string("op_4507_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4508_cast_fp16 = add(x = var_4507_cast_fp16, y = bd_25_cast_fp16)[name = string("op_4508_cast_fp16")];
            fp16 var_4509_to_fp16 = const()[name = string("op_4509_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_4510_cast_fp16 = mul(x = var_4508_cast_fp16, y = var_4509_to_fp16)[name = string("op_4510_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_51_cast_fp16 = add(x = var_4510_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_51_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4512_cast_fp16 = softmax(axis = var_4418, x = scores_51_cast_fp16)[name = string("op_4512_cast_fp16")];
            bool out_25_transpose_x_0 = const()[name = string("out_25_transpose_x_0"), val = bool(false)];
            bool out_25_transpose_y_0 = const()[name = string("out_25_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_25_cast_fp16 = transpose(perm = v_25_perm_0, x = var_4478_cast_fp16)[name = string("transpose_380")];
            tensor<fp16, [1, 8, 188, 128]> out_25_cast_fp16 = matmul(transpose_x = out_25_transpose_x_0, transpose_y = out_25_transpose_y_0, x = var_4512_cast_fp16, y = v_25_cast_fp16)[name = string("out_25_cast_fp16")];
            tensor<int32, [4]> var_4514_perm_0 = const()[name = string("op_4514_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4515 = const()[name = string("op_4515"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_4514_cast_fp16 = transpose(perm = var_4514_perm_0, x = out_25_cast_fp16)[name = string("transpose_379")];
            tensor<fp16, [1, 188, 1024]> x_215_cast_fp16 = reshape(shape = var_4515, x = var_4514_cast_fp16)[name = string("x_215_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_12_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103278080))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103540288))))[name = string("blocks_12_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_139_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_12_o_weight_to_fp16_palettized, x = x_215_cast_fp16)[name = string("linear_139_cast_fp16")];
            tensor<fp16, [1024]> blocks_12_o_scale_to_fp16 = const()[name = string("blocks_12_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103540864)))];
            tensor<fp16, [1, 188, 1024]> var_4521_cast_fp16 = mul(x = linear_139_cast_fp16, y = blocks_12_o_scale_to_fp16)[name = string("op_4521_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_315_cast_fp16 = add(x = input_311_cast_fp16, y = var_4521_cast_fp16)[name = string("input_315_cast_fp16")];
            tensor<int32, [1]> x_217_axes_0 = const()[name = string("x_217_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_12_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_12_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103542976)))];
            tensor<fp16, [1024]> enc_layers_12_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_12_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103545088)))];
            tensor<fp16, [1, 188, 1024]> x_217_cast_fp16 = layer_norm(axes = x_217_axes_0, beta = enc_layers_12_norm_conv_bias_to_fp16, epsilon = var_4425_to_fp16, gamma = enc_layers_12_norm_conv_weight_to_fp16, x = input_315_cast_fp16)[name = string("x_217_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_12_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103547200))), lut = tensor<fp16, [128, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104071552))))[name = string("blocks_12_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_140_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_12_pw1_weight_to_fp16_palettized, x = x_217_cast_fp16)[name = string("linear_140_cast_fp16")];
            tensor<fp16, [2048]> blocks_12_pw1_scale_to_fp16 = const()[name = string("blocks_12_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104072640)))];
            tensor<fp16, [1, 188, 2048]> input_317_cast_fp16 = mul(x = linear_140_cast_fp16, y = blocks_12_pw1_scale_to_fp16)[name = string("input_317_cast_fp16")];
            int32 var_4530_split_num_splits_0 = const()[name = string("op_4530_split_num_splits_0"), val = int32(2)];
            int32 var_4530_split_axis_0 = const()[name = string("op_4530_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_4530_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_4530_split_cast_fp16_1 = split(axis = var_4530_split_axis_0, num_splits = var_4530_split_num_splits_0, x = input_317_cast_fp16)[name = string("op_4530_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_4530_split_1_sigmoid_cast_fp16 = sigmoid(x = var_4530_split_cast_fp16_1)[name = string("op_4530_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_4530_cast_fp16 = mul(x = var_4530_split_cast_fp16_0, y = var_4530_split_1_sigmoid_cast_fp16)[name = string("op_4530_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_49_cast_fp16 = mul(x = var_4530_cast_fp16, y = valid_cast_fp16)[name = string("h_49_cast_fp16")];
            tensor<int32, [3]> input_319_perm_0 = const()[name = string("input_319_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_321_pad_type_0 = const()[name = string("input_321_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_321_pad_0 = const()[name = string("input_321_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_321_groups_0 = const()[name = string("input_321_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_321_strides_0 = const()[name = string("input_321_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_321_dilations_0 = const()[name = string("input_321_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_60_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104076800))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104086080))))[name = string("const_60_to_fp16_quantized")];
            tensor<fp16, [1024]> const_61_to_fp16 = const()[name = string("const_61_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104088192)))];
            tensor<fp16, [1, 1024, 188]> input_319_cast_fp16 = transpose(perm = input_319_perm_0, x = h_49_cast_fp16)[name = string("transpose_378")];
            tensor<fp16, [1, 1024, 188]> input_323_cast_fp16 = conv(bias = const_61_to_fp16, dilations = input_321_dilations_0, groups = input_321_groups_0, pad = input_321_pad_0, pad_type = input_321_pad_type_0, strides = input_321_strides_0, weight = const_60_to_fp16_quantized, x = input_319_cast_fp16)[name = string("input_323_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_51_cast_fp16 = silu(x = input_323_cast_fp16)[name = string("h_51_cast_fp16")];
            tensor<int32, [3]> x_219_perm_0 = const()[name = string("x_219_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_12_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104090304))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104352512))))[name = string("blocks_12_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_219_cast_fp16 = transpose(perm = x_219_perm_0, x = h_51_cast_fp16)[name = string("transpose_377")];
            tensor<fp16, [1, 188, 1024]> linear_141_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_12_pw2_weight_to_fp16_palettized, x = x_219_cast_fp16)[name = string("linear_141_cast_fp16")];
            tensor<fp16, [1024]> blocks_12_pw2_scale_to_fp16 = const()[name = string("blocks_12_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104353088)))];
            tensor<fp16, [1, 188, 1024]> var_4545_cast_fp16 = mul(x = linear_141_cast_fp16, y = blocks_12_pw2_scale_to_fp16)[name = string("op_4545_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_325_cast_fp16 = add(x = input_315_cast_fp16, y = var_4545_cast_fp16)[name = string("input_325_cast_fp16")];
            tensor<int32, [1]> x_221_axes_0 = const()[name = string("x_221_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_12_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_12_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104355200)))];
            tensor<fp16, [1024]> enc_layers_12_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_12_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104357312)))];
            tensor<fp16, [1, 188, 1024]> x_221_cast_fp16 = layer_norm(axes = x_221_axes_0, beta = enc_layers_12_norm_feed_forward2_bias_to_fp16, epsilon = var_4425_to_fp16, gamma = enc_layers_12_norm_feed_forward2_weight_to_fp16, x = input_325_cast_fp16)[name = string("x_221_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_12_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104359424))), lut = tensor<fp16, [256, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105932352))))[name = string("blocks_12_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_142_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_12_ff2_1_weight_to_fp16_palettized, x = x_221_cast_fp16)[name = string("linear_142_cast_fp16")];
            tensor<fp16, [4096]> blocks_12_ff2_1_scale_to_fp16 = const()[name = string("blocks_12_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105936512)))];
            tensor<fp16, [1, 188, 4096]> input_327_cast_fp16 = mul(x = linear_142_cast_fp16, y = blocks_12_ff2_1_scale_to_fp16)[name = string("input_327_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_223_cast_fp16 = silu(x = input_327_cast_fp16)[name = string("x_223_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_12_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105944768))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107517696))))[name = string("blocks_12_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_143_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_12_ff2_2_weight_to_fp16_palettized, x = x_223_cast_fp16)[name = string("linear_143_cast_fp16")];
            tensor<fp16, [1024]> blocks_12_ff2_2_scale_to_fp16 = const()[name = string("blocks_12_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107518784)))];
            tensor<fp16, [1, 188, 1024]> var_4559_cast_fp16 = mul(x = linear_143_cast_fp16, y = blocks_12_ff2_2_scale_to_fp16)[name = string("op_4559_cast_fp16")];
            fp16 var_4560_to_fp16 = const()[name = string("op_4560_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4561_cast_fp16 = mul(x = var_4559_cast_fp16, y = var_4560_to_fp16)[name = string("op_4561_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_329_cast_fp16 = add(x = input_325_cast_fp16, y = var_4561_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]> enc_layers_12_norm_out_weight_to_fp16 = const()[name = string("enc_layers_12_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107520896)))];
            tensor<fp16, [1024]> enc_layers_12_norm_out_bias_to_fp16 = const()[name = string("enc_layers_12_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107523008)))];
            tensor<fp16, [1, 188, 1024]> input_331_cast_fp16 = layer_norm(axes = input_331_axes_0, beta = enc_layers_12_norm_out_bias_to_fp16, epsilon = var_4425_to_fp16, gamma = enc_layers_12_norm_out_weight_to_fp16, x = input_329_cast_fp16)[name = string("input_331_cast_fp16")];
            int32 var_4575 = const()[name = string("op_4575"), val = int32(-1)];
            tensor<int32, [1]> x_225_axes_0 = const()[name = string("x_225_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_13_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_13_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107525120)))];
            tensor<fp16, [1024]> enc_layers_13_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_13_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107527232)))];
            fp16 var_4582_to_fp16 = const()[name = string("op_4582_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_225_cast_fp16 = layer_norm(axes = x_225_axes_0, beta = enc_layers_13_norm_feed_forward1_bias_to_fp16, epsilon = var_4582_to_fp16, gamma = enc_layers_13_norm_feed_forward1_weight_to_fp16, x = input_331_cast_fp16)[name = string("x_225_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_13_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107529344))), lut = tensor<fp16, [256, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109102272))))[name = string("blocks_13_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_144_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_13_ff1_1_weight_to_fp16_palettized, x = x_225_cast_fp16)[name = string("linear_144_cast_fp16")];
            tensor<fp16, [4096]> blocks_13_ff1_1_scale_to_fp16 = const()[name = string("blocks_13_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109106432)))];
            tensor<fp16, [1, 188, 4096]> input_333_cast_fp16 = mul(x = linear_144_cast_fp16, y = blocks_13_ff1_1_scale_to_fp16)[name = string("input_333_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_227_cast_fp16 = silu(x = input_333_cast_fp16)[name = string("x_227_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_13_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109114688))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110687616))))[name = string("blocks_13_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_145_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_13_ff1_2_weight_to_fp16_palettized, x = x_227_cast_fp16)[name = string("linear_145_cast_fp16")];
            tensor<fp16, [1024]> blocks_13_ff1_2_scale_to_fp16 = const()[name = string("blocks_13_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110688704)))];
            tensor<fp16, [1, 188, 1024]> var_4607_cast_fp16 = mul(x = linear_145_cast_fp16, y = blocks_13_ff1_2_scale_to_fp16)[name = string("op_4607_cast_fp16")];
            fp16 var_4608_to_fp16 = const()[name = string("op_4608_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4609_cast_fp16 = mul(x = var_4607_cast_fp16, y = var_4608_to_fp16)[name = string("op_4609_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_335_cast_fp16 = add(x = input_331_cast_fp16, y = var_4609_cast_fp16)[name = string("input_335_cast_fp16")];
            tensor<int32, [1]> x_229_axes_0 = const()[name = string("x_229_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_13_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_13_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110690816)))];
            tensor<fp16, [1024]> enc_layers_13_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_13_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110692928)))];
            tensor<fp16, [1, 188, 1024]> x_229_cast_fp16 = layer_norm(axes = x_229_axes_0, beta = enc_layers_13_norm_self_att_bias_to_fp16, epsilon = var_4582_to_fp16, gamma = enc_layers_13_norm_self_att_weight_to_fp16, x = input_335_cast_fp16)[name = string("x_229_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_13_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110695040))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111088320))))[name = string("blocks_13_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_146_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_13_q_weight_to_fp16_palettized, x = x_229_cast_fp16)[name = string("linear_146_cast_fp16")];
            tensor<fp16, [1024]> blocks_13_q_scale_to_fp16 = const()[name = string("blocks_13_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111089408)))];
            tensor<fp16, [1, 188, 1024]> var_4617_cast_fp16 = mul(x = linear_146_cast_fp16, y = blocks_13_q_scale_to_fp16)[name = string("op_4617_cast_fp16")];
            tensor<int32, [4]> var_4618 = const()[name = string("op_4618"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4619_cast_fp16 = reshape(shape = var_4618, x = var_4617_cast_fp16)[name = string("op_4619_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_13_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111091520))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111484800))))[name = string("blocks_13_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_147_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_13_k_weight_to_fp16_palettized, x = x_229_cast_fp16)[name = string("linear_147_cast_fp16")];
            tensor<fp16, [1024]> blocks_13_k_scale_to_fp16 = const()[name = string("blocks_13_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111485888)))];
            tensor<fp16, [1, 188, 1024]> var_4625_cast_fp16 = mul(x = linear_147_cast_fp16, y = blocks_13_k_scale_to_fp16)[name = string("op_4625_cast_fp16")];
            tensor<int32, [4]> var_4626 = const()[name = string("op_4626"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4627_cast_fp16 = reshape(shape = var_4626, x = var_4625_cast_fp16)[name = string("op_4627_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_13_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111488000))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111881280))))[name = string("blocks_13_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_148_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_13_v_weight_to_fp16_palettized, x = x_229_cast_fp16)[name = string("linear_148_cast_fp16")];
            tensor<fp16, [1024]> blocks_13_v_scale_to_fp16 = const()[name = string("blocks_13_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111882368)))];
            tensor<fp16, [1, 188, 1024]> var_4633_cast_fp16 = mul(x = linear_148_cast_fp16, y = blocks_13_v_scale_to_fp16)[name = string("op_4633_cast_fp16")];
            tensor<int32, [4]> var_4634 = const()[name = string("op_4634"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4635_cast_fp16 = reshape(shape = var_4634, x = var_4633_cast_fp16)[name = string("op_4635_cast_fp16")];
            tensor<int32, [4]> v_27_perm_0 = const()[name = string("v_27_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_110_to_fp16 = const()[name = string("const_110_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111884480)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_27_cast_fp16 = add(x = var_4619_cast_fp16, y = const_110_to_fp16)[name = string("q_u_27_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_111_to_fp16 = const()[name = string("const_111_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111886592)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_27_cast_fp16 = add(x = var_4619_cast_fp16, y = const_111_to_fp16)[name = string("q_v_27_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_13_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111888704))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112281984))))[name = string("blocks_13_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_149_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_13_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_149_cast_fp16")];
            tensor<fp16, [1024]> blocks_13_pos_scale_to_fp16 = const()[name = string("blocks_13_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112283072)))];
            tensor<fp16, [1, 375, 1024]> var_4647_cast_fp16 = mul(x = linear_149_cast_fp16, y = blocks_13_pos_scale_to_fp16)[name = string("op_4647_cast_fp16")];
            tensor<int32, [4]> var_4648 = const()[name = string("op_4648"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_27_cast_fp16 = reshape(shape = var_4648, x = var_4647_cast_fp16)[name = string("p_27_cast_fp16")];
            bool input_337_transpose_x_0 = const()[name = string("input_337_transpose_x_0"), val = bool(false)];
            bool input_337_transpose_y_0 = const()[name = string("input_337_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_244_perm_0 = const()[name = string("transpose_244_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_245_perm_0 = const()[name = string("transpose_245_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_245 = transpose(perm = transpose_245_perm_0, x = p_27_cast_fp16)[name = string("transpose_375")];
            tensor<fp16, [1, 8, 188, 128]> transpose_244 = transpose(perm = transpose_244_perm_0, x = q_v_27_cast_fp16)[name = string("transpose_376")];
            tensor<fp16, [1, 8, 188, 375]> input_337_cast_fp16 = matmul(transpose_x = input_337_transpose_x_0, transpose_y = input_337_transpose_y_0, x = transpose_244, y = transpose_245)[name = string("input_337_cast_fp16")];
            tensor<int32, [8]> var_4654_pad_0 = const()[name = string("op_4654_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_4654_mode_0 = const()[name = string("op_4654_mode_0"), val = string("constant")];
            fp16 const_25_to_fp16 = const()[name = string("const_25_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_4654_cast_fp16 = pad(constant_val = const_25_to_fp16, mode = var_4654_mode_0, pad = var_4654_pad_0, x = input_337_cast_fp16)[name = string("op_4654_cast_fp16")];
            tensor<int32, [4]> var_4655 = const()[name = string("op_4655"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_53_cast_fp16 = reshape(shape = var_4655, x = var_4654_cast_fp16)[name = string("scores_53_cast_fp16")];
            tensor<int32, [4]> var_4659_begin_0 = const()[name = string("op_4659_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_4659_end_0 = const()[name = string("op_4659_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_4659_end_mask_0 = const()[name = string("op_4659_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_4659_cast_fp16 = slice_by_index(begin = var_4659_begin_0, end = var_4659_end_0, end_mask = var_4659_end_mask_0, x = scores_53_cast_fp16)[name = string("op_4659_cast_fp16")];
            tensor<int32, [4]> var_4660 = const()[name = string("op_4660"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_4661_cast_fp16 = reshape(shape = var_4660, x = var_4659_cast_fp16)[name = string("op_4661_cast_fp16")];
            tensor<int32, [4]> bd_27_begin_0 = const()[name = string("bd_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_27_end_0 = const()[name = string("bd_27_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_27_end_mask_0 = const()[name = string("bd_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_27_cast_fp16 = slice_by_index(begin = bd_27_begin_0, end = bd_27_end_0, end_mask = bd_27_end_mask_0, x = var_4661_cast_fp16)[name = string("bd_27_cast_fp16")];
            bool var_4664_transpose_x_0 = const()[name = string("op_4664_transpose_x_0"), val = bool(false)];
            bool var_4664_transpose_y_0 = const()[name = string("op_4664_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_246_perm_0 = const()[name = string("transpose_246_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_247_perm_0 = const()[name = string("transpose_247_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_247 = transpose(perm = transpose_247_perm_0, x = var_4627_cast_fp16)[name = string("transpose_373")];
            tensor<fp16, [1, 8, 188, 128]> transpose_246 = transpose(perm = transpose_246_perm_0, x = q_u_27_cast_fp16)[name = string("transpose_374")];
            tensor<fp16, [1, 8, 188, 188]> var_4664_cast_fp16 = matmul(transpose_x = var_4664_transpose_x_0, transpose_y = var_4664_transpose_y_0, x = transpose_246, y = transpose_247)[name = string("op_4664_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4665_cast_fp16 = add(x = var_4664_cast_fp16, y = bd_27_cast_fp16)[name = string("op_4665_cast_fp16")];
            fp16 var_4666_to_fp16 = const()[name = string("op_4666_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_4667_cast_fp16 = mul(x = var_4665_cast_fp16, y = var_4666_to_fp16)[name = string("op_4667_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_55_cast_fp16 = add(x = var_4667_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_55_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4669_cast_fp16 = softmax(axis = var_4575, x = scores_55_cast_fp16)[name = string("op_4669_cast_fp16")];
            bool out_27_transpose_x_0 = const()[name = string("out_27_transpose_x_0"), val = bool(false)];
            bool out_27_transpose_y_0 = const()[name = string("out_27_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_27_cast_fp16 = transpose(perm = v_27_perm_0, x = var_4635_cast_fp16)[name = string("transpose_372")];
            tensor<fp16, [1, 8, 188, 128]> out_27_cast_fp16 = matmul(transpose_x = out_27_transpose_x_0, transpose_y = out_27_transpose_y_0, x = var_4669_cast_fp16, y = v_27_cast_fp16)[name = string("out_27_cast_fp16")];
            tensor<int32, [4]> var_4671_perm_0 = const()[name = string("op_4671_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4672 = const()[name = string("op_4672"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_4671_cast_fp16 = transpose(perm = var_4671_perm_0, x = out_27_cast_fp16)[name = string("transpose_371")];
            tensor<fp16, [1, 188, 1024]> x_231_cast_fp16 = reshape(shape = var_4672, x = var_4671_cast_fp16)[name = string("x_231_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_13_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112285184))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112678464))))[name = string("blocks_13_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_150_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_13_o_weight_to_fp16_palettized, x = x_231_cast_fp16)[name = string("linear_150_cast_fp16")];
            tensor<fp16, [1024]> blocks_13_o_scale_to_fp16 = const()[name = string("blocks_13_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112679552)))];
            tensor<fp16, [1, 188, 1024]> var_4678_cast_fp16 = mul(x = linear_150_cast_fp16, y = blocks_13_o_scale_to_fp16)[name = string("op_4678_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_339_cast_fp16 = add(x = input_335_cast_fp16, y = var_4678_cast_fp16)[name = string("input_339_cast_fp16")];
            tensor<int32, [1]> x_233_axes_0 = const()[name = string("x_233_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_13_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_13_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112681664)))];
            tensor<fp16, [1024]> enc_layers_13_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_13_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112683776)))];
            tensor<fp16, [1, 188, 1024]> x_233_cast_fp16 = layer_norm(axes = x_233_axes_0, beta = enc_layers_13_norm_conv_bias_to_fp16, epsilon = var_4582_to_fp16, gamma = enc_layers_13_norm_conv_weight_to_fp16, x = input_339_cast_fp16)[name = string("x_233_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_13_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112685888))), lut = tensor<fp16, [128, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113472384))))[name = string("blocks_13_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_151_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_13_pw1_weight_to_fp16_palettized, x = x_233_cast_fp16)[name = string("linear_151_cast_fp16")];
            tensor<fp16, [2048]> blocks_13_pw1_scale_to_fp16 = const()[name = string("blocks_13_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113474496)))];
            tensor<fp16, [1, 188, 2048]> input_341_cast_fp16 = mul(x = linear_151_cast_fp16, y = blocks_13_pw1_scale_to_fp16)[name = string("input_341_cast_fp16")];
            int32 var_4687_split_num_splits_0 = const()[name = string("op_4687_split_num_splits_0"), val = int32(2)];
            int32 var_4687_split_axis_0 = const()[name = string("op_4687_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_4687_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_4687_split_cast_fp16_1 = split(axis = var_4687_split_axis_0, num_splits = var_4687_split_num_splits_0, x = input_341_cast_fp16)[name = string("op_4687_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_4687_split_1_sigmoid_cast_fp16 = sigmoid(x = var_4687_split_cast_fp16_1)[name = string("op_4687_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_4687_cast_fp16 = mul(x = var_4687_split_cast_fp16_0, y = var_4687_split_1_sigmoid_cast_fp16)[name = string("op_4687_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_53_cast_fp16 = mul(x = var_4687_cast_fp16, y = valid_cast_fp16)[name = string("h_53_cast_fp16")];
            tensor<int32, [3]> input_343_perm_0 = const()[name = string("input_343_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_345_pad_type_0 = const()[name = string("input_345_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_345_pad_0 = const()[name = string("input_345_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_345_groups_0 = const()[name = string("input_345_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_345_strides_0 = const()[name = string("input_345_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_345_dilations_0 = const()[name = string("input_345_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_62_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113478656))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113487936))))[name = string("const_62_to_fp16_quantized")];
            tensor<fp16, [1024]> const_63_to_fp16 = const()[name = string("const_63_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113490048)))];
            tensor<fp16, [1, 1024, 188]> input_343_cast_fp16 = transpose(perm = input_343_perm_0, x = h_53_cast_fp16)[name = string("transpose_370")];
            tensor<fp16, [1, 1024, 188]> input_347_cast_fp16 = conv(bias = const_63_to_fp16, dilations = input_345_dilations_0, groups = input_345_groups_0, pad = input_345_pad_0, pad_type = input_345_pad_type_0, strides = input_345_strides_0, weight = const_62_to_fp16_quantized, x = input_343_cast_fp16)[name = string("input_347_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_55_cast_fp16 = silu(x = input_347_cast_fp16)[name = string("h_55_cast_fp16")];
            tensor<int32, [3]> x_235_perm_0 = const()[name = string("x_235_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_13_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113492160))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113885440))))[name = string("blocks_13_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_235_cast_fp16 = transpose(perm = x_235_perm_0, x = h_55_cast_fp16)[name = string("transpose_369")];
            tensor<fp16, [1, 188, 1024]> linear_152_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_13_pw2_weight_to_fp16_palettized, x = x_235_cast_fp16)[name = string("linear_152_cast_fp16")];
            tensor<fp16, [1024]> blocks_13_pw2_scale_to_fp16 = const()[name = string("blocks_13_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113886528)))];
            tensor<fp16, [1, 188, 1024]> var_4702_cast_fp16 = mul(x = linear_152_cast_fp16, y = blocks_13_pw2_scale_to_fp16)[name = string("op_4702_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_349_cast_fp16 = add(x = input_339_cast_fp16, y = var_4702_cast_fp16)[name = string("input_349_cast_fp16")];
            tensor<int32, [1]> x_237_axes_0 = const()[name = string("x_237_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_13_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_13_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113888640)))];
            tensor<fp16, [1024]> enc_layers_13_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_13_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113890752)))];
            tensor<fp16, [1, 188, 1024]> x_237_cast_fp16 = layer_norm(axes = x_237_axes_0, beta = enc_layers_13_norm_feed_forward2_bias_to_fp16, epsilon = var_4582_to_fp16, gamma = enc_layers_13_norm_feed_forward2_weight_to_fp16, x = input_349_cast_fp16)[name = string("x_237_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_13_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113892864))), lut = tensor<fp16, [256, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115465792))))[name = string("blocks_13_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_153_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_13_ff2_1_weight_to_fp16_palettized, x = x_237_cast_fp16)[name = string("linear_153_cast_fp16")];
            tensor<fp16, [4096]> blocks_13_ff2_1_scale_to_fp16 = const()[name = string("blocks_13_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115469952)))];
            tensor<fp16, [1, 188, 4096]> input_351_cast_fp16 = mul(x = linear_153_cast_fp16, y = blocks_13_ff2_1_scale_to_fp16)[name = string("input_351_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_239_cast_fp16 = silu(x = input_351_cast_fp16)[name = string("x_239_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_13_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115478208))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117051136))))[name = string("blocks_13_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_154_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_13_ff2_2_weight_to_fp16_palettized, x = x_239_cast_fp16)[name = string("linear_154_cast_fp16")];
            tensor<fp16, [1024]> blocks_13_ff2_2_scale_to_fp16 = const()[name = string("blocks_13_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117052224)))];
            tensor<fp16, [1, 188, 1024]> var_4716_cast_fp16 = mul(x = linear_154_cast_fp16, y = blocks_13_ff2_2_scale_to_fp16)[name = string("op_4716_cast_fp16")];
            fp16 var_4717_to_fp16 = const()[name = string("op_4717_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4718_cast_fp16 = mul(x = var_4716_cast_fp16, y = var_4717_to_fp16)[name = string("op_4718_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_353_cast_fp16 = add(x = input_349_cast_fp16, y = var_4718_cast_fp16)[name = string("input_353_cast_fp16")];
            tensor<int32, [1]> input_355_axes_0 = const()[name = string("input_355_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_13_norm_out_weight_to_fp16 = const()[name = string("enc_layers_13_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117054336)))];
            tensor<fp16, [1024]> enc_layers_13_norm_out_bias_to_fp16 = const()[name = string("enc_layers_13_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117056448)))];
            tensor<fp16, [1, 188, 1024]> input_355_cast_fp16 = layer_norm(axes = input_355_axes_0, beta = enc_layers_13_norm_out_bias_to_fp16, epsilon = var_4582_to_fp16, gamma = enc_layers_13_norm_out_weight_to_fp16, x = input_353_cast_fp16)[name = string("input_355_cast_fp16")];
            int32 var_4732 = const()[name = string("op_4732"), val = int32(-1)];
            tensor<int32, [1]> x_241_axes_0 = const()[name = string("x_241_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_14_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_14_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117058560)))];
            tensor<fp16, [1024]> enc_layers_14_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_14_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117060672)))];
            fp16 var_4739_to_fp16 = const()[name = string("op_4739_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_241_cast_fp16 = layer_norm(axes = x_241_axes_0, beta = enc_layers_14_norm_feed_forward1_bias_to_fp16, epsilon = var_4739_to_fp16, gamma = enc_layers_14_norm_feed_forward1_weight_to_fp16, x = input_355_cast_fp16)[name = string("x_241_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_14_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117062784))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119160000))))[name = string("blocks_14_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_155_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_14_ff1_1_weight_to_fp16_palettized, x = x_241_cast_fp16)[name = string("linear_155_cast_fp16")];
            tensor<fp16, [4096]> blocks_14_ff1_1_scale_to_fp16 = const()[name = string("blocks_14_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119168256)))];
            tensor<fp16, [1, 188, 4096]> input_357_cast_fp16 = mul(x = linear_155_cast_fp16, y = blocks_14_ff1_1_scale_to_fp16)[name = string("input_357_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_243_cast_fp16 = silu(x = input_357_cast_fp16)[name = string("x_243_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_14_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119176512))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121273728))))[name = string("blocks_14_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_156_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_14_ff1_2_weight_to_fp16_palettized, x = x_243_cast_fp16)[name = string("linear_156_cast_fp16")];
            tensor<fp16, [1024]> blocks_14_ff1_2_scale_to_fp16 = const()[name = string("blocks_14_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121275840)))];
            tensor<fp16, [1, 188, 1024]> var_4764_cast_fp16 = mul(x = linear_156_cast_fp16, y = blocks_14_ff1_2_scale_to_fp16)[name = string("op_4764_cast_fp16")];
            fp16 var_4765_to_fp16 = const()[name = string("op_4765_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4766_cast_fp16 = mul(x = var_4764_cast_fp16, y = var_4765_to_fp16)[name = string("op_4766_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_359_cast_fp16 = add(x = input_355_cast_fp16, y = var_4766_cast_fp16)[name = string("input_359_cast_fp16")];
            tensor<int32, [1]> x_245_axes_0 = const()[name = string("x_245_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_14_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_14_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121277952)))];
            tensor<fp16, [1024]> enc_layers_14_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_14_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121280064)))];
            tensor<fp16, [1, 188, 1024]> x_245_cast_fp16 = layer_norm(axes = x_245_axes_0, beta = enc_layers_14_norm_self_att_bias_to_fp16, epsilon = var_4739_to_fp16, gamma = enc_layers_14_norm_self_att_weight_to_fp16, x = input_359_cast_fp16)[name = string("x_245_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_14_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121282176))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121675456))))[name = string("blocks_14_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_157_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_14_q_weight_to_fp16_palettized, x = x_245_cast_fp16)[name = string("linear_157_cast_fp16")];
            tensor<fp16, [1024]> blocks_14_q_scale_to_fp16 = const()[name = string("blocks_14_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121676544)))];
            tensor<fp16, [1, 188, 1024]> var_4774_cast_fp16 = mul(x = linear_157_cast_fp16, y = blocks_14_q_scale_to_fp16)[name = string("op_4774_cast_fp16")];
            tensor<int32, [4]> var_4775 = const()[name = string("op_4775"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4776_cast_fp16 = reshape(shape = var_4775, x = var_4774_cast_fp16)[name = string("op_4776_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_14_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121678656))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122071936))))[name = string("blocks_14_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_158_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_14_k_weight_to_fp16_palettized, x = x_245_cast_fp16)[name = string("linear_158_cast_fp16")];
            tensor<fp16, [1024]> blocks_14_k_scale_to_fp16 = const()[name = string("blocks_14_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122073024)))];
            tensor<fp16, [1, 188, 1024]> var_4782_cast_fp16 = mul(x = linear_158_cast_fp16, y = blocks_14_k_scale_to_fp16)[name = string("op_4782_cast_fp16")];
            tensor<int32, [4]> var_4783 = const()[name = string("op_4783"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4784_cast_fp16 = reshape(shape = var_4783, x = var_4782_cast_fp16)[name = string("op_4784_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_14_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122075136))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122468416))))[name = string("blocks_14_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_159_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_14_v_weight_to_fp16_palettized, x = x_245_cast_fp16)[name = string("linear_159_cast_fp16")];
            tensor<fp16, [1024]> blocks_14_v_scale_to_fp16 = const()[name = string("blocks_14_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122469504)))];
            tensor<fp16, [1, 188, 1024]> var_4790_cast_fp16 = mul(x = linear_159_cast_fp16, y = blocks_14_v_scale_to_fp16)[name = string("op_4790_cast_fp16")];
            tensor<int32, [4]> var_4791 = const()[name = string("op_4791"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4792_cast_fp16 = reshape(shape = var_4791, x = var_4790_cast_fp16)[name = string("op_4792_cast_fp16")];
            tensor<int32, [4]> v_29_perm_0 = const()[name = string("v_29_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_112_to_fp16 = const()[name = string("const_112_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122471616)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_29_cast_fp16 = add(x = var_4776_cast_fp16, y = const_112_to_fp16)[name = string("q_u_29_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_113_to_fp16 = const()[name = string("const_113_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122473728)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_29_cast_fp16 = add(x = var_4776_cast_fp16, y = const_113_to_fp16)[name = string("q_v_29_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_14_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122475840))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122869120))))[name = string("blocks_14_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_160_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_14_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_160_cast_fp16")];
            tensor<fp16, [1024]> blocks_14_pos_scale_to_fp16 = const()[name = string("blocks_14_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122870208)))];
            tensor<fp16, [1, 375, 1024]> var_4804_cast_fp16 = mul(x = linear_160_cast_fp16, y = blocks_14_pos_scale_to_fp16)[name = string("op_4804_cast_fp16")];
            tensor<int32, [4]> var_4805 = const()[name = string("op_4805"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_29_cast_fp16 = reshape(shape = var_4805, x = var_4804_cast_fp16)[name = string("p_29_cast_fp16")];
            bool input_361_transpose_x_0 = const()[name = string("input_361_transpose_x_0"), val = bool(false)];
            bool input_361_transpose_y_0 = const()[name = string("input_361_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_248_perm_0 = const()[name = string("transpose_248_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_249_perm_0 = const()[name = string("transpose_249_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_249 = transpose(perm = transpose_249_perm_0, x = p_29_cast_fp16)[name = string("transpose_367")];
            tensor<fp16, [1, 8, 188, 128]> transpose_248 = transpose(perm = transpose_248_perm_0, x = q_v_29_cast_fp16)[name = string("transpose_368")];
            tensor<fp16, [1, 8, 188, 375]> input_361_cast_fp16 = matmul(transpose_x = input_361_transpose_x_0, transpose_y = input_361_transpose_y_0, x = transpose_248, y = transpose_249)[name = string("input_361_cast_fp16")];
            tensor<int32, [8]> var_4811_pad_0 = const()[name = string("op_4811_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_4811_mode_0 = const()[name = string("op_4811_mode_0"), val = string("constant")];
            fp16 const_26_to_fp16 = const()[name = string("const_26_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_4811_cast_fp16 = pad(constant_val = const_26_to_fp16, mode = var_4811_mode_0, pad = var_4811_pad_0, x = input_361_cast_fp16)[name = string("op_4811_cast_fp16")];
            tensor<int32, [4]> var_4812 = const()[name = string("op_4812"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_57_cast_fp16 = reshape(shape = var_4812, x = var_4811_cast_fp16)[name = string("scores_57_cast_fp16")];
            tensor<int32, [4]> var_4816_begin_0 = const()[name = string("op_4816_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_4816_end_0 = const()[name = string("op_4816_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_4816_end_mask_0 = const()[name = string("op_4816_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_4816_cast_fp16 = slice_by_index(begin = var_4816_begin_0, end = var_4816_end_0, end_mask = var_4816_end_mask_0, x = scores_57_cast_fp16)[name = string("op_4816_cast_fp16")];
            tensor<int32, [4]> var_4817 = const()[name = string("op_4817"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_4818_cast_fp16 = reshape(shape = var_4817, x = var_4816_cast_fp16)[name = string("op_4818_cast_fp16")];
            tensor<int32, [4]> bd_29_begin_0 = const()[name = string("bd_29_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_29_end_0 = const()[name = string("bd_29_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_29_end_mask_0 = const()[name = string("bd_29_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_29_cast_fp16 = slice_by_index(begin = bd_29_begin_0, end = bd_29_end_0, end_mask = bd_29_end_mask_0, x = var_4818_cast_fp16)[name = string("bd_29_cast_fp16")];
            bool var_4821_transpose_x_0 = const()[name = string("op_4821_transpose_x_0"), val = bool(false)];
            bool var_4821_transpose_y_0 = const()[name = string("op_4821_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_250_perm_0 = const()[name = string("transpose_250_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_251_perm_0 = const()[name = string("transpose_251_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_251 = transpose(perm = transpose_251_perm_0, x = var_4784_cast_fp16)[name = string("transpose_365")];
            tensor<fp16, [1, 8, 188, 128]> transpose_250 = transpose(perm = transpose_250_perm_0, x = q_u_29_cast_fp16)[name = string("transpose_366")];
            tensor<fp16, [1, 8, 188, 188]> var_4821_cast_fp16 = matmul(transpose_x = var_4821_transpose_x_0, transpose_y = var_4821_transpose_y_0, x = transpose_250, y = transpose_251)[name = string("op_4821_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4822_cast_fp16 = add(x = var_4821_cast_fp16, y = bd_29_cast_fp16)[name = string("op_4822_cast_fp16")];
            fp16 var_4823_to_fp16 = const()[name = string("op_4823_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_4824_cast_fp16 = mul(x = var_4822_cast_fp16, y = var_4823_to_fp16)[name = string("op_4824_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_59_cast_fp16 = add(x = var_4824_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_59_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4826_cast_fp16 = softmax(axis = var_4732, x = scores_59_cast_fp16)[name = string("op_4826_cast_fp16")];
            bool out_29_transpose_x_0 = const()[name = string("out_29_transpose_x_0"), val = bool(false)];
            bool out_29_transpose_y_0 = const()[name = string("out_29_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_29_cast_fp16 = transpose(perm = v_29_perm_0, x = var_4792_cast_fp16)[name = string("transpose_364")];
            tensor<fp16, [1, 8, 188, 128]> out_29_cast_fp16 = matmul(transpose_x = out_29_transpose_x_0, transpose_y = out_29_transpose_y_0, x = var_4826_cast_fp16, y = v_29_cast_fp16)[name = string("out_29_cast_fp16")];
            tensor<int32, [4]> var_4828_perm_0 = const()[name = string("op_4828_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4829 = const()[name = string("op_4829"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_4828_cast_fp16 = transpose(perm = var_4828_perm_0, x = out_29_cast_fp16)[name = string("transpose_363")];
            tensor<fp16, [1, 188, 1024]> x_247_cast_fp16 = reshape(shape = var_4829, x = var_4828_cast_fp16)[name = string("x_247_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_14_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122872320))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123265600))))[name = string("blocks_14_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_161_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_14_o_weight_to_fp16_palettized, x = x_247_cast_fp16)[name = string("linear_161_cast_fp16")];
            tensor<fp16, [1024]> blocks_14_o_scale_to_fp16 = const()[name = string("blocks_14_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123266688)))];
            tensor<fp16, [1, 188, 1024]> var_4835_cast_fp16 = mul(x = linear_161_cast_fp16, y = blocks_14_o_scale_to_fp16)[name = string("op_4835_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_363_cast_fp16 = add(x = input_359_cast_fp16, y = var_4835_cast_fp16)[name = string("input_363_cast_fp16")];
            tensor<int32, [1]> x_249_axes_0 = const()[name = string("x_249_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_14_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_14_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123268800)))];
            tensor<fp16, [1024]> enc_layers_14_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_14_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123270912)))];
            tensor<fp16, [1, 188, 1024]> x_249_cast_fp16 = layer_norm(axes = x_249_axes_0, beta = enc_layers_14_norm_conv_bias_to_fp16, epsilon = var_4739_to_fp16, gamma = enc_layers_14_norm_conv_weight_to_fp16, x = input_363_cast_fp16)[name = string("x_249_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_14_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123273024))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124321664))))[name = string("blocks_14_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_162_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_14_pw1_weight_to_fp16_palettized, x = x_249_cast_fp16)[name = string("linear_162_cast_fp16")];
            tensor<fp16, [2048]> blocks_14_pw1_scale_to_fp16 = const()[name = string("blocks_14_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124325824)))];
            tensor<fp16, [1, 188, 2048]> input_365_cast_fp16 = mul(x = linear_162_cast_fp16, y = blocks_14_pw1_scale_to_fp16)[name = string("input_365_cast_fp16")];
            int32 var_4844_split_num_splits_0 = const()[name = string("op_4844_split_num_splits_0"), val = int32(2)];
            int32 var_4844_split_axis_0 = const()[name = string("op_4844_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_4844_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_4844_split_cast_fp16_1 = split(axis = var_4844_split_axis_0, num_splits = var_4844_split_num_splits_0, x = input_365_cast_fp16)[name = string("op_4844_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_4844_split_1_sigmoid_cast_fp16 = sigmoid(x = var_4844_split_cast_fp16_1)[name = string("op_4844_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_4844_cast_fp16 = mul(x = var_4844_split_cast_fp16_0, y = var_4844_split_1_sigmoid_cast_fp16)[name = string("op_4844_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_57_cast_fp16 = mul(x = var_4844_cast_fp16, y = valid_cast_fp16)[name = string("h_57_cast_fp16")];
            tensor<int32, [3]> input_367_perm_0 = const()[name = string("input_367_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_369_pad_type_0 = const()[name = string("input_369_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_369_pad_0 = const()[name = string("input_369_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_369_groups_0 = const()[name = string("input_369_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_369_strides_0 = const()[name = string("input_369_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_369_dilations_0 = const()[name = string("input_369_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_64_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124329984))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124339264))))[name = string("const_64_to_fp16_quantized")];
            tensor<fp16, [1024]> const_65_to_fp16 = const()[name = string("const_65_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124341376)))];
            tensor<fp16, [1, 1024, 188]> input_367_cast_fp16 = transpose(perm = input_367_perm_0, x = h_57_cast_fp16)[name = string("transpose_362")];
            tensor<fp16, [1, 1024, 188]> input_371_cast_fp16 = conv(bias = const_65_to_fp16, dilations = input_369_dilations_0, groups = input_369_groups_0, pad = input_369_pad_0, pad_type = input_369_pad_type_0, strides = input_369_strides_0, weight = const_64_to_fp16_quantized, x = input_367_cast_fp16)[name = string("input_371_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_59_cast_fp16 = silu(x = input_371_cast_fp16)[name = string("h_59_cast_fp16")];
            tensor<int32, [3]> x_251_perm_0 = const()[name = string("x_251_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_14_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124343488))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124867840))))[name = string("blocks_14_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_251_cast_fp16 = transpose(perm = x_251_perm_0, x = h_59_cast_fp16)[name = string("transpose_361")];
            tensor<fp16, [1, 188, 1024]> linear_163_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_14_pw2_weight_to_fp16_palettized, x = x_251_cast_fp16)[name = string("linear_163_cast_fp16")];
            tensor<fp16, [1024]> blocks_14_pw2_scale_to_fp16 = const()[name = string("blocks_14_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124869952)))];
            tensor<fp16, [1, 188, 1024]> var_4859_cast_fp16 = mul(x = linear_163_cast_fp16, y = blocks_14_pw2_scale_to_fp16)[name = string("op_4859_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_373_cast_fp16 = add(x = input_363_cast_fp16, y = var_4859_cast_fp16)[name = string("input_373_cast_fp16")];
            tensor<int32, [1]> x_253_axes_0 = const()[name = string("x_253_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_14_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_14_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124872064)))];
            tensor<fp16, [1024]> enc_layers_14_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_14_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124874176)))];
            tensor<fp16, [1, 188, 1024]> x_253_cast_fp16 = layer_norm(axes = x_253_axes_0, beta = enc_layers_14_norm_feed_forward2_bias_to_fp16, epsilon = var_4739_to_fp16, gamma = enc_layers_14_norm_feed_forward2_weight_to_fp16, x = input_373_cast_fp16)[name = string("x_253_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_14_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124876288))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126973504))))[name = string("blocks_14_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_164_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_14_ff2_1_weight_to_fp16_palettized, x = x_253_cast_fp16)[name = string("linear_164_cast_fp16")];
            tensor<fp16, [4096]> blocks_14_ff2_1_scale_to_fp16 = const()[name = string("blocks_14_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126981760)))];
            tensor<fp16, [1, 188, 4096]> input_375_cast_fp16 = mul(x = linear_164_cast_fp16, y = blocks_14_ff2_1_scale_to_fp16)[name = string("input_375_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_255_cast_fp16 = silu(x = input_375_cast_fp16)[name = string("x_255_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_14_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126990016))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129087232))))[name = string("blocks_14_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_165_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_14_ff2_2_weight_to_fp16_palettized, x = x_255_cast_fp16)[name = string("linear_165_cast_fp16")];
            tensor<fp16, [1024]> blocks_14_ff2_2_scale_to_fp16 = const()[name = string("blocks_14_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129089344)))];
            tensor<fp16, [1, 188, 1024]> var_4873_cast_fp16 = mul(x = linear_165_cast_fp16, y = blocks_14_ff2_2_scale_to_fp16)[name = string("op_4873_cast_fp16")];
            fp16 var_4874_to_fp16 = const()[name = string("op_4874_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4875_cast_fp16 = mul(x = var_4873_cast_fp16, y = var_4874_to_fp16)[name = string("op_4875_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_377_cast_fp16 = add(x = input_373_cast_fp16, y = var_4875_cast_fp16)[name = string("input_377_cast_fp16")];
            tensor<int32, [1]> input_379_axes_0 = const()[name = string("input_379_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_14_norm_out_weight_to_fp16 = const()[name = string("enc_layers_14_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129091456)))];
            tensor<fp16, [1024]> enc_layers_14_norm_out_bias_to_fp16 = const()[name = string("enc_layers_14_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129093568)))];
            tensor<fp16, [1, 188, 1024]> input_379_cast_fp16 = layer_norm(axes = input_379_axes_0, beta = enc_layers_14_norm_out_bias_to_fp16, epsilon = var_4739_to_fp16, gamma = enc_layers_14_norm_out_weight_to_fp16, x = input_377_cast_fp16)[name = string("input_379_cast_fp16")];
            int32 var_4889 = const()[name = string("op_4889"), val = int32(-1)];
            tensor<int32, [1]> x_257_axes_0 = const()[name = string("x_257_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_15_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_15_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129095680)))];
            tensor<fp16, [1024]> enc_layers_15_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_15_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129097792)))];
            fp16 var_4896_to_fp16 = const()[name = string("op_4896_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_257_cast_fp16 = layer_norm(axes = x_257_axes_0, beta = enc_layers_15_norm_feed_forward1_bias_to_fp16, epsilon = var_4896_to_fp16, gamma = enc_layers_15_norm_feed_forward1_weight_to_fp16, x = input_379_cast_fp16)[name = string("x_257_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_15_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129099904))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131197120))))[name = string("blocks_15_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_166_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_15_ff1_1_weight_to_fp16_palettized, x = x_257_cast_fp16)[name = string("linear_166_cast_fp16")];
            tensor<fp16, [4096]> blocks_15_ff1_1_scale_to_fp16 = const()[name = string("blocks_15_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131205376)))];
            tensor<fp16, [1, 188, 4096]> input_381_cast_fp16 = mul(x = linear_166_cast_fp16, y = blocks_15_ff1_1_scale_to_fp16)[name = string("input_381_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_259_cast_fp16 = silu(x = input_381_cast_fp16)[name = string("x_259_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_15_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131213632))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133310848))))[name = string("blocks_15_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_167_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_15_ff1_2_weight_to_fp16_palettized, x = x_259_cast_fp16)[name = string("linear_167_cast_fp16")];
            tensor<fp16, [1024]> blocks_15_ff1_2_scale_to_fp16 = const()[name = string("blocks_15_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133312960)))];
            tensor<fp16, [1, 188, 1024]> var_4921_cast_fp16 = mul(x = linear_167_cast_fp16, y = blocks_15_ff1_2_scale_to_fp16)[name = string("op_4921_cast_fp16")];
            fp16 var_4922_to_fp16 = const()[name = string("op_4922_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4923_cast_fp16 = mul(x = var_4921_cast_fp16, y = var_4922_to_fp16)[name = string("op_4923_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_383_cast_fp16 = add(x = input_379_cast_fp16, y = var_4923_cast_fp16)[name = string("input_383_cast_fp16")];
            tensor<int32, [1]> x_261_axes_0 = const()[name = string("x_261_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_15_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_15_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133315072)))];
            tensor<fp16, [1024]> enc_layers_15_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_15_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133317184)))];
            tensor<fp16, [1, 188, 1024]> x_261_cast_fp16 = layer_norm(axes = x_261_axes_0, beta = enc_layers_15_norm_self_att_bias_to_fp16, epsilon = var_4896_to_fp16, gamma = enc_layers_15_norm_self_att_weight_to_fp16, x = input_383_cast_fp16)[name = string("x_261_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_15_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133319296))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133581504))))[name = string("blocks_15_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_168_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_15_q_weight_to_fp16_palettized, x = x_261_cast_fp16)[name = string("linear_168_cast_fp16")];
            tensor<fp16, [1024]> blocks_15_q_scale_to_fp16 = const()[name = string("blocks_15_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133582080)))];
            tensor<fp16, [1, 188, 1024]> var_4931_cast_fp16 = mul(x = linear_168_cast_fp16, y = blocks_15_q_scale_to_fp16)[name = string("op_4931_cast_fp16")];
            tensor<int32, [4]> var_4932 = const()[name = string("op_4932"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4933_cast_fp16 = reshape(shape = var_4932, x = var_4931_cast_fp16)[name = string("op_4933_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_15_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133584192))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133846400))))[name = string("blocks_15_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_169_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_15_k_weight_to_fp16_palettized, x = x_261_cast_fp16)[name = string("linear_169_cast_fp16")];
            tensor<fp16, [1024]> blocks_15_k_scale_to_fp16 = const()[name = string("blocks_15_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133846976)))];
            tensor<fp16, [1, 188, 1024]> var_4939_cast_fp16 = mul(x = linear_169_cast_fp16, y = blocks_15_k_scale_to_fp16)[name = string("op_4939_cast_fp16")];
            tensor<int32, [4]> var_4940 = const()[name = string("op_4940"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4941_cast_fp16 = reshape(shape = var_4940, x = var_4939_cast_fp16)[name = string("op_4941_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_15_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133849088))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134111296))))[name = string("blocks_15_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_170_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_15_v_weight_to_fp16_palettized, x = x_261_cast_fp16)[name = string("linear_170_cast_fp16")];
            tensor<fp16, [1024]> blocks_15_v_scale_to_fp16 = const()[name = string("blocks_15_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134111872)))];
            tensor<fp16, [1, 188, 1024]> var_4947_cast_fp16 = mul(x = linear_170_cast_fp16, y = blocks_15_v_scale_to_fp16)[name = string("op_4947_cast_fp16")];
            tensor<int32, [4]> var_4948 = const()[name = string("op_4948"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4949_cast_fp16 = reshape(shape = var_4948, x = var_4947_cast_fp16)[name = string("op_4949_cast_fp16")];
            tensor<int32, [4]> v_31_perm_0 = const()[name = string("v_31_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_114_to_fp16 = const()[name = string("const_114_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134113984)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_31_cast_fp16 = add(x = var_4933_cast_fp16, y = const_114_to_fp16)[name = string("q_u_31_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_115_to_fp16 = const()[name = string("const_115_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134116096)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_31_cast_fp16 = add(x = var_4933_cast_fp16, y = const_115_to_fp16)[name = string("q_v_31_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_15_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134118208))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134380416))))[name = string("blocks_15_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_171_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_15_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_171_cast_fp16")];
            tensor<fp16, [1024]> blocks_15_pos_scale_to_fp16 = const()[name = string("blocks_15_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134380992)))];
            tensor<fp16, [1, 375, 1024]> var_4961_cast_fp16 = mul(x = linear_171_cast_fp16, y = blocks_15_pos_scale_to_fp16)[name = string("op_4961_cast_fp16")];
            tensor<int32, [4]> var_4962 = const()[name = string("op_4962"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_31_cast_fp16 = reshape(shape = var_4962, x = var_4961_cast_fp16)[name = string("p_31_cast_fp16")];
            bool input_385_transpose_x_0 = const()[name = string("input_385_transpose_x_0"), val = bool(false)];
            bool input_385_transpose_y_0 = const()[name = string("input_385_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_252_perm_0 = const()[name = string("transpose_252_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_253_perm_0 = const()[name = string("transpose_253_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_253 = transpose(perm = transpose_253_perm_0, x = p_31_cast_fp16)[name = string("transpose_359")];
            tensor<fp16, [1, 8, 188, 128]> transpose_252 = transpose(perm = transpose_252_perm_0, x = q_v_31_cast_fp16)[name = string("transpose_360")];
            tensor<fp16, [1, 8, 188, 375]> input_385_cast_fp16 = matmul(transpose_x = input_385_transpose_x_0, transpose_y = input_385_transpose_y_0, x = transpose_252, y = transpose_253)[name = string("input_385_cast_fp16")];
            tensor<int32, [8]> var_4968_pad_0 = const()[name = string("op_4968_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_4968_mode_0 = const()[name = string("op_4968_mode_0"), val = string("constant")];
            fp16 const_27_to_fp16 = const()[name = string("const_27_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_4968_cast_fp16 = pad(constant_val = const_27_to_fp16, mode = var_4968_mode_0, pad = var_4968_pad_0, x = input_385_cast_fp16)[name = string("op_4968_cast_fp16")];
            tensor<int32, [4]> var_4969 = const()[name = string("op_4969"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_61_cast_fp16 = reshape(shape = var_4969, x = var_4968_cast_fp16)[name = string("scores_61_cast_fp16")];
            tensor<int32, [4]> var_4973_begin_0 = const()[name = string("op_4973_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_4973_end_0 = const()[name = string("op_4973_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_4973_end_mask_0 = const()[name = string("op_4973_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_4973_cast_fp16 = slice_by_index(begin = var_4973_begin_0, end = var_4973_end_0, end_mask = var_4973_end_mask_0, x = scores_61_cast_fp16)[name = string("op_4973_cast_fp16")];
            tensor<int32, [4]> var_4974 = const()[name = string("op_4974"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_4975_cast_fp16 = reshape(shape = var_4974, x = var_4973_cast_fp16)[name = string("op_4975_cast_fp16")];
            tensor<int32, [4]> bd_31_begin_0 = const()[name = string("bd_31_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_31_end_0 = const()[name = string("bd_31_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_31_end_mask_0 = const()[name = string("bd_31_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_31_cast_fp16 = slice_by_index(begin = bd_31_begin_0, end = bd_31_end_0, end_mask = bd_31_end_mask_0, x = var_4975_cast_fp16)[name = string("bd_31_cast_fp16")];
            bool var_4978_transpose_x_0 = const()[name = string("op_4978_transpose_x_0"), val = bool(false)];
            bool var_4978_transpose_y_0 = const()[name = string("op_4978_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_254_perm_0 = const()[name = string("transpose_254_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_255_perm_0 = const()[name = string("transpose_255_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_255 = transpose(perm = transpose_255_perm_0, x = var_4941_cast_fp16)[name = string("transpose_357")];
            tensor<fp16, [1, 8, 188, 128]> transpose_254 = transpose(perm = transpose_254_perm_0, x = q_u_31_cast_fp16)[name = string("transpose_358")];
            tensor<fp16, [1, 8, 188, 188]> var_4978_cast_fp16 = matmul(transpose_x = var_4978_transpose_x_0, transpose_y = var_4978_transpose_y_0, x = transpose_254, y = transpose_255)[name = string("op_4978_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4979_cast_fp16 = add(x = var_4978_cast_fp16, y = bd_31_cast_fp16)[name = string("op_4979_cast_fp16")];
            fp16 var_4980_to_fp16 = const()[name = string("op_4980_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_4981_cast_fp16 = mul(x = var_4979_cast_fp16, y = var_4980_to_fp16)[name = string("op_4981_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_63_cast_fp16 = add(x = var_4981_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_63_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4983_cast_fp16 = softmax(axis = var_4889, x = scores_63_cast_fp16)[name = string("op_4983_cast_fp16")];
            bool out_31_transpose_x_0 = const()[name = string("out_31_transpose_x_0"), val = bool(false)];
            bool out_31_transpose_y_0 = const()[name = string("out_31_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_31_cast_fp16 = transpose(perm = v_31_perm_0, x = var_4949_cast_fp16)[name = string("transpose_356")];
            tensor<fp16, [1, 8, 188, 128]> out_31_cast_fp16 = matmul(transpose_x = out_31_transpose_x_0, transpose_y = out_31_transpose_y_0, x = var_4983_cast_fp16, y = v_31_cast_fp16)[name = string("out_31_cast_fp16")];
            tensor<int32, [4]> var_4985_perm_0 = const()[name = string("op_4985_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4986 = const()[name = string("op_4986"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_4985_cast_fp16 = transpose(perm = var_4985_perm_0, x = out_31_cast_fp16)[name = string("transpose_355")];
            tensor<fp16, [1, 188, 1024]> x_263_cast_fp16 = reshape(shape = var_4986, x = var_4985_cast_fp16)[name = string("x_263_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_15_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134383104))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134645312))))[name = string("blocks_15_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_172_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_15_o_weight_to_fp16_palettized, x = x_263_cast_fp16)[name = string("linear_172_cast_fp16")];
            tensor<fp16, [1024]> blocks_15_o_scale_to_fp16 = const()[name = string("blocks_15_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134645888)))];
            tensor<fp16, [1, 188, 1024]> var_4992_cast_fp16 = mul(x = linear_172_cast_fp16, y = blocks_15_o_scale_to_fp16)[name = string("op_4992_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_387_cast_fp16 = add(x = input_383_cast_fp16, y = var_4992_cast_fp16)[name = string("input_387_cast_fp16")];
            tensor<int32, [1]> x_265_axes_0 = const()[name = string("x_265_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_15_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_15_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134648000)))];
            tensor<fp16, [1024]> enc_layers_15_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_15_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134650112)))];
            tensor<fp16, [1, 188, 1024]> x_265_cast_fp16 = layer_norm(axes = x_265_axes_0, beta = enc_layers_15_norm_conv_bias_to_fp16, epsilon = var_4896_to_fp16, gamma = enc_layers_15_norm_conv_weight_to_fp16, x = input_387_cast_fp16)[name = string("x_265_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_15_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134652224))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135700864))))[name = string("blocks_15_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_173_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_15_pw1_weight_to_fp16_palettized, x = x_265_cast_fp16)[name = string("linear_173_cast_fp16")];
            tensor<fp16, [2048]> blocks_15_pw1_scale_to_fp16 = const()[name = string("blocks_15_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135705024)))];
            tensor<fp16, [1, 188, 2048]> input_389_cast_fp16 = mul(x = linear_173_cast_fp16, y = blocks_15_pw1_scale_to_fp16)[name = string("input_389_cast_fp16")];
            int32 var_5001_split_num_splits_0 = const()[name = string("op_5001_split_num_splits_0"), val = int32(2)];
            int32 var_5001_split_axis_0 = const()[name = string("op_5001_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_5001_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_5001_split_cast_fp16_1 = split(axis = var_5001_split_axis_0, num_splits = var_5001_split_num_splits_0, x = input_389_cast_fp16)[name = string("op_5001_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_5001_split_1_sigmoid_cast_fp16 = sigmoid(x = var_5001_split_cast_fp16_1)[name = string("op_5001_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_5001_cast_fp16 = mul(x = var_5001_split_cast_fp16_0, y = var_5001_split_1_sigmoid_cast_fp16)[name = string("op_5001_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_61_cast_fp16 = mul(x = var_5001_cast_fp16, y = valid_cast_fp16)[name = string("h_61_cast_fp16")];
            tensor<int32, [3]> input_391_perm_0 = const()[name = string("input_391_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_393_pad_type_0 = const()[name = string("input_393_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_393_pad_0 = const()[name = string("input_393_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_393_groups_0 = const()[name = string("input_393_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_393_strides_0 = const()[name = string("input_393_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_393_dilations_0 = const()[name = string("input_393_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_66_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135709184))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135718464))))[name = string("const_66_to_fp16_quantized")];
            tensor<fp16, [1024]> const_67_to_fp16 = const()[name = string("const_67_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135720576)))];
            tensor<fp16, [1, 1024, 188]> input_391_cast_fp16 = transpose(perm = input_391_perm_0, x = h_61_cast_fp16)[name = string("transpose_354")];
            tensor<fp16, [1, 1024, 188]> input_395_cast_fp16 = conv(bias = const_67_to_fp16, dilations = input_393_dilations_0, groups = input_393_groups_0, pad = input_393_pad_0, pad_type = input_393_pad_type_0, strides = input_393_strides_0, weight = const_66_to_fp16_quantized, x = input_391_cast_fp16)[name = string("input_395_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_63_cast_fp16 = silu(x = input_395_cast_fp16)[name = string("h_63_cast_fp16")];
            tensor<int32, [3]> x_267_perm_0 = const()[name = string("x_267_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_15_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135722688))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136247040))))[name = string("blocks_15_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_267_cast_fp16 = transpose(perm = x_267_perm_0, x = h_63_cast_fp16)[name = string("transpose_353")];
            tensor<fp16, [1, 188, 1024]> linear_174_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_15_pw2_weight_to_fp16_palettized, x = x_267_cast_fp16)[name = string("linear_174_cast_fp16")];
            tensor<fp16, [1024]> blocks_15_pw2_scale_to_fp16 = const()[name = string("blocks_15_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136249152)))];
            tensor<fp16, [1, 188, 1024]> var_5016_cast_fp16 = mul(x = linear_174_cast_fp16, y = blocks_15_pw2_scale_to_fp16)[name = string("op_5016_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_397_cast_fp16 = add(x = input_387_cast_fp16, y = var_5016_cast_fp16)[name = string("input_397_cast_fp16")];
            tensor<int32, [1]> x_269_axes_0 = const()[name = string("x_269_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_15_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_15_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136251264)))];
            tensor<fp16, [1024]> enc_layers_15_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_15_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136253376)))];
            tensor<fp16, [1, 188, 1024]> x_269_cast_fp16 = layer_norm(axes = x_269_axes_0, beta = enc_layers_15_norm_feed_forward2_bias_to_fp16, epsilon = var_4896_to_fp16, gamma = enc_layers_15_norm_feed_forward2_weight_to_fp16, x = input_397_cast_fp16)[name = string("x_269_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_15_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136255488))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138352704))))[name = string("blocks_15_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_175_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_15_ff2_1_weight_to_fp16_palettized, x = x_269_cast_fp16)[name = string("linear_175_cast_fp16")];
            tensor<fp16, [4096]> blocks_15_ff2_1_scale_to_fp16 = const()[name = string("blocks_15_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138360960)))];
            tensor<fp16, [1, 188, 4096]> input_399_cast_fp16 = mul(x = linear_175_cast_fp16, y = blocks_15_ff2_1_scale_to_fp16)[name = string("input_399_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_271_cast_fp16 = silu(x = input_399_cast_fp16)[name = string("x_271_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_15_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138369216))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140466432))))[name = string("blocks_15_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_176_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_15_ff2_2_weight_to_fp16_palettized, x = x_271_cast_fp16)[name = string("linear_176_cast_fp16")];
            tensor<fp16, [1024]> blocks_15_ff2_2_scale_to_fp16 = const()[name = string("blocks_15_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140468544)))];
            tensor<fp16, [1, 188, 1024]> var_5030_cast_fp16 = mul(x = linear_176_cast_fp16, y = blocks_15_ff2_2_scale_to_fp16)[name = string("op_5030_cast_fp16")];
            fp16 var_5031_to_fp16 = const()[name = string("op_5031_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5032_cast_fp16 = mul(x = var_5030_cast_fp16, y = var_5031_to_fp16)[name = string("op_5032_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_401_cast_fp16 = add(x = input_397_cast_fp16, y = var_5032_cast_fp16)[name = string("input_401_cast_fp16")];
            tensor<int32, [1]> input_403_axes_0 = const()[name = string("input_403_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_15_norm_out_weight_to_fp16 = const()[name = string("enc_layers_15_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140470656)))];
            tensor<fp16, [1024]> enc_layers_15_norm_out_bias_to_fp16 = const()[name = string("enc_layers_15_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140472768)))];
            tensor<fp16, [1, 188, 1024]> input_403_cast_fp16 = layer_norm(axes = input_403_axes_0, beta = enc_layers_15_norm_out_bias_to_fp16, epsilon = var_4896_to_fp16, gamma = enc_layers_15_norm_out_weight_to_fp16, x = input_401_cast_fp16)[name = string("input_403_cast_fp16")];
            int32 var_5046 = const()[name = string("op_5046"), val = int32(-1)];
            tensor<int32, [1]> x_273_axes_0 = const()[name = string("x_273_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_16_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_16_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140474880)))];
            tensor<fp16, [1024]> enc_layers_16_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_16_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140476992)))];
            fp16 var_5053_to_fp16 = const()[name = string("op_5053_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_273_cast_fp16 = layer_norm(axes = x_273_axes_0, beta = enc_layers_16_norm_feed_forward1_bias_to_fp16, epsilon = var_5053_to_fp16, gamma = enc_layers_16_norm_feed_forward1_weight_to_fp16, x = input_403_cast_fp16)[name = string("x_273_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_16_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140479104))), lut = tensor<fp16, [256, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142052032))))[name = string("blocks_16_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_177_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_16_ff1_1_weight_to_fp16_palettized, x = x_273_cast_fp16)[name = string("linear_177_cast_fp16")];
            tensor<fp16, [4096]> blocks_16_ff1_1_scale_to_fp16 = const()[name = string("blocks_16_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142056192)))];
            tensor<fp16, [1, 188, 4096]> input_405_cast_fp16 = mul(x = linear_177_cast_fp16, y = blocks_16_ff1_1_scale_to_fp16)[name = string("input_405_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_275_cast_fp16 = silu(x = input_405_cast_fp16)[name = string("x_275_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_16_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142064448))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143637376))))[name = string("blocks_16_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_178_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_16_ff1_2_weight_to_fp16_palettized, x = x_275_cast_fp16)[name = string("linear_178_cast_fp16")];
            tensor<fp16, [1024]> blocks_16_ff1_2_scale_to_fp16 = const()[name = string("blocks_16_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143638464)))];
            tensor<fp16, [1, 188, 1024]> var_5078_cast_fp16 = mul(x = linear_178_cast_fp16, y = blocks_16_ff1_2_scale_to_fp16)[name = string("op_5078_cast_fp16")];
            fp16 var_5079_to_fp16 = const()[name = string("op_5079_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5080_cast_fp16 = mul(x = var_5078_cast_fp16, y = var_5079_to_fp16)[name = string("op_5080_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_407_cast_fp16 = add(x = input_403_cast_fp16, y = var_5080_cast_fp16)[name = string("input_407_cast_fp16")];
            tensor<int32, [1]> x_277_axes_0 = const()[name = string("x_277_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_16_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_16_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143640576)))];
            tensor<fp16, [1024]> enc_layers_16_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_16_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143642688)))];
            tensor<fp16, [1, 188, 1024]> x_277_cast_fp16 = layer_norm(axes = x_277_axes_0, beta = enc_layers_16_norm_self_att_bias_to_fp16, epsilon = var_5053_to_fp16, gamma = enc_layers_16_norm_self_att_weight_to_fp16, x = input_407_cast_fp16)[name = string("x_277_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_16_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143644800))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143907008))))[name = string("blocks_16_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_179_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_16_q_weight_to_fp16_palettized, x = x_277_cast_fp16)[name = string("linear_179_cast_fp16")];
            tensor<fp16, [1024]> blocks_16_q_scale_to_fp16 = const()[name = string("blocks_16_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143907584)))];
            tensor<fp16, [1, 188, 1024]> var_5088_cast_fp16 = mul(x = linear_179_cast_fp16, y = blocks_16_q_scale_to_fp16)[name = string("op_5088_cast_fp16")];
            tensor<int32, [4]> var_5089 = const()[name = string("op_5089"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5090_cast_fp16 = reshape(shape = var_5089, x = var_5088_cast_fp16)[name = string("op_5090_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_16_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143909696))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144171904))))[name = string("blocks_16_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_180_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_16_k_weight_to_fp16_palettized, x = x_277_cast_fp16)[name = string("linear_180_cast_fp16")];
            tensor<fp16, [1024]> blocks_16_k_scale_to_fp16 = const()[name = string("blocks_16_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144172480)))];
            tensor<fp16, [1, 188, 1024]> var_5096_cast_fp16 = mul(x = linear_180_cast_fp16, y = blocks_16_k_scale_to_fp16)[name = string("op_5096_cast_fp16")];
            tensor<int32, [4]> var_5097 = const()[name = string("op_5097"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5098_cast_fp16 = reshape(shape = var_5097, x = var_5096_cast_fp16)[name = string("op_5098_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_16_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144174592))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144436800))))[name = string("blocks_16_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_181_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_16_v_weight_to_fp16_palettized, x = x_277_cast_fp16)[name = string("linear_181_cast_fp16")];
            tensor<fp16, [1024]> blocks_16_v_scale_to_fp16 = const()[name = string("blocks_16_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144437376)))];
            tensor<fp16, [1, 188, 1024]> var_5104_cast_fp16 = mul(x = linear_181_cast_fp16, y = blocks_16_v_scale_to_fp16)[name = string("op_5104_cast_fp16")];
            tensor<int32, [4]> var_5105 = const()[name = string("op_5105"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5106_cast_fp16 = reshape(shape = var_5105, x = var_5104_cast_fp16)[name = string("op_5106_cast_fp16")];
            tensor<int32, [4]> v_33_perm_0 = const()[name = string("v_33_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_116_to_fp16 = const()[name = string("const_116_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144439488)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_33_cast_fp16 = add(x = var_5090_cast_fp16, y = const_116_to_fp16)[name = string("q_u_33_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_117_to_fp16 = const()[name = string("const_117_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144441600)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_33_cast_fp16 = add(x = var_5090_cast_fp16, y = const_117_to_fp16)[name = string("q_v_33_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_16_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144443712))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144705920))))[name = string("blocks_16_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_182_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_16_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_182_cast_fp16")];
            tensor<fp16, [1024]> blocks_16_pos_scale_to_fp16 = const()[name = string("blocks_16_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144706496)))];
            tensor<fp16, [1, 375, 1024]> var_5118_cast_fp16 = mul(x = linear_182_cast_fp16, y = blocks_16_pos_scale_to_fp16)[name = string("op_5118_cast_fp16")];
            tensor<int32, [4]> var_5119 = const()[name = string("op_5119"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_33_cast_fp16 = reshape(shape = var_5119, x = var_5118_cast_fp16)[name = string("p_33_cast_fp16")];
            bool input_409_transpose_x_0 = const()[name = string("input_409_transpose_x_0"), val = bool(false)];
            bool input_409_transpose_y_0 = const()[name = string("input_409_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_256_perm_0 = const()[name = string("transpose_256_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_257_perm_0 = const()[name = string("transpose_257_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_257 = transpose(perm = transpose_257_perm_0, x = p_33_cast_fp16)[name = string("transpose_351")];
            tensor<fp16, [1, 8, 188, 128]> transpose_256 = transpose(perm = transpose_256_perm_0, x = q_v_33_cast_fp16)[name = string("transpose_352")];
            tensor<fp16, [1, 8, 188, 375]> input_409_cast_fp16 = matmul(transpose_x = input_409_transpose_x_0, transpose_y = input_409_transpose_y_0, x = transpose_256, y = transpose_257)[name = string("input_409_cast_fp16")];
            tensor<int32, [8]> var_5125_pad_0 = const()[name = string("op_5125_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_5125_mode_0 = const()[name = string("op_5125_mode_0"), val = string("constant")];
            fp16 const_28_to_fp16 = const()[name = string("const_28_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_5125_cast_fp16 = pad(constant_val = const_28_to_fp16, mode = var_5125_mode_0, pad = var_5125_pad_0, x = input_409_cast_fp16)[name = string("op_5125_cast_fp16")];
            tensor<int32, [4]> var_5126 = const()[name = string("op_5126"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_65_cast_fp16 = reshape(shape = var_5126, x = var_5125_cast_fp16)[name = string("scores_65_cast_fp16")];
            tensor<int32, [4]> var_5130_begin_0 = const()[name = string("op_5130_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_5130_end_0 = const()[name = string("op_5130_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_5130_end_mask_0 = const()[name = string("op_5130_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_5130_cast_fp16 = slice_by_index(begin = var_5130_begin_0, end = var_5130_end_0, end_mask = var_5130_end_mask_0, x = scores_65_cast_fp16)[name = string("op_5130_cast_fp16")];
            tensor<int32, [4]> var_5131 = const()[name = string("op_5131"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_5132_cast_fp16 = reshape(shape = var_5131, x = var_5130_cast_fp16)[name = string("op_5132_cast_fp16")];
            tensor<int32, [4]> bd_33_begin_0 = const()[name = string("bd_33_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_33_end_0 = const()[name = string("bd_33_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_33_end_mask_0 = const()[name = string("bd_33_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_33_cast_fp16 = slice_by_index(begin = bd_33_begin_0, end = bd_33_end_0, end_mask = bd_33_end_mask_0, x = var_5132_cast_fp16)[name = string("bd_33_cast_fp16")];
            bool var_5135_transpose_x_0 = const()[name = string("op_5135_transpose_x_0"), val = bool(false)];
            bool var_5135_transpose_y_0 = const()[name = string("op_5135_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_258_perm_0 = const()[name = string("transpose_258_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_259_perm_0 = const()[name = string("transpose_259_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_259 = transpose(perm = transpose_259_perm_0, x = var_5098_cast_fp16)[name = string("transpose_349")];
            tensor<fp16, [1, 8, 188, 128]> transpose_258 = transpose(perm = transpose_258_perm_0, x = q_u_33_cast_fp16)[name = string("transpose_350")];
            tensor<fp16, [1, 8, 188, 188]> var_5135_cast_fp16 = matmul(transpose_x = var_5135_transpose_x_0, transpose_y = var_5135_transpose_y_0, x = transpose_258, y = transpose_259)[name = string("op_5135_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5136_cast_fp16 = add(x = var_5135_cast_fp16, y = bd_33_cast_fp16)[name = string("op_5136_cast_fp16")];
            fp16 var_5137_to_fp16 = const()[name = string("op_5137_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_5138_cast_fp16 = mul(x = var_5136_cast_fp16, y = var_5137_to_fp16)[name = string("op_5138_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_67_cast_fp16 = add(x = var_5138_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_67_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5140_cast_fp16 = softmax(axis = var_5046, x = scores_67_cast_fp16)[name = string("op_5140_cast_fp16")];
            bool out_33_transpose_x_0 = const()[name = string("out_33_transpose_x_0"), val = bool(false)];
            bool out_33_transpose_y_0 = const()[name = string("out_33_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_33_cast_fp16 = transpose(perm = v_33_perm_0, x = var_5106_cast_fp16)[name = string("transpose_348")];
            tensor<fp16, [1, 8, 188, 128]> out_33_cast_fp16 = matmul(transpose_x = out_33_transpose_x_0, transpose_y = out_33_transpose_y_0, x = var_5140_cast_fp16, y = v_33_cast_fp16)[name = string("out_33_cast_fp16")];
            tensor<int32, [4]> var_5142_perm_0 = const()[name = string("op_5142_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5143 = const()[name = string("op_5143"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_5142_cast_fp16 = transpose(perm = var_5142_perm_0, x = out_33_cast_fp16)[name = string("transpose_347")];
            tensor<fp16, [1, 188, 1024]> x_279_cast_fp16 = reshape(shape = var_5143, x = var_5142_cast_fp16)[name = string("x_279_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_16_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144708608))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144970816))))[name = string("blocks_16_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_183_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_16_o_weight_to_fp16_palettized, x = x_279_cast_fp16)[name = string("linear_183_cast_fp16")];
            tensor<fp16, [1024]> blocks_16_o_scale_to_fp16 = const()[name = string("blocks_16_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144971392)))];
            tensor<fp16, [1, 188, 1024]> var_5149_cast_fp16 = mul(x = linear_183_cast_fp16, y = blocks_16_o_scale_to_fp16)[name = string("op_5149_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_411_cast_fp16 = add(x = input_407_cast_fp16, y = var_5149_cast_fp16)[name = string("input_411_cast_fp16")];
            tensor<int32, [1]> x_281_axes_0 = const()[name = string("x_281_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_16_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_16_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144973504)))];
            tensor<fp16, [1024]> enc_layers_16_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_16_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144975616)))];
            tensor<fp16, [1, 188, 1024]> x_281_cast_fp16 = layer_norm(axes = x_281_axes_0, beta = enc_layers_16_norm_conv_bias_to_fp16, epsilon = var_5053_to_fp16, gamma = enc_layers_16_norm_conv_weight_to_fp16, x = input_411_cast_fp16)[name = string("x_281_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_16_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144977728))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146026368))))[name = string("blocks_16_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_184_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_16_pw1_weight_to_fp16_palettized, x = x_281_cast_fp16)[name = string("linear_184_cast_fp16")];
            tensor<fp16, [2048]> blocks_16_pw1_scale_to_fp16 = const()[name = string("blocks_16_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146030528)))];
            tensor<fp16, [1, 188, 2048]> input_413_cast_fp16 = mul(x = linear_184_cast_fp16, y = blocks_16_pw1_scale_to_fp16)[name = string("input_413_cast_fp16")];
            int32 var_5158_split_num_splits_0 = const()[name = string("op_5158_split_num_splits_0"), val = int32(2)];
            int32 var_5158_split_axis_0 = const()[name = string("op_5158_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_5158_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_5158_split_cast_fp16_1 = split(axis = var_5158_split_axis_0, num_splits = var_5158_split_num_splits_0, x = input_413_cast_fp16)[name = string("op_5158_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_5158_split_1_sigmoid_cast_fp16 = sigmoid(x = var_5158_split_cast_fp16_1)[name = string("op_5158_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_5158_cast_fp16 = mul(x = var_5158_split_cast_fp16_0, y = var_5158_split_1_sigmoid_cast_fp16)[name = string("op_5158_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_65_cast_fp16 = mul(x = var_5158_cast_fp16, y = valid_cast_fp16)[name = string("h_65_cast_fp16")];
            tensor<int32, [3]> input_415_perm_0 = const()[name = string("input_415_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_417_pad_type_0 = const()[name = string("input_417_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_417_pad_0 = const()[name = string("input_417_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_417_groups_0 = const()[name = string("input_417_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_417_strides_0 = const()[name = string("input_417_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_417_dilations_0 = const()[name = string("input_417_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_68_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146034688))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146043968))))[name = string("const_68_to_fp16_quantized")];
            tensor<fp16, [1024]> const_69_to_fp16 = const()[name = string("const_69_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146046080)))];
            tensor<fp16, [1, 1024, 188]> input_415_cast_fp16 = transpose(perm = input_415_perm_0, x = h_65_cast_fp16)[name = string("transpose_346")];
            tensor<fp16, [1, 1024, 188]> input_419_cast_fp16 = conv(bias = const_69_to_fp16, dilations = input_417_dilations_0, groups = input_417_groups_0, pad = input_417_pad_0, pad_type = input_417_pad_type_0, strides = input_417_strides_0, weight = const_68_to_fp16_quantized, x = input_415_cast_fp16)[name = string("input_419_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_67_cast_fp16 = silu(x = input_419_cast_fp16)[name = string("h_67_cast_fp16")];
            tensor<int32, [3]> x_283_perm_0 = const()[name = string("x_283_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_16_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146048192))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146572544))))[name = string("blocks_16_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_283_cast_fp16 = transpose(perm = x_283_perm_0, x = h_67_cast_fp16)[name = string("transpose_345")];
            tensor<fp16, [1, 188, 1024]> linear_185_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_16_pw2_weight_to_fp16_palettized, x = x_283_cast_fp16)[name = string("linear_185_cast_fp16")];
            tensor<fp16, [1024]> blocks_16_pw2_scale_to_fp16 = const()[name = string("blocks_16_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146574656)))];
            tensor<fp16, [1, 188, 1024]> var_5173_cast_fp16 = mul(x = linear_185_cast_fp16, y = blocks_16_pw2_scale_to_fp16)[name = string("op_5173_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_421_cast_fp16 = add(x = input_411_cast_fp16, y = var_5173_cast_fp16)[name = string("input_421_cast_fp16")];
            tensor<int32, [1]> x_285_axes_0 = const()[name = string("x_285_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_16_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_16_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146576768)))];
            tensor<fp16, [1024]> enc_layers_16_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_16_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146578880)))];
            tensor<fp16, [1, 188, 1024]> x_285_cast_fp16 = layer_norm(axes = x_285_axes_0, beta = enc_layers_16_norm_feed_forward2_bias_to_fp16, epsilon = var_5053_to_fp16, gamma = enc_layers_16_norm_feed_forward2_weight_to_fp16, x = input_421_cast_fp16)[name = string("x_285_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_16_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146580992))), lut = tensor<fp16, [256, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148153920))))[name = string("blocks_16_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_186_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_16_ff2_1_weight_to_fp16_palettized, x = x_285_cast_fp16)[name = string("linear_186_cast_fp16")];
            tensor<fp16, [4096]> blocks_16_ff2_1_scale_to_fp16 = const()[name = string("blocks_16_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148158080)))];
            tensor<fp16, [1, 188, 4096]> input_423_cast_fp16 = mul(x = linear_186_cast_fp16, y = blocks_16_ff2_1_scale_to_fp16)[name = string("input_423_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_287_cast_fp16 = silu(x = input_423_cast_fp16)[name = string("x_287_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_16_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148166336))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149739264))))[name = string("blocks_16_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_187_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_16_ff2_2_weight_to_fp16_palettized, x = x_287_cast_fp16)[name = string("linear_187_cast_fp16")];
            tensor<fp16, [1024]> blocks_16_ff2_2_scale_to_fp16 = const()[name = string("blocks_16_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149740352)))];
            tensor<fp16, [1, 188, 1024]> var_5187_cast_fp16 = mul(x = linear_187_cast_fp16, y = blocks_16_ff2_2_scale_to_fp16)[name = string("op_5187_cast_fp16")];
            fp16 var_5188_to_fp16 = const()[name = string("op_5188_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5189_cast_fp16 = mul(x = var_5187_cast_fp16, y = var_5188_to_fp16)[name = string("op_5189_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_425_cast_fp16 = add(x = input_421_cast_fp16, y = var_5189_cast_fp16)[name = string("input_425_cast_fp16")];
            tensor<int32, [1]> input_427_axes_0 = const()[name = string("input_427_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_16_norm_out_weight_to_fp16 = const()[name = string("enc_layers_16_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149742464)))];
            tensor<fp16, [1024]> enc_layers_16_norm_out_bias_to_fp16 = const()[name = string("enc_layers_16_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149744576)))];
            tensor<fp16, [1, 188, 1024]> input_427_cast_fp16 = layer_norm(axes = input_427_axes_0, beta = enc_layers_16_norm_out_bias_to_fp16, epsilon = var_5053_to_fp16, gamma = enc_layers_16_norm_out_weight_to_fp16, x = input_425_cast_fp16)[name = string("input_427_cast_fp16")];
            int32 var_5203 = const()[name = string("op_5203"), val = int32(-1)];
            tensor<int32, [1]> x_289_axes_0 = const()[name = string("x_289_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_17_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_17_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149746688)))];
            tensor<fp16, [1024]> enc_layers_17_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_17_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149748800)))];
            fp16 var_5210_to_fp16 = const()[name = string("op_5210_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_289_cast_fp16 = layer_norm(axes = x_289_axes_0, beta = enc_layers_17_norm_feed_forward1_bias_to_fp16, epsilon = var_5210_to_fp16, gamma = enc_layers_17_norm_feed_forward1_weight_to_fp16, x = input_427_cast_fp16)[name = string("x_289_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_17_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149750912))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151848128))))[name = string("blocks_17_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_188_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_17_ff1_1_weight_to_fp16_palettized, x = x_289_cast_fp16)[name = string("linear_188_cast_fp16")];
            tensor<fp16, [4096]> blocks_17_ff1_1_scale_to_fp16 = const()[name = string("blocks_17_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151856384)))];
            tensor<fp16, [1, 188, 4096]> input_429_cast_fp16 = mul(x = linear_188_cast_fp16, y = blocks_17_ff1_1_scale_to_fp16)[name = string("input_429_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_291_cast_fp16 = silu(x = input_429_cast_fp16)[name = string("x_291_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_17_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151864640))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153961856))))[name = string("blocks_17_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_189_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_17_ff1_2_weight_to_fp16_palettized, x = x_291_cast_fp16)[name = string("linear_189_cast_fp16")];
            tensor<fp16, [1024]> blocks_17_ff1_2_scale_to_fp16 = const()[name = string("blocks_17_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153963968)))];
            tensor<fp16, [1, 188, 1024]> var_5235_cast_fp16 = mul(x = linear_189_cast_fp16, y = blocks_17_ff1_2_scale_to_fp16)[name = string("op_5235_cast_fp16")];
            fp16 var_5236_to_fp16 = const()[name = string("op_5236_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5237_cast_fp16 = mul(x = var_5235_cast_fp16, y = var_5236_to_fp16)[name = string("op_5237_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_431_cast_fp16 = add(x = input_427_cast_fp16, y = var_5237_cast_fp16)[name = string("input_431_cast_fp16")];
            tensor<int32, [1]> x_293_axes_0 = const()[name = string("x_293_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_17_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_17_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153966080)))];
            tensor<fp16, [1024]> enc_layers_17_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_17_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153968192)))];
            tensor<fp16, [1, 188, 1024]> x_293_cast_fp16 = layer_norm(axes = x_293_axes_0, beta = enc_layers_17_norm_self_att_bias_to_fp16, epsilon = var_5210_to_fp16, gamma = enc_layers_17_norm_self_att_weight_to_fp16, x = input_431_cast_fp16)[name = string("x_293_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_17_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153970304))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154232512))))[name = string("blocks_17_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_190_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_17_q_weight_to_fp16_palettized, x = x_293_cast_fp16)[name = string("linear_190_cast_fp16")];
            tensor<fp16, [1024]> blocks_17_q_scale_to_fp16 = const()[name = string("blocks_17_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154233088)))];
            tensor<fp16, [1, 188, 1024]> var_5245_cast_fp16 = mul(x = linear_190_cast_fp16, y = blocks_17_q_scale_to_fp16)[name = string("op_5245_cast_fp16")];
            tensor<int32, [4]> var_5246 = const()[name = string("op_5246"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5247_cast_fp16 = reshape(shape = var_5246, x = var_5245_cast_fp16)[name = string("op_5247_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_17_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154235200))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154497408))))[name = string("blocks_17_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_191_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_17_k_weight_to_fp16_palettized, x = x_293_cast_fp16)[name = string("linear_191_cast_fp16")];
            tensor<fp16, [1024]> blocks_17_k_scale_to_fp16 = const()[name = string("blocks_17_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154497984)))];
            tensor<fp16, [1, 188, 1024]> var_5253_cast_fp16 = mul(x = linear_191_cast_fp16, y = blocks_17_k_scale_to_fp16)[name = string("op_5253_cast_fp16")];
            tensor<int32, [4]> var_5254 = const()[name = string("op_5254"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5255_cast_fp16 = reshape(shape = var_5254, x = var_5253_cast_fp16)[name = string("op_5255_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_17_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154500096))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154762304))))[name = string("blocks_17_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_192_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_17_v_weight_to_fp16_palettized, x = x_293_cast_fp16)[name = string("linear_192_cast_fp16")];
            tensor<fp16, [1024]> blocks_17_v_scale_to_fp16 = const()[name = string("blocks_17_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154762880)))];
            tensor<fp16, [1, 188, 1024]> var_5261_cast_fp16 = mul(x = linear_192_cast_fp16, y = blocks_17_v_scale_to_fp16)[name = string("op_5261_cast_fp16")];
            tensor<int32, [4]> var_5262 = const()[name = string("op_5262"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5263_cast_fp16 = reshape(shape = var_5262, x = var_5261_cast_fp16)[name = string("op_5263_cast_fp16")];
            tensor<int32, [4]> v_35_perm_0 = const()[name = string("v_35_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_118_to_fp16 = const()[name = string("const_118_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154764992)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_35_cast_fp16 = add(x = var_5247_cast_fp16, y = const_118_to_fp16)[name = string("q_u_35_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_119_to_fp16 = const()[name = string("const_119_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154767104)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_35_cast_fp16 = add(x = var_5247_cast_fp16, y = const_119_to_fp16)[name = string("q_v_35_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_17_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154769216))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155031424))))[name = string("blocks_17_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_193_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_17_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_193_cast_fp16")];
            tensor<fp16, [1024]> blocks_17_pos_scale_to_fp16 = const()[name = string("blocks_17_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155032000)))];
            tensor<fp16, [1, 375, 1024]> var_5275_cast_fp16 = mul(x = linear_193_cast_fp16, y = blocks_17_pos_scale_to_fp16)[name = string("op_5275_cast_fp16")];
            tensor<int32, [4]> var_5276 = const()[name = string("op_5276"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_35_cast_fp16 = reshape(shape = var_5276, x = var_5275_cast_fp16)[name = string("p_35_cast_fp16")];
            bool input_433_transpose_x_0 = const()[name = string("input_433_transpose_x_0"), val = bool(false)];
            bool input_433_transpose_y_0 = const()[name = string("input_433_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_260_perm_0 = const()[name = string("transpose_260_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_261_perm_0 = const()[name = string("transpose_261_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_261 = transpose(perm = transpose_261_perm_0, x = p_35_cast_fp16)[name = string("transpose_343")];
            tensor<fp16, [1, 8, 188, 128]> transpose_260 = transpose(perm = transpose_260_perm_0, x = q_v_35_cast_fp16)[name = string("transpose_344")];
            tensor<fp16, [1, 8, 188, 375]> input_433_cast_fp16 = matmul(transpose_x = input_433_transpose_x_0, transpose_y = input_433_transpose_y_0, x = transpose_260, y = transpose_261)[name = string("input_433_cast_fp16")];
            tensor<int32, [8]> var_5282_pad_0 = const()[name = string("op_5282_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_5282_mode_0 = const()[name = string("op_5282_mode_0"), val = string("constant")];
            fp16 const_29_to_fp16 = const()[name = string("const_29_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_5282_cast_fp16 = pad(constant_val = const_29_to_fp16, mode = var_5282_mode_0, pad = var_5282_pad_0, x = input_433_cast_fp16)[name = string("op_5282_cast_fp16")];
            tensor<int32, [4]> var_5283 = const()[name = string("op_5283"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_69_cast_fp16 = reshape(shape = var_5283, x = var_5282_cast_fp16)[name = string("scores_69_cast_fp16")];
            tensor<int32, [4]> var_5287_begin_0 = const()[name = string("op_5287_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_5287_end_0 = const()[name = string("op_5287_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_5287_end_mask_0 = const()[name = string("op_5287_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_5287_cast_fp16 = slice_by_index(begin = var_5287_begin_0, end = var_5287_end_0, end_mask = var_5287_end_mask_0, x = scores_69_cast_fp16)[name = string("op_5287_cast_fp16")];
            tensor<int32, [4]> var_5288 = const()[name = string("op_5288"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_5289_cast_fp16 = reshape(shape = var_5288, x = var_5287_cast_fp16)[name = string("op_5289_cast_fp16")];
            tensor<int32, [4]> bd_35_begin_0 = const()[name = string("bd_35_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_35_end_0 = const()[name = string("bd_35_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_35_end_mask_0 = const()[name = string("bd_35_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_35_cast_fp16 = slice_by_index(begin = bd_35_begin_0, end = bd_35_end_0, end_mask = bd_35_end_mask_0, x = var_5289_cast_fp16)[name = string("bd_35_cast_fp16")];
            bool var_5292_transpose_x_0 = const()[name = string("op_5292_transpose_x_0"), val = bool(false)];
            bool var_5292_transpose_y_0 = const()[name = string("op_5292_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_262_perm_0 = const()[name = string("transpose_262_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_263_perm_0 = const()[name = string("transpose_263_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_263 = transpose(perm = transpose_263_perm_0, x = var_5255_cast_fp16)[name = string("transpose_341")];
            tensor<fp16, [1, 8, 188, 128]> transpose_262 = transpose(perm = transpose_262_perm_0, x = q_u_35_cast_fp16)[name = string("transpose_342")];
            tensor<fp16, [1, 8, 188, 188]> var_5292_cast_fp16 = matmul(transpose_x = var_5292_transpose_x_0, transpose_y = var_5292_transpose_y_0, x = transpose_262, y = transpose_263)[name = string("op_5292_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5293_cast_fp16 = add(x = var_5292_cast_fp16, y = bd_35_cast_fp16)[name = string("op_5293_cast_fp16")];
            fp16 var_5294_to_fp16 = const()[name = string("op_5294_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_5295_cast_fp16 = mul(x = var_5293_cast_fp16, y = var_5294_to_fp16)[name = string("op_5295_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_71_cast_fp16 = add(x = var_5295_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_71_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5297_cast_fp16 = softmax(axis = var_5203, x = scores_71_cast_fp16)[name = string("op_5297_cast_fp16")];
            bool out_35_transpose_x_0 = const()[name = string("out_35_transpose_x_0"), val = bool(false)];
            bool out_35_transpose_y_0 = const()[name = string("out_35_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_35_cast_fp16 = transpose(perm = v_35_perm_0, x = var_5263_cast_fp16)[name = string("transpose_340")];
            tensor<fp16, [1, 8, 188, 128]> out_35_cast_fp16 = matmul(transpose_x = out_35_transpose_x_0, transpose_y = out_35_transpose_y_0, x = var_5297_cast_fp16, y = v_35_cast_fp16)[name = string("out_35_cast_fp16")];
            tensor<int32, [4]> var_5299_perm_0 = const()[name = string("op_5299_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5300 = const()[name = string("op_5300"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_5299_cast_fp16 = transpose(perm = var_5299_perm_0, x = out_35_cast_fp16)[name = string("transpose_339")];
            tensor<fp16, [1, 188, 1024]> x_295_cast_fp16 = reshape(shape = var_5300, x = var_5299_cast_fp16)[name = string("x_295_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_17_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155034112))), lut = tensor<fp16, [64, 1, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155296320))))[name = string("blocks_17_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_194_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_17_o_weight_to_fp16_palettized, x = x_295_cast_fp16)[name = string("linear_194_cast_fp16")];
            tensor<fp16, [1024]> blocks_17_o_scale_to_fp16 = const()[name = string("blocks_17_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155296896)))];
            tensor<fp16, [1, 188, 1024]> var_5306_cast_fp16 = mul(x = linear_194_cast_fp16, y = blocks_17_o_scale_to_fp16)[name = string("op_5306_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_435_cast_fp16 = add(x = input_431_cast_fp16, y = var_5306_cast_fp16)[name = string("input_435_cast_fp16")];
            tensor<int32, [1]> x_297_axes_0 = const()[name = string("x_297_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_17_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_17_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155299008)))];
            tensor<fp16, [1024]> enc_layers_17_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_17_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155301120)))];
            tensor<fp16, [1, 188, 1024]> x_297_cast_fp16 = layer_norm(axes = x_297_axes_0, beta = enc_layers_17_norm_conv_bias_to_fp16, epsilon = var_5210_to_fp16, gamma = enc_layers_17_norm_conv_weight_to_fp16, x = input_435_cast_fp16)[name = string("x_297_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_17_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155303232))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156351872))))[name = string("blocks_17_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_195_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_17_pw1_weight_to_fp16_palettized, x = x_297_cast_fp16)[name = string("linear_195_cast_fp16")];
            tensor<fp16, [2048]> blocks_17_pw1_scale_to_fp16 = const()[name = string("blocks_17_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156356032)))];
            tensor<fp16, [1, 188, 2048]> input_437_cast_fp16 = mul(x = linear_195_cast_fp16, y = blocks_17_pw1_scale_to_fp16)[name = string("input_437_cast_fp16")];
            int32 var_5315_split_num_splits_0 = const()[name = string("op_5315_split_num_splits_0"), val = int32(2)];
            int32 var_5315_split_axis_0 = const()[name = string("op_5315_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_5315_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_5315_split_cast_fp16_1 = split(axis = var_5315_split_axis_0, num_splits = var_5315_split_num_splits_0, x = input_437_cast_fp16)[name = string("op_5315_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_5315_split_1_sigmoid_cast_fp16 = sigmoid(x = var_5315_split_cast_fp16_1)[name = string("op_5315_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_5315_cast_fp16 = mul(x = var_5315_split_cast_fp16_0, y = var_5315_split_1_sigmoid_cast_fp16)[name = string("op_5315_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_69_cast_fp16 = mul(x = var_5315_cast_fp16, y = valid_cast_fp16)[name = string("h_69_cast_fp16")];
            tensor<int32, [3]> input_439_perm_0 = const()[name = string("input_439_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_441_pad_type_0 = const()[name = string("input_441_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_441_pad_0 = const()[name = string("input_441_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_441_groups_0 = const()[name = string("input_441_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_441_strides_0 = const()[name = string("input_441_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_441_dilations_0 = const()[name = string("input_441_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_70_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156360192))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156369472))))[name = string("const_70_to_fp16_quantized")];
            tensor<fp16, [1024]> const_71_to_fp16 = const()[name = string("const_71_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156371584)))];
            tensor<fp16, [1, 1024, 188]> input_439_cast_fp16 = transpose(perm = input_439_perm_0, x = h_69_cast_fp16)[name = string("transpose_338")];
            tensor<fp16, [1, 1024, 188]> input_443_cast_fp16 = conv(bias = const_71_to_fp16, dilations = input_441_dilations_0, groups = input_441_groups_0, pad = input_441_pad_0, pad_type = input_441_pad_type_0, strides = input_441_strides_0, weight = const_70_to_fp16_quantized, x = input_439_cast_fp16)[name = string("input_443_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_71_cast_fp16 = silu(x = input_443_cast_fp16)[name = string("h_71_cast_fp16")];
            tensor<int32, [3]> x_299_perm_0 = const()[name = string("x_299_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_17_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156373696))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156898048))))[name = string("blocks_17_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_299_cast_fp16 = transpose(perm = x_299_perm_0, x = h_71_cast_fp16)[name = string("transpose_337")];
            tensor<fp16, [1, 188, 1024]> linear_196_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_17_pw2_weight_to_fp16_palettized, x = x_299_cast_fp16)[name = string("linear_196_cast_fp16")];
            tensor<fp16, [1024]> blocks_17_pw2_scale_to_fp16 = const()[name = string("blocks_17_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156900160)))];
            tensor<fp16, [1, 188, 1024]> var_5330_cast_fp16 = mul(x = linear_196_cast_fp16, y = blocks_17_pw2_scale_to_fp16)[name = string("op_5330_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_445_cast_fp16 = add(x = input_435_cast_fp16, y = var_5330_cast_fp16)[name = string("input_445_cast_fp16")];
            tensor<int32, [1]> x_301_axes_0 = const()[name = string("x_301_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_17_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_17_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156902272)))];
            tensor<fp16, [1024]> enc_layers_17_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_17_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156904384)))];
            tensor<fp16, [1, 188, 1024]> x_301_cast_fp16 = layer_norm(axes = x_301_axes_0, beta = enc_layers_17_norm_feed_forward2_bias_to_fp16, epsilon = var_5210_to_fp16, gamma = enc_layers_17_norm_feed_forward2_weight_to_fp16, x = input_445_cast_fp16)[name = string("x_301_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_17_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156906496))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159003712))))[name = string("blocks_17_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_197_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_17_ff2_1_weight_to_fp16_palettized, x = x_301_cast_fp16)[name = string("linear_197_cast_fp16")];
            tensor<fp16, [4096]> blocks_17_ff2_1_scale_to_fp16 = const()[name = string("blocks_17_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159011968)))];
            tensor<fp16, [1, 188, 4096]> input_447_cast_fp16 = mul(x = linear_197_cast_fp16, y = blocks_17_ff2_1_scale_to_fp16)[name = string("input_447_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_303_cast_fp16 = silu(x = input_447_cast_fp16)[name = string("x_303_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_17_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159020224))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161117440))))[name = string("blocks_17_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_198_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_17_ff2_2_weight_to_fp16_palettized, x = x_303_cast_fp16)[name = string("linear_198_cast_fp16")];
            tensor<fp16, [1024]> blocks_17_ff2_2_scale_to_fp16 = const()[name = string("blocks_17_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161119552)))];
            tensor<fp16, [1, 188, 1024]> var_5344_cast_fp16 = mul(x = linear_198_cast_fp16, y = blocks_17_ff2_2_scale_to_fp16)[name = string("op_5344_cast_fp16")];
            fp16 var_5345_to_fp16 = const()[name = string("op_5345_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5346_cast_fp16 = mul(x = var_5344_cast_fp16, y = var_5345_to_fp16)[name = string("op_5346_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_449_cast_fp16 = add(x = input_445_cast_fp16, y = var_5346_cast_fp16)[name = string("input_449_cast_fp16")];
            tensor<int32, [1]> input_451_axes_0 = const()[name = string("input_451_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_17_norm_out_weight_to_fp16 = const()[name = string("enc_layers_17_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161121664)))];
            tensor<fp16, [1024]> enc_layers_17_norm_out_bias_to_fp16 = const()[name = string("enc_layers_17_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161123776)))];
            tensor<fp16, [1, 188, 1024]> input_451_cast_fp16 = layer_norm(axes = input_451_axes_0, beta = enc_layers_17_norm_out_bias_to_fp16, epsilon = var_5210_to_fp16, gamma = enc_layers_17_norm_out_weight_to_fp16, x = input_449_cast_fp16)[name = string("input_451_cast_fp16")];
            int32 var_5360 = const()[name = string("op_5360"), val = int32(-1)];
            tensor<int32, [1]> x_305_axes_0 = const()[name = string("x_305_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_18_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_18_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161125888)))];
            tensor<fp16, [1024]> enc_layers_18_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_18_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161128000)))];
            fp16 var_5367_to_fp16 = const()[name = string("op_5367_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_305_cast_fp16 = layer_norm(axes = x_305_axes_0, beta = enc_layers_18_norm_feed_forward1_bias_to_fp16, epsilon = var_5367_to_fp16, gamma = enc_layers_18_norm_feed_forward1_weight_to_fp16, x = input_451_cast_fp16)[name = string("x_305_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_18_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161130112))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163227328))))[name = string("blocks_18_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_199_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_18_ff1_1_weight_to_fp16_palettized, x = x_305_cast_fp16)[name = string("linear_199_cast_fp16")];
            tensor<fp16, [4096]> blocks_18_ff1_1_scale_to_fp16 = const()[name = string("blocks_18_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163235584)))];
            tensor<fp16, [1, 188, 4096]> input_453_cast_fp16 = mul(x = linear_199_cast_fp16, y = blocks_18_ff1_1_scale_to_fp16)[name = string("input_453_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_307_cast_fp16 = silu(x = input_453_cast_fp16)[name = string("x_307_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_18_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163243840))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165341056))))[name = string("blocks_18_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_200_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_18_ff1_2_weight_to_fp16_palettized, x = x_307_cast_fp16)[name = string("linear_200_cast_fp16")];
            tensor<fp16, [1024]> blocks_18_ff1_2_scale_to_fp16 = const()[name = string("blocks_18_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165343168)))];
            tensor<fp16, [1, 188, 1024]> var_5392_cast_fp16 = mul(x = linear_200_cast_fp16, y = blocks_18_ff1_2_scale_to_fp16)[name = string("op_5392_cast_fp16")];
            fp16 var_5393_to_fp16 = const()[name = string("op_5393_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5394_cast_fp16 = mul(x = var_5392_cast_fp16, y = var_5393_to_fp16)[name = string("op_5394_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_455_cast_fp16 = add(x = input_451_cast_fp16, y = var_5394_cast_fp16)[name = string("input_455_cast_fp16")];
            tensor<int32, [1]> x_309_axes_0 = const()[name = string("x_309_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_18_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_18_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165345280)))];
            tensor<fp16, [1024]> enc_layers_18_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_18_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165347392)))];
            tensor<fp16, [1, 188, 1024]> x_309_cast_fp16 = layer_norm(axes = x_309_axes_0, beta = enc_layers_18_norm_self_att_bias_to_fp16, epsilon = var_5367_to_fp16, gamma = enc_layers_18_norm_self_att_weight_to_fp16, x = input_455_cast_fp16)[name = string("x_309_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_18_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165349504))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165873856))))[name = string("blocks_18_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_201_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_18_q_weight_to_fp16_palettized, x = x_309_cast_fp16)[name = string("linear_201_cast_fp16")];
            tensor<fp16, [1024]> blocks_18_q_scale_to_fp16 = const()[name = string("blocks_18_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165875968)))];
            tensor<fp16, [1, 188, 1024]> var_5402_cast_fp16 = mul(x = linear_201_cast_fp16, y = blocks_18_q_scale_to_fp16)[name = string("op_5402_cast_fp16")];
            tensor<int32, [4]> var_5403 = const()[name = string("op_5403"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5404_cast_fp16 = reshape(shape = var_5403, x = var_5402_cast_fp16)[name = string("op_5404_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_18_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165878080))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166402432))))[name = string("blocks_18_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_202_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_18_k_weight_to_fp16_palettized, x = x_309_cast_fp16)[name = string("linear_202_cast_fp16")];
            tensor<fp16, [1024]> blocks_18_k_scale_to_fp16 = const()[name = string("blocks_18_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166404544)))];
            tensor<fp16, [1, 188, 1024]> var_5410_cast_fp16 = mul(x = linear_202_cast_fp16, y = blocks_18_k_scale_to_fp16)[name = string("op_5410_cast_fp16")];
            tensor<int32, [4]> var_5411 = const()[name = string("op_5411"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5412_cast_fp16 = reshape(shape = var_5411, x = var_5410_cast_fp16)[name = string("op_5412_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_18_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166406656))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166931008))))[name = string("blocks_18_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_203_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_18_v_weight_to_fp16_palettized, x = x_309_cast_fp16)[name = string("linear_203_cast_fp16")];
            tensor<fp16, [1024]> blocks_18_v_scale_to_fp16 = const()[name = string("blocks_18_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166933120)))];
            tensor<fp16, [1, 188, 1024]> var_5418_cast_fp16 = mul(x = linear_203_cast_fp16, y = blocks_18_v_scale_to_fp16)[name = string("op_5418_cast_fp16")];
            tensor<int32, [4]> var_5419 = const()[name = string("op_5419"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5420_cast_fp16 = reshape(shape = var_5419, x = var_5418_cast_fp16)[name = string("op_5420_cast_fp16")];
            tensor<int32, [4]> v_37_perm_0 = const()[name = string("v_37_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_120_to_fp16 = const()[name = string("const_120_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166935232)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_37_cast_fp16 = add(x = var_5404_cast_fp16, y = const_120_to_fp16)[name = string("q_u_37_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_121_to_fp16 = const()[name = string("const_121_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166937344)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_37_cast_fp16 = add(x = var_5404_cast_fp16, y = const_121_to_fp16)[name = string("q_v_37_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_18_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166939456))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167463808))))[name = string("blocks_18_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_204_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_18_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_204_cast_fp16")];
            tensor<fp16, [1024]> blocks_18_pos_scale_to_fp16 = const()[name = string("blocks_18_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167465920)))];
            tensor<fp16, [1, 375, 1024]> var_5432_cast_fp16 = mul(x = linear_204_cast_fp16, y = blocks_18_pos_scale_to_fp16)[name = string("op_5432_cast_fp16")];
            tensor<int32, [4]> var_5433 = const()[name = string("op_5433"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_37_cast_fp16 = reshape(shape = var_5433, x = var_5432_cast_fp16)[name = string("p_37_cast_fp16")];
            bool input_457_transpose_x_0 = const()[name = string("input_457_transpose_x_0"), val = bool(false)];
            bool input_457_transpose_y_0 = const()[name = string("input_457_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_264_perm_0 = const()[name = string("transpose_264_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_265_perm_0 = const()[name = string("transpose_265_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_265 = transpose(perm = transpose_265_perm_0, x = p_37_cast_fp16)[name = string("transpose_335")];
            tensor<fp16, [1, 8, 188, 128]> transpose_264 = transpose(perm = transpose_264_perm_0, x = q_v_37_cast_fp16)[name = string("transpose_336")];
            tensor<fp16, [1, 8, 188, 375]> input_457_cast_fp16 = matmul(transpose_x = input_457_transpose_x_0, transpose_y = input_457_transpose_y_0, x = transpose_264, y = transpose_265)[name = string("input_457_cast_fp16")];
            tensor<int32, [8]> var_5439_pad_0 = const()[name = string("op_5439_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_5439_mode_0 = const()[name = string("op_5439_mode_0"), val = string("constant")];
            fp16 const_30_to_fp16 = const()[name = string("const_30_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_5439_cast_fp16 = pad(constant_val = const_30_to_fp16, mode = var_5439_mode_0, pad = var_5439_pad_0, x = input_457_cast_fp16)[name = string("op_5439_cast_fp16")];
            tensor<int32, [4]> var_5440 = const()[name = string("op_5440"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_73_cast_fp16 = reshape(shape = var_5440, x = var_5439_cast_fp16)[name = string("scores_73_cast_fp16")];
            tensor<int32, [4]> var_5444_begin_0 = const()[name = string("op_5444_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_5444_end_0 = const()[name = string("op_5444_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_5444_end_mask_0 = const()[name = string("op_5444_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_5444_cast_fp16 = slice_by_index(begin = var_5444_begin_0, end = var_5444_end_0, end_mask = var_5444_end_mask_0, x = scores_73_cast_fp16)[name = string("op_5444_cast_fp16")];
            tensor<int32, [4]> var_5445 = const()[name = string("op_5445"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_5446_cast_fp16 = reshape(shape = var_5445, x = var_5444_cast_fp16)[name = string("op_5446_cast_fp16")];
            tensor<int32, [4]> bd_37_begin_0 = const()[name = string("bd_37_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_37_end_0 = const()[name = string("bd_37_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_37_end_mask_0 = const()[name = string("bd_37_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_37_cast_fp16 = slice_by_index(begin = bd_37_begin_0, end = bd_37_end_0, end_mask = bd_37_end_mask_0, x = var_5446_cast_fp16)[name = string("bd_37_cast_fp16")];
            bool var_5449_transpose_x_0 = const()[name = string("op_5449_transpose_x_0"), val = bool(false)];
            bool var_5449_transpose_y_0 = const()[name = string("op_5449_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_266_perm_0 = const()[name = string("transpose_266_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_267_perm_0 = const()[name = string("transpose_267_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_267 = transpose(perm = transpose_267_perm_0, x = var_5412_cast_fp16)[name = string("transpose_333")];
            tensor<fp16, [1, 8, 188, 128]> transpose_266 = transpose(perm = transpose_266_perm_0, x = q_u_37_cast_fp16)[name = string("transpose_334")];
            tensor<fp16, [1, 8, 188, 188]> var_5449_cast_fp16 = matmul(transpose_x = var_5449_transpose_x_0, transpose_y = var_5449_transpose_y_0, x = transpose_266, y = transpose_267)[name = string("op_5449_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5450_cast_fp16 = add(x = var_5449_cast_fp16, y = bd_37_cast_fp16)[name = string("op_5450_cast_fp16")];
            fp16 var_5451_to_fp16 = const()[name = string("op_5451_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_5452_cast_fp16 = mul(x = var_5450_cast_fp16, y = var_5451_to_fp16)[name = string("op_5452_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_75_cast_fp16 = add(x = var_5452_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_75_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5454_cast_fp16 = softmax(axis = var_5360, x = scores_75_cast_fp16)[name = string("op_5454_cast_fp16")];
            bool out_37_transpose_x_0 = const()[name = string("out_37_transpose_x_0"), val = bool(false)];
            bool out_37_transpose_y_0 = const()[name = string("out_37_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_37_cast_fp16 = transpose(perm = v_37_perm_0, x = var_5420_cast_fp16)[name = string("transpose_332")];
            tensor<fp16, [1, 8, 188, 128]> out_37_cast_fp16 = matmul(transpose_x = out_37_transpose_x_0, transpose_y = out_37_transpose_y_0, x = var_5454_cast_fp16, y = v_37_cast_fp16)[name = string("out_37_cast_fp16")];
            tensor<int32, [4]> var_5456_perm_0 = const()[name = string("op_5456_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5457 = const()[name = string("op_5457"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_5456_cast_fp16 = transpose(perm = var_5456_perm_0, x = out_37_cast_fp16)[name = string("transpose_331")];
            tensor<fp16, [1, 188, 1024]> x_311_cast_fp16 = reshape(shape = var_5457, x = var_5456_cast_fp16)[name = string("x_311_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_18_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167468032))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167992384))))[name = string("blocks_18_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_205_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_18_o_weight_to_fp16_palettized, x = x_311_cast_fp16)[name = string("linear_205_cast_fp16")];
            tensor<fp16, [1024]> blocks_18_o_scale_to_fp16 = const()[name = string("blocks_18_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167994496)))];
            tensor<fp16, [1, 188, 1024]> var_5463_cast_fp16 = mul(x = linear_205_cast_fp16, y = blocks_18_o_scale_to_fp16)[name = string("op_5463_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_459_cast_fp16 = add(x = input_455_cast_fp16, y = var_5463_cast_fp16)[name = string("input_459_cast_fp16")];
            tensor<int32, [1]> x_313_axes_0 = const()[name = string("x_313_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_18_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_18_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167996608)))];
            tensor<fp16, [1024]> enc_layers_18_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_18_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167998720)))];
            tensor<fp16, [1, 188, 1024]> x_313_cast_fp16 = layer_norm(axes = x_313_axes_0, beta = enc_layers_18_norm_conv_bias_to_fp16, epsilon = var_5367_to_fp16, gamma = enc_layers_18_norm_conv_weight_to_fp16, x = input_459_cast_fp16)[name = string("x_313_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_18_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168000832))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169049472))))[name = string("blocks_18_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_206_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_18_pw1_weight_to_fp16_palettized, x = x_313_cast_fp16)[name = string("linear_206_cast_fp16")];
            tensor<fp16, [2048]> blocks_18_pw1_scale_to_fp16 = const()[name = string("blocks_18_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169053632)))];
            tensor<fp16, [1, 188, 2048]> input_461_cast_fp16 = mul(x = linear_206_cast_fp16, y = blocks_18_pw1_scale_to_fp16)[name = string("input_461_cast_fp16")];
            int32 var_5472_split_num_splits_0 = const()[name = string("op_5472_split_num_splits_0"), val = int32(2)];
            int32 var_5472_split_axis_0 = const()[name = string("op_5472_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_5472_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_5472_split_cast_fp16_1 = split(axis = var_5472_split_axis_0, num_splits = var_5472_split_num_splits_0, x = input_461_cast_fp16)[name = string("op_5472_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_5472_split_1_sigmoid_cast_fp16 = sigmoid(x = var_5472_split_cast_fp16_1)[name = string("op_5472_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_5472_cast_fp16 = mul(x = var_5472_split_cast_fp16_0, y = var_5472_split_1_sigmoid_cast_fp16)[name = string("op_5472_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_73_cast_fp16 = mul(x = var_5472_cast_fp16, y = valid_cast_fp16)[name = string("h_73_cast_fp16")];
            tensor<int32, [3]> input_463_perm_0 = const()[name = string("input_463_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_465_pad_type_0 = const()[name = string("input_465_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_465_pad_0 = const()[name = string("input_465_pad_0"), val = tensor<int32, [2]>([4, 4])];
            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, [1]> input_465_dilations_0 = const()[name = string("input_465_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_72_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169057792))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169067072))))[name = string("const_72_to_fp16_quantized")];
            tensor<fp16, [1024]> const_73_to_fp16 = const()[name = string("const_73_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169069184)))];
            tensor<fp16, [1, 1024, 188]> input_463_cast_fp16 = transpose(perm = input_463_perm_0, x = h_73_cast_fp16)[name = string("transpose_330")];
            tensor<fp16, [1, 1024, 188]> input_467_cast_fp16 = conv(bias = const_73_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_72_to_fp16_quantized, x = input_463_cast_fp16)[name = string("input_467_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_75_cast_fp16 = silu(x = input_467_cast_fp16)[name = string("h_75_cast_fp16")];
            tensor<int32, [3]> x_315_perm_0 = const()[name = string("x_315_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_18_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169071296))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169595648))))[name = string("blocks_18_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_315_cast_fp16 = transpose(perm = x_315_perm_0, x = h_75_cast_fp16)[name = string("transpose_329")];
            tensor<fp16, [1, 188, 1024]> linear_207_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_18_pw2_weight_to_fp16_palettized, x = x_315_cast_fp16)[name = string("linear_207_cast_fp16")];
            tensor<fp16, [1024]> blocks_18_pw2_scale_to_fp16 = const()[name = string("blocks_18_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169597760)))];
            tensor<fp16, [1, 188, 1024]> var_5487_cast_fp16 = mul(x = linear_207_cast_fp16, y = blocks_18_pw2_scale_to_fp16)[name = string("op_5487_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_469_cast_fp16 = add(x = input_459_cast_fp16, y = var_5487_cast_fp16)[name = string("input_469_cast_fp16")];
            tensor<int32, [1]> x_317_axes_0 = const()[name = string("x_317_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_18_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_18_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169599872)))];
            tensor<fp16, [1024]> enc_layers_18_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_18_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169601984)))];
            tensor<fp16, [1, 188, 1024]> x_317_cast_fp16 = layer_norm(axes = x_317_axes_0, beta = enc_layers_18_norm_feed_forward2_bias_to_fp16, epsilon = var_5367_to_fp16, gamma = enc_layers_18_norm_feed_forward2_weight_to_fp16, x = input_469_cast_fp16)[name = string("x_317_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_18_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169604096))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171701312))))[name = string("blocks_18_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_208_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_18_ff2_1_weight_to_fp16_palettized, x = x_317_cast_fp16)[name = string("linear_208_cast_fp16")];
            tensor<fp16, [4096]> blocks_18_ff2_1_scale_to_fp16 = const()[name = string("blocks_18_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171709568)))];
            tensor<fp16, [1, 188, 4096]> input_471_cast_fp16 = mul(x = linear_208_cast_fp16, y = blocks_18_ff2_1_scale_to_fp16)[name = string("input_471_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_319_cast_fp16 = silu(x = input_471_cast_fp16)[name = string("x_319_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_18_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171717824))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173815040))))[name = string("blocks_18_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_209_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_18_ff2_2_weight_to_fp16_palettized, x = x_319_cast_fp16)[name = string("linear_209_cast_fp16")];
            tensor<fp16, [1024]> blocks_18_ff2_2_scale_to_fp16 = const()[name = string("blocks_18_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173817152)))];
            tensor<fp16, [1, 188, 1024]> var_5501_cast_fp16 = mul(x = linear_209_cast_fp16, y = blocks_18_ff2_2_scale_to_fp16)[name = string("op_5501_cast_fp16")];
            fp16 var_5502_to_fp16 = const()[name = string("op_5502_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5503_cast_fp16 = mul(x = var_5501_cast_fp16, y = var_5502_to_fp16)[name = string("op_5503_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_473_cast_fp16 = add(x = input_469_cast_fp16, y = var_5503_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]> enc_layers_18_norm_out_weight_to_fp16 = const()[name = string("enc_layers_18_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173819264)))];
            tensor<fp16, [1024]> enc_layers_18_norm_out_bias_to_fp16 = const()[name = string("enc_layers_18_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173821376)))];
            tensor<fp16, [1, 188, 1024]> input_475_cast_fp16 = layer_norm(axes = input_475_axes_0, beta = enc_layers_18_norm_out_bias_to_fp16, epsilon = var_5367_to_fp16, gamma = enc_layers_18_norm_out_weight_to_fp16, x = input_473_cast_fp16)[name = string("input_475_cast_fp16")];
            int32 var_5517 = const()[name = string("op_5517"), val = int32(-1)];
            tensor<int32, [1]> x_321_axes_0 = const()[name = string("x_321_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_19_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_19_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173823488)))];
            tensor<fp16, [1024]> enc_layers_19_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_19_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173825600)))];
            fp16 var_5524_to_fp16 = const()[name = string("op_5524_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_321_cast_fp16 = layer_norm(axes = x_321_axes_0, beta = enc_layers_19_norm_feed_forward1_bias_to_fp16, epsilon = var_5524_to_fp16, gamma = enc_layers_19_norm_feed_forward1_weight_to_fp16, x = input_475_cast_fp16)[name = string("x_321_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_19_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173827712))), lut = tensor<fp16, [256, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175400640))))[name = string("blocks_19_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_210_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_19_ff1_1_weight_to_fp16_palettized, x = x_321_cast_fp16)[name = string("linear_210_cast_fp16")];
            tensor<fp16, [4096]> blocks_19_ff1_1_scale_to_fp16 = const()[name = string("blocks_19_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175404800)))];
            tensor<fp16, [1, 188, 4096]> input_477_cast_fp16 = mul(x = linear_210_cast_fp16, y = blocks_19_ff1_1_scale_to_fp16)[name = string("input_477_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_323_cast_fp16 = silu(x = input_477_cast_fp16)[name = string("x_323_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_19_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175413056))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176985984))))[name = string("blocks_19_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_211_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_19_ff1_2_weight_to_fp16_palettized, x = x_323_cast_fp16)[name = string("linear_211_cast_fp16")];
            tensor<fp16, [1024]> blocks_19_ff1_2_scale_to_fp16 = const()[name = string("blocks_19_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176987072)))];
            tensor<fp16, [1, 188, 1024]> var_5549_cast_fp16 = mul(x = linear_211_cast_fp16, y = blocks_19_ff1_2_scale_to_fp16)[name = string("op_5549_cast_fp16")];
            fp16 var_5550_to_fp16 = const()[name = string("op_5550_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5551_cast_fp16 = mul(x = var_5549_cast_fp16, y = var_5550_to_fp16)[name = string("op_5551_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_479_cast_fp16 = add(x = input_475_cast_fp16, y = var_5551_cast_fp16)[name = string("input_479_cast_fp16")];
            tensor<int32, [1]> x_325_axes_0 = const()[name = string("x_325_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_19_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_19_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176989184)))];
            tensor<fp16, [1024]> enc_layers_19_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_19_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176991296)))];
            tensor<fp16, [1, 188, 1024]> x_325_cast_fp16 = layer_norm(axes = x_325_axes_0, beta = enc_layers_19_norm_self_att_bias_to_fp16, epsilon = var_5524_to_fp16, gamma = enc_layers_19_norm_self_att_weight_to_fp16, x = input_479_cast_fp16)[name = string("x_325_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_19_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176993408))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177517760))))[name = string("blocks_19_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_212_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_19_q_weight_to_fp16_palettized, x = x_325_cast_fp16)[name = string("linear_212_cast_fp16")];
            tensor<fp16, [1024]> blocks_19_q_scale_to_fp16 = const()[name = string("blocks_19_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177519872)))];
            tensor<fp16, [1, 188, 1024]> var_5559_cast_fp16 = mul(x = linear_212_cast_fp16, y = blocks_19_q_scale_to_fp16)[name = string("op_5559_cast_fp16")];
            tensor<int32, [4]> var_5560 = const()[name = string("op_5560"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5561_cast_fp16 = reshape(shape = var_5560, x = var_5559_cast_fp16)[name = string("op_5561_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_19_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177521984))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178046336))))[name = string("blocks_19_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_213_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_19_k_weight_to_fp16_palettized, x = x_325_cast_fp16)[name = string("linear_213_cast_fp16")];
            tensor<fp16, [1024]> blocks_19_k_scale_to_fp16 = const()[name = string("blocks_19_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178048448)))];
            tensor<fp16, [1, 188, 1024]> var_5567_cast_fp16 = mul(x = linear_213_cast_fp16, y = blocks_19_k_scale_to_fp16)[name = string("op_5567_cast_fp16")];
            tensor<int32, [4]> var_5568 = const()[name = string("op_5568"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5569_cast_fp16 = reshape(shape = var_5568, x = var_5567_cast_fp16)[name = string("op_5569_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_19_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178050560))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178574912))))[name = string("blocks_19_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_214_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_19_v_weight_to_fp16_palettized, x = x_325_cast_fp16)[name = string("linear_214_cast_fp16")];
            tensor<fp16, [1024]> blocks_19_v_scale_to_fp16 = const()[name = string("blocks_19_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178577024)))];
            tensor<fp16, [1, 188, 1024]> var_5575_cast_fp16 = mul(x = linear_214_cast_fp16, y = blocks_19_v_scale_to_fp16)[name = string("op_5575_cast_fp16")];
            tensor<int32, [4]> var_5576 = const()[name = string("op_5576"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5577_cast_fp16 = reshape(shape = var_5576, x = var_5575_cast_fp16)[name = string("op_5577_cast_fp16")];
            tensor<int32, [4]> v_39_perm_0 = const()[name = string("v_39_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_122_to_fp16 = const()[name = string("const_122_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178579136)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_39_cast_fp16 = add(x = var_5561_cast_fp16, y = const_122_to_fp16)[name = string("q_u_39_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_123_to_fp16 = const()[name = string("const_123_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178581248)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_39_cast_fp16 = add(x = var_5561_cast_fp16, y = const_123_to_fp16)[name = string("q_v_39_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_19_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178583360))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179107712))))[name = string("blocks_19_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_215_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_19_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_215_cast_fp16")];
            tensor<fp16, [1024]> blocks_19_pos_scale_to_fp16 = const()[name = string("blocks_19_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179109824)))];
            tensor<fp16, [1, 375, 1024]> var_5589_cast_fp16 = mul(x = linear_215_cast_fp16, y = blocks_19_pos_scale_to_fp16)[name = string("op_5589_cast_fp16")];
            tensor<int32, [4]> var_5590 = const()[name = string("op_5590"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_39_cast_fp16 = reshape(shape = var_5590, x = var_5589_cast_fp16)[name = string("p_39_cast_fp16")];
            bool input_481_transpose_x_0 = const()[name = string("input_481_transpose_x_0"), val = bool(false)];
            bool input_481_transpose_y_0 = const()[name = string("input_481_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_268_perm_0 = const()[name = string("transpose_268_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_269_perm_0 = const()[name = string("transpose_269_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_269 = transpose(perm = transpose_269_perm_0, x = p_39_cast_fp16)[name = string("transpose_327")];
            tensor<fp16, [1, 8, 188, 128]> transpose_268 = transpose(perm = transpose_268_perm_0, x = q_v_39_cast_fp16)[name = string("transpose_328")];
            tensor<fp16, [1, 8, 188, 375]> input_481_cast_fp16 = matmul(transpose_x = input_481_transpose_x_0, transpose_y = input_481_transpose_y_0, x = transpose_268, y = transpose_269)[name = string("input_481_cast_fp16")];
            tensor<int32, [8]> var_5596_pad_0 = const()[name = string("op_5596_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_5596_mode_0 = const()[name = string("op_5596_mode_0"), val = string("constant")];
            fp16 const_31_to_fp16 = const()[name = string("const_31_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_5596_cast_fp16 = pad(constant_val = const_31_to_fp16, mode = var_5596_mode_0, pad = var_5596_pad_0, x = input_481_cast_fp16)[name = string("op_5596_cast_fp16")];
            tensor<int32, [4]> var_5597 = const()[name = string("op_5597"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_77_cast_fp16 = reshape(shape = var_5597, x = var_5596_cast_fp16)[name = string("scores_77_cast_fp16")];
            tensor<int32, [4]> var_5601_begin_0 = const()[name = string("op_5601_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_5601_end_0 = const()[name = string("op_5601_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_5601_end_mask_0 = const()[name = string("op_5601_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_5601_cast_fp16 = slice_by_index(begin = var_5601_begin_0, end = var_5601_end_0, end_mask = var_5601_end_mask_0, x = scores_77_cast_fp16)[name = string("op_5601_cast_fp16")];
            tensor<int32, [4]> var_5602 = const()[name = string("op_5602"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_5603_cast_fp16 = reshape(shape = var_5602, x = var_5601_cast_fp16)[name = string("op_5603_cast_fp16")];
            tensor<int32, [4]> bd_39_begin_0 = const()[name = string("bd_39_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_39_end_0 = const()[name = string("bd_39_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_39_end_mask_0 = const()[name = string("bd_39_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_39_cast_fp16 = slice_by_index(begin = bd_39_begin_0, end = bd_39_end_0, end_mask = bd_39_end_mask_0, x = var_5603_cast_fp16)[name = string("bd_39_cast_fp16")];
            bool var_5606_transpose_x_0 = const()[name = string("op_5606_transpose_x_0"), val = bool(false)];
            bool var_5606_transpose_y_0 = const()[name = string("op_5606_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_270_perm_0 = const()[name = string("transpose_270_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_271_perm_0 = const()[name = string("transpose_271_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_271 = transpose(perm = transpose_271_perm_0, x = var_5569_cast_fp16)[name = string("transpose_325")];
            tensor<fp16, [1, 8, 188, 128]> transpose_270 = transpose(perm = transpose_270_perm_0, x = q_u_39_cast_fp16)[name = string("transpose_326")];
            tensor<fp16, [1, 8, 188, 188]> var_5606_cast_fp16 = matmul(transpose_x = var_5606_transpose_x_0, transpose_y = var_5606_transpose_y_0, x = transpose_270, y = transpose_271)[name = string("op_5606_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5607_cast_fp16 = add(x = var_5606_cast_fp16, y = bd_39_cast_fp16)[name = string("op_5607_cast_fp16")];
            fp16 var_5608_to_fp16 = const()[name = string("op_5608_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_5609_cast_fp16 = mul(x = var_5607_cast_fp16, y = var_5608_to_fp16)[name = string("op_5609_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_79_cast_fp16 = add(x = var_5609_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_79_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5611_cast_fp16 = softmax(axis = var_5517, x = scores_79_cast_fp16)[name = string("op_5611_cast_fp16")];
            bool out_39_transpose_x_0 = const()[name = string("out_39_transpose_x_0"), val = bool(false)];
            bool out_39_transpose_y_0 = const()[name = string("out_39_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_39_cast_fp16 = transpose(perm = v_39_perm_0, x = var_5577_cast_fp16)[name = string("transpose_324")];
            tensor<fp16, [1, 8, 188, 128]> out_39_cast_fp16 = matmul(transpose_x = out_39_transpose_x_0, transpose_y = out_39_transpose_y_0, x = var_5611_cast_fp16, y = v_39_cast_fp16)[name = string("out_39_cast_fp16")];
            tensor<int32, [4]> var_5613_perm_0 = const()[name = string("op_5613_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5614 = const()[name = string("op_5614"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_5613_cast_fp16 = transpose(perm = var_5613_perm_0, x = out_39_cast_fp16)[name = string("transpose_323")];
            tensor<fp16, [1, 188, 1024]> x_327_cast_fp16 = reshape(shape = var_5614, x = var_5613_cast_fp16)[name = string("x_327_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_19_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179111936))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179636288))))[name = string("blocks_19_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_216_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_19_o_weight_to_fp16_palettized, x = x_327_cast_fp16)[name = string("linear_216_cast_fp16")];
            tensor<fp16, [1024]> blocks_19_o_scale_to_fp16 = const()[name = string("blocks_19_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179638400)))];
            tensor<fp16, [1, 188, 1024]> var_5620_cast_fp16 = mul(x = linear_216_cast_fp16, y = blocks_19_o_scale_to_fp16)[name = string("op_5620_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_483_cast_fp16 = add(x = input_479_cast_fp16, y = var_5620_cast_fp16)[name = string("input_483_cast_fp16")];
            tensor<int32, [1]> x_329_axes_0 = const()[name = string("x_329_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_19_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_19_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179640512)))];
            tensor<fp16, [1024]> enc_layers_19_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_19_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179642624)))];
            tensor<fp16, [1, 188, 1024]> x_329_cast_fp16 = layer_norm(axes = x_329_axes_0, beta = enc_layers_19_norm_conv_bias_to_fp16, epsilon = var_5524_to_fp16, gamma = enc_layers_19_norm_conv_weight_to_fp16, x = input_483_cast_fp16)[name = string("x_329_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_19_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179644736))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180693376))))[name = string("blocks_19_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_217_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_19_pw1_weight_to_fp16_palettized, x = x_329_cast_fp16)[name = string("linear_217_cast_fp16")];
            tensor<fp16, [2048]> blocks_19_pw1_scale_to_fp16 = const()[name = string("blocks_19_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180697536)))];
            tensor<fp16, [1, 188, 2048]> input_485_cast_fp16 = mul(x = linear_217_cast_fp16, y = blocks_19_pw1_scale_to_fp16)[name = string("input_485_cast_fp16")];
            int32 var_5629_split_num_splits_0 = const()[name = string("op_5629_split_num_splits_0"), val = int32(2)];
            int32 var_5629_split_axis_0 = const()[name = string("op_5629_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_5629_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_5629_split_cast_fp16_1 = split(axis = var_5629_split_axis_0, num_splits = var_5629_split_num_splits_0, x = input_485_cast_fp16)[name = string("op_5629_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_5629_split_1_sigmoid_cast_fp16 = sigmoid(x = var_5629_split_cast_fp16_1)[name = string("op_5629_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_5629_cast_fp16 = mul(x = var_5629_split_cast_fp16_0, y = var_5629_split_1_sigmoid_cast_fp16)[name = string("op_5629_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_77_cast_fp16 = mul(x = var_5629_cast_fp16, y = valid_cast_fp16)[name = string("h_77_cast_fp16")];
            tensor<int32, [3]> input_487_perm_0 = const()[name = string("input_487_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_489_pad_type_0 = const()[name = string("input_489_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_489_pad_0 = const()[name = string("input_489_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_489_groups_0 = const()[name = string("input_489_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_489_strides_0 = const()[name = string("input_489_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_489_dilations_0 = const()[name = string("input_489_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_74_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180701696))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180710976))))[name = string("const_74_to_fp16_quantized")];
            tensor<fp16, [1024]> const_75_to_fp16 = const()[name = string("const_75_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180713088)))];
            tensor<fp16, [1, 1024, 188]> input_487_cast_fp16 = transpose(perm = input_487_perm_0, x = h_77_cast_fp16)[name = string("transpose_322")];
            tensor<fp16, [1, 1024, 188]> input_491_cast_fp16 = conv(bias = const_75_to_fp16, dilations = input_489_dilations_0, groups = input_489_groups_0, pad = input_489_pad_0, pad_type = input_489_pad_type_0, strides = input_489_strides_0, weight = const_74_to_fp16_quantized, x = input_487_cast_fp16)[name = string("input_491_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_79_cast_fp16 = silu(x = input_491_cast_fp16)[name = string("h_79_cast_fp16")];
            tensor<int32, [3]> x_331_perm_0 = const()[name = string("x_331_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_19_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180715200))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181239552))))[name = string("blocks_19_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_331_cast_fp16 = transpose(perm = x_331_perm_0, x = h_79_cast_fp16)[name = string("transpose_321")];
            tensor<fp16, [1, 188, 1024]> linear_218_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_19_pw2_weight_to_fp16_palettized, x = x_331_cast_fp16)[name = string("linear_218_cast_fp16")];
            tensor<fp16, [1024]> blocks_19_pw2_scale_to_fp16 = const()[name = string("blocks_19_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181241664)))];
            tensor<fp16, [1, 188, 1024]> var_5644_cast_fp16 = mul(x = linear_218_cast_fp16, y = blocks_19_pw2_scale_to_fp16)[name = string("op_5644_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_493_cast_fp16 = add(x = input_483_cast_fp16, y = var_5644_cast_fp16)[name = string("input_493_cast_fp16")];
            tensor<int32, [1]> x_333_axes_0 = const()[name = string("x_333_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_19_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_19_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181243776)))];
            tensor<fp16, [1024]> enc_layers_19_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_19_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181245888)))];
            tensor<fp16, [1, 188, 1024]> x_333_cast_fp16 = layer_norm(axes = x_333_axes_0, beta = enc_layers_19_norm_feed_forward2_bias_to_fp16, epsilon = var_5524_to_fp16, gamma = enc_layers_19_norm_feed_forward2_weight_to_fp16, x = input_493_cast_fp16)[name = string("x_333_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_19_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181248000))), lut = tensor<fp16, [256, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182820928))))[name = string("blocks_19_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_219_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_19_ff2_1_weight_to_fp16_palettized, x = x_333_cast_fp16)[name = string("linear_219_cast_fp16")];
            tensor<fp16, [4096]> blocks_19_ff2_1_scale_to_fp16 = const()[name = string("blocks_19_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182825088)))];
            tensor<fp16, [1, 188, 4096]> input_495_cast_fp16 = mul(x = linear_219_cast_fp16, y = blocks_19_ff2_1_scale_to_fp16)[name = string("input_495_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_335_cast_fp16 = silu(x = input_495_cast_fp16)[name = string("x_335_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_19_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182833344))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184406272))))[name = string("blocks_19_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_220_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_19_ff2_2_weight_to_fp16_palettized, x = x_335_cast_fp16)[name = string("linear_220_cast_fp16")];
            tensor<fp16, [1024]> blocks_19_ff2_2_scale_to_fp16 = const()[name = string("blocks_19_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184407360)))];
            tensor<fp16, [1, 188, 1024]> var_5658_cast_fp16 = mul(x = linear_220_cast_fp16, y = blocks_19_ff2_2_scale_to_fp16)[name = string("op_5658_cast_fp16")];
            fp16 var_5659_to_fp16 = const()[name = string("op_5659_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5660_cast_fp16 = mul(x = var_5658_cast_fp16, y = var_5659_to_fp16)[name = string("op_5660_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_497_cast_fp16 = add(x = input_493_cast_fp16, y = var_5660_cast_fp16)[name = string("input_497_cast_fp16")];
            tensor<int32, [1]> input_499_axes_0 = const()[name = string("input_499_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_19_norm_out_weight_to_fp16 = const()[name = string("enc_layers_19_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184409472)))];
            tensor<fp16, [1024]> enc_layers_19_norm_out_bias_to_fp16 = const()[name = string("enc_layers_19_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184411584)))];
            tensor<fp16, [1, 188, 1024]> input_499_cast_fp16 = layer_norm(axes = input_499_axes_0, beta = enc_layers_19_norm_out_bias_to_fp16, epsilon = var_5524_to_fp16, gamma = enc_layers_19_norm_out_weight_to_fp16, x = input_497_cast_fp16)[name = string("input_499_cast_fp16")];
            int32 var_5674 = const()[name = string("op_5674"), val = int32(-1)];
            tensor<int32, [1]> x_337_axes_0 = const()[name = string("x_337_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_20_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_20_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184413696)))];
            tensor<fp16, [1024]> enc_layers_20_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_20_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184415808)))];
            fp16 var_5681_to_fp16 = const()[name = string("op_5681_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_337_cast_fp16 = layer_norm(axes = x_337_axes_0, beta = enc_layers_20_norm_feed_forward1_bias_to_fp16, epsilon = var_5681_to_fp16, gamma = enc_layers_20_norm_feed_forward1_weight_to_fp16, x = input_499_cast_fp16)[name = string("x_337_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_20_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184417920))), lut = tensor<fp16, [256, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185990848))))[name = string("blocks_20_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_221_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_20_ff1_1_weight_to_fp16_palettized, x = x_337_cast_fp16)[name = string("linear_221_cast_fp16")];
            tensor<fp16, [4096]> blocks_20_ff1_1_scale_to_fp16 = const()[name = string("blocks_20_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185995008)))];
            tensor<fp16, [1, 188, 4096]> input_501_cast_fp16 = mul(x = linear_221_cast_fp16, y = blocks_20_ff1_1_scale_to_fp16)[name = string("input_501_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_339_cast_fp16 = silu(x = input_501_cast_fp16)[name = string("x_339_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_20_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186003264))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187576192))))[name = string("blocks_20_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_222_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_20_ff1_2_weight_to_fp16_palettized, x = x_339_cast_fp16)[name = string("linear_222_cast_fp16")];
            tensor<fp16, [1024]> blocks_20_ff1_2_scale_to_fp16 = const()[name = string("blocks_20_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187577280)))];
            tensor<fp16, [1, 188, 1024]> var_5706_cast_fp16 = mul(x = linear_222_cast_fp16, y = blocks_20_ff1_2_scale_to_fp16)[name = string("op_5706_cast_fp16")];
            fp16 var_5707_to_fp16 = const()[name = string("op_5707_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5708_cast_fp16 = mul(x = var_5706_cast_fp16, y = var_5707_to_fp16)[name = string("op_5708_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_503_cast_fp16 = add(x = input_499_cast_fp16, y = var_5708_cast_fp16)[name = string("input_503_cast_fp16")];
            tensor<int32, [1]> x_341_axes_0 = const()[name = string("x_341_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_20_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_20_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187579392)))];
            tensor<fp16, [1024]> enc_layers_20_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_20_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187581504)))];
            tensor<fp16, [1, 188, 1024]> x_341_cast_fp16 = layer_norm(axes = x_341_axes_0, beta = enc_layers_20_norm_self_att_bias_to_fp16, epsilon = var_5681_to_fp16, gamma = enc_layers_20_norm_self_att_weight_to_fp16, x = input_503_cast_fp16)[name = string("x_341_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_20_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187583616))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188107968))))[name = string("blocks_20_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_223_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_20_q_weight_to_fp16_palettized, x = x_341_cast_fp16)[name = string("linear_223_cast_fp16")];
            tensor<fp16, [1024]> blocks_20_q_scale_to_fp16 = const()[name = string("blocks_20_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188110080)))];
            tensor<fp16, [1, 188, 1024]> var_5716_cast_fp16 = mul(x = linear_223_cast_fp16, y = blocks_20_q_scale_to_fp16)[name = string("op_5716_cast_fp16")];
            tensor<int32, [4]> var_5717 = const()[name = string("op_5717"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5718_cast_fp16 = reshape(shape = var_5717, x = var_5716_cast_fp16)[name = string("op_5718_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_20_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188112192))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188636544))))[name = string("blocks_20_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_224_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_20_k_weight_to_fp16_palettized, x = x_341_cast_fp16)[name = string("linear_224_cast_fp16")];
            tensor<fp16, [1024]> blocks_20_k_scale_to_fp16 = const()[name = string("blocks_20_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188638656)))];
            tensor<fp16, [1, 188, 1024]> var_5724_cast_fp16 = mul(x = linear_224_cast_fp16, y = blocks_20_k_scale_to_fp16)[name = string("op_5724_cast_fp16")];
            tensor<int32, [4]> var_5725 = const()[name = string("op_5725"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5726_cast_fp16 = reshape(shape = var_5725, x = var_5724_cast_fp16)[name = string("op_5726_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_20_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188640768))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189165120))))[name = string("blocks_20_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_225_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_20_v_weight_to_fp16_palettized, x = x_341_cast_fp16)[name = string("linear_225_cast_fp16")];
            tensor<fp16, [1024]> blocks_20_v_scale_to_fp16 = const()[name = string("blocks_20_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189167232)))];
            tensor<fp16, [1, 188, 1024]> var_5732_cast_fp16 = mul(x = linear_225_cast_fp16, y = blocks_20_v_scale_to_fp16)[name = string("op_5732_cast_fp16")];
            tensor<int32, [4]> var_5733 = const()[name = string("op_5733"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5734_cast_fp16 = reshape(shape = var_5733, x = var_5732_cast_fp16)[name = string("op_5734_cast_fp16")];
            tensor<int32, [4]> v_41_perm_0 = const()[name = string("v_41_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_124_to_fp16 = const()[name = string("const_124_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189169344)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_41_cast_fp16 = add(x = var_5718_cast_fp16, y = const_124_to_fp16)[name = string("q_u_41_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_125_to_fp16 = const()[name = string("const_125_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189171456)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_41_cast_fp16 = add(x = var_5718_cast_fp16, y = const_125_to_fp16)[name = string("q_v_41_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_20_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189173568))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189697920))))[name = string("blocks_20_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_226_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_20_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_226_cast_fp16")];
            tensor<fp16, [1024]> blocks_20_pos_scale_to_fp16 = const()[name = string("blocks_20_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189700032)))];
            tensor<fp16, [1, 375, 1024]> var_5746_cast_fp16 = mul(x = linear_226_cast_fp16, y = blocks_20_pos_scale_to_fp16)[name = string("op_5746_cast_fp16")];
            tensor<int32, [4]> var_5747 = const()[name = string("op_5747"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_41_cast_fp16 = reshape(shape = var_5747, x = var_5746_cast_fp16)[name = string("p_41_cast_fp16")];
            bool input_505_transpose_x_0 = const()[name = string("input_505_transpose_x_0"), val = bool(false)];
            bool input_505_transpose_y_0 = const()[name = string("input_505_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_272_perm_0 = const()[name = string("transpose_272_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_273_perm_0 = const()[name = string("transpose_273_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_273 = transpose(perm = transpose_273_perm_0, x = p_41_cast_fp16)[name = string("transpose_319")];
            tensor<fp16, [1, 8, 188, 128]> transpose_272 = transpose(perm = transpose_272_perm_0, x = q_v_41_cast_fp16)[name = string("transpose_320")];
            tensor<fp16, [1, 8, 188, 375]> input_505_cast_fp16 = matmul(transpose_x = input_505_transpose_x_0, transpose_y = input_505_transpose_y_0, x = transpose_272, y = transpose_273)[name = string("input_505_cast_fp16")];
            tensor<int32, [8]> var_5753_pad_0 = const()[name = string("op_5753_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_5753_mode_0 = const()[name = string("op_5753_mode_0"), val = string("constant")];
            fp16 const_32_to_fp16 = const()[name = string("const_32_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_5753_cast_fp16 = pad(constant_val = const_32_to_fp16, mode = var_5753_mode_0, pad = var_5753_pad_0, x = input_505_cast_fp16)[name = string("op_5753_cast_fp16")];
            tensor<int32, [4]> var_5754 = const()[name = string("op_5754"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_81_cast_fp16 = reshape(shape = var_5754, x = var_5753_cast_fp16)[name = string("scores_81_cast_fp16")];
            tensor<int32, [4]> var_5758_begin_0 = const()[name = string("op_5758_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_5758_end_0 = const()[name = string("op_5758_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_5758_end_mask_0 = const()[name = string("op_5758_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_5758_cast_fp16 = slice_by_index(begin = var_5758_begin_0, end = var_5758_end_0, end_mask = var_5758_end_mask_0, x = scores_81_cast_fp16)[name = string("op_5758_cast_fp16")];
            tensor<int32, [4]> var_5759 = const()[name = string("op_5759"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_5760_cast_fp16 = reshape(shape = var_5759, x = var_5758_cast_fp16)[name = string("op_5760_cast_fp16")];
            tensor<int32, [4]> bd_41_begin_0 = const()[name = string("bd_41_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_41_end_0 = const()[name = string("bd_41_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_41_end_mask_0 = const()[name = string("bd_41_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_41_cast_fp16 = slice_by_index(begin = bd_41_begin_0, end = bd_41_end_0, end_mask = bd_41_end_mask_0, x = var_5760_cast_fp16)[name = string("bd_41_cast_fp16")];
            bool var_5763_transpose_x_0 = const()[name = string("op_5763_transpose_x_0"), val = bool(false)];
            bool var_5763_transpose_y_0 = const()[name = string("op_5763_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_274_perm_0 = const()[name = string("transpose_274_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_275_perm_0 = const()[name = string("transpose_275_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_275 = transpose(perm = transpose_275_perm_0, x = var_5726_cast_fp16)[name = string("transpose_317")];
            tensor<fp16, [1, 8, 188, 128]> transpose_274 = transpose(perm = transpose_274_perm_0, x = q_u_41_cast_fp16)[name = string("transpose_318")];
            tensor<fp16, [1, 8, 188, 188]> var_5763_cast_fp16 = matmul(transpose_x = var_5763_transpose_x_0, transpose_y = var_5763_transpose_y_0, x = transpose_274, y = transpose_275)[name = string("op_5763_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5764_cast_fp16 = add(x = var_5763_cast_fp16, y = bd_41_cast_fp16)[name = string("op_5764_cast_fp16")];
            fp16 var_5765_to_fp16 = const()[name = string("op_5765_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_5766_cast_fp16 = mul(x = var_5764_cast_fp16, y = var_5765_to_fp16)[name = string("op_5766_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_83_cast_fp16 = add(x = var_5766_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_83_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5768_cast_fp16 = softmax(axis = var_5674, x = scores_83_cast_fp16)[name = string("op_5768_cast_fp16")];
            bool out_41_transpose_x_0 = const()[name = string("out_41_transpose_x_0"), val = bool(false)];
            bool out_41_transpose_y_0 = const()[name = string("out_41_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_41_cast_fp16 = transpose(perm = v_41_perm_0, x = var_5734_cast_fp16)[name = string("transpose_316")];
            tensor<fp16, [1, 8, 188, 128]> out_41_cast_fp16 = matmul(transpose_x = out_41_transpose_x_0, transpose_y = out_41_transpose_y_0, x = var_5768_cast_fp16, y = v_41_cast_fp16)[name = string("out_41_cast_fp16")];
            tensor<int32, [4]> var_5770_perm_0 = const()[name = string("op_5770_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5771 = const()[name = string("op_5771"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_5770_cast_fp16 = transpose(perm = var_5770_perm_0, x = out_41_cast_fp16)[name = string("transpose_315")];
            tensor<fp16, [1, 188, 1024]> x_343_cast_fp16 = reshape(shape = var_5771, x = var_5770_cast_fp16)[name = string("x_343_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_20_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189702144))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190226496))))[name = string("blocks_20_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_227_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_20_o_weight_to_fp16_palettized, x = x_343_cast_fp16)[name = string("linear_227_cast_fp16")];
            tensor<fp16, [1024]> blocks_20_o_scale_to_fp16 = const()[name = string("blocks_20_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190228608)))];
            tensor<fp16, [1, 188, 1024]> var_5777_cast_fp16 = mul(x = linear_227_cast_fp16, y = blocks_20_o_scale_to_fp16)[name = string("op_5777_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_507_cast_fp16 = add(x = input_503_cast_fp16, y = var_5777_cast_fp16)[name = string("input_507_cast_fp16")];
            tensor<int32, [1]> x_345_axes_0 = const()[name = string("x_345_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_20_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_20_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190230720)))];
            tensor<fp16, [1024]> enc_layers_20_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_20_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190232832)))];
            tensor<fp16, [1, 188, 1024]> x_345_cast_fp16 = layer_norm(axes = x_345_axes_0, beta = enc_layers_20_norm_conv_bias_to_fp16, epsilon = var_5681_to_fp16, gamma = enc_layers_20_norm_conv_weight_to_fp16, x = input_507_cast_fp16)[name = string("x_345_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_20_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190234944))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191283584))))[name = string("blocks_20_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_228_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_20_pw1_weight_to_fp16_palettized, x = x_345_cast_fp16)[name = string("linear_228_cast_fp16")];
            tensor<fp16, [2048]> blocks_20_pw1_scale_to_fp16 = const()[name = string("blocks_20_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191287744)))];
            tensor<fp16, [1, 188, 2048]> input_509_cast_fp16 = mul(x = linear_228_cast_fp16, y = blocks_20_pw1_scale_to_fp16)[name = string("input_509_cast_fp16")];
            int32 var_5786_split_num_splits_0 = const()[name = string("op_5786_split_num_splits_0"), val = int32(2)];
            int32 var_5786_split_axis_0 = const()[name = string("op_5786_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_5786_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_5786_split_cast_fp16_1 = split(axis = var_5786_split_axis_0, num_splits = var_5786_split_num_splits_0, x = input_509_cast_fp16)[name = string("op_5786_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_5786_split_1_sigmoid_cast_fp16 = sigmoid(x = var_5786_split_cast_fp16_1)[name = string("op_5786_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_5786_cast_fp16 = mul(x = var_5786_split_cast_fp16_0, y = var_5786_split_1_sigmoid_cast_fp16)[name = string("op_5786_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_81_cast_fp16 = mul(x = var_5786_cast_fp16, y = valid_cast_fp16)[name = string("h_81_cast_fp16")];
            tensor<int32, [3]> input_511_perm_0 = const()[name = string("input_511_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_513_pad_type_0 = const()[name = string("input_513_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_513_pad_0 = const()[name = string("input_513_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_513_groups_0 = const()[name = string("input_513_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_513_strides_0 = const()[name = string("input_513_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_513_dilations_0 = const()[name = string("input_513_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_76_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191291904))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191301184))))[name = string("const_76_to_fp16_quantized")];
            tensor<fp16, [1024]> const_77_to_fp16 = const()[name = string("const_77_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191303296)))];
            tensor<fp16, [1, 1024, 188]> input_511_cast_fp16 = transpose(perm = input_511_perm_0, x = h_81_cast_fp16)[name = string("transpose_314")];
            tensor<fp16, [1, 1024, 188]> input_515_cast_fp16 = conv(bias = const_77_to_fp16, dilations = input_513_dilations_0, groups = input_513_groups_0, pad = input_513_pad_0, pad_type = input_513_pad_type_0, strides = input_513_strides_0, weight = const_76_to_fp16_quantized, x = input_511_cast_fp16)[name = string("input_515_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_83_cast_fp16 = silu(x = input_515_cast_fp16)[name = string("h_83_cast_fp16")];
            tensor<int32, [3]> x_347_perm_0 = const()[name = string("x_347_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_20_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191305408))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191829760))))[name = string("blocks_20_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_347_cast_fp16 = transpose(perm = x_347_perm_0, x = h_83_cast_fp16)[name = string("transpose_313")];
            tensor<fp16, [1, 188, 1024]> linear_229_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_20_pw2_weight_to_fp16_palettized, x = x_347_cast_fp16)[name = string("linear_229_cast_fp16")];
            tensor<fp16, [1024]> blocks_20_pw2_scale_to_fp16 = const()[name = string("blocks_20_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191831872)))];
            tensor<fp16, [1, 188, 1024]> var_5801_cast_fp16 = mul(x = linear_229_cast_fp16, y = blocks_20_pw2_scale_to_fp16)[name = string("op_5801_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_517_cast_fp16 = add(x = input_507_cast_fp16, y = var_5801_cast_fp16)[name = string("input_517_cast_fp16")];
            tensor<int32, [1]> x_349_axes_0 = const()[name = string("x_349_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_20_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_20_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191833984)))];
            tensor<fp16, [1024]> enc_layers_20_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_20_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191836096)))];
            tensor<fp16, [1, 188, 1024]> x_349_cast_fp16 = layer_norm(axes = x_349_axes_0, beta = enc_layers_20_norm_feed_forward2_bias_to_fp16, epsilon = var_5681_to_fp16, gamma = enc_layers_20_norm_feed_forward2_weight_to_fp16, x = input_517_cast_fp16)[name = string("x_349_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_20_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191838208))), lut = tensor<fp16, [256, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193411136))))[name = string("blocks_20_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_230_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_20_ff2_1_weight_to_fp16_palettized, x = x_349_cast_fp16)[name = string("linear_230_cast_fp16")];
            tensor<fp16, [4096]> blocks_20_ff2_1_scale_to_fp16 = const()[name = string("blocks_20_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193415296)))];
            tensor<fp16, [1, 188, 4096]> input_519_cast_fp16 = mul(x = linear_230_cast_fp16, y = blocks_20_ff2_1_scale_to_fp16)[name = string("input_519_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_351_cast_fp16 = silu(x = input_519_cast_fp16)[name = string("x_351_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_20_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint3, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193423552))), lut = tensor<fp16, [64, 1, 8, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194996480))))[name = string("blocks_20_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_231_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_20_ff2_2_weight_to_fp16_palettized, x = x_351_cast_fp16)[name = string("linear_231_cast_fp16")];
            tensor<fp16, [1024]> blocks_20_ff2_2_scale_to_fp16 = const()[name = string("blocks_20_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194997568)))];
            tensor<fp16, [1, 188, 1024]> var_5815_cast_fp16 = mul(x = linear_231_cast_fp16, y = blocks_20_ff2_2_scale_to_fp16)[name = string("op_5815_cast_fp16")];
            fp16 var_5816_to_fp16 = const()[name = string("op_5816_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5817_cast_fp16 = mul(x = var_5815_cast_fp16, y = var_5816_to_fp16)[name = string("op_5817_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_521_cast_fp16 = add(x = input_517_cast_fp16, y = var_5817_cast_fp16)[name = string("input_521_cast_fp16")];
            tensor<int32, [1]> input_523_axes_0 = const()[name = string("input_523_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_20_norm_out_weight_to_fp16 = const()[name = string("enc_layers_20_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194999680)))];
            tensor<fp16, [1024]> enc_layers_20_norm_out_bias_to_fp16 = const()[name = string("enc_layers_20_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195001792)))];
            tensor<fp16, [1, 188, 1024]> input_523_cast_fp16 = layer_norm(axes = input_523_axes_0, beta = enc_layers_20_norm_out_bias_to_fp16, epsilon = var_5681_to_fp16, gamma = enc_layers_20_norm_out_weight_to_fp16, x = input_521_cast_fp16)[name = string("input_523_cast_fp16")];
            int32 var_5831 = const()[name = string("op_5831"), val = int32(-1)];
            tensor<int32, [1]> x_353_axes_0 = const()[name = string("x_353_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_21_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_21_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195003904)))];
            tensor<fp16, [1024]> enc_layers_21_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_21_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195006016)))];
            fp16 var_5838_to_fp16 = const()[name = string("op_5838_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_353_cast_fp16 = layer_norm(axes = x_353_axes_0, beta = enc_layers_21_norm_feed_forward1_bias_to_fp16, epsilon = var_5838_to_fp16, gamma = enc_layers_21_norm_feed_forward1_weight_to_fp16, x = input_523_cast_fp16)[name = string("x_353_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_21_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195008128))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197105344))))[name = string("blocks_21_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_232_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_21_ff1_1_weight_to_fp16_palettized, x = x_353_cast_fp16)[name = string("linear_232_cast_fp16")];
            tensor<fp16, [4096]> blocks_21_ff1_1_scale_to_fp16 = const()[name = string("blocks_21_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197113600)))];
            tensor<fp16, [1, 188, 4096]> input_525_cast_fp16 = mul(x = linear_232_cast_fp16, y = blocks_21_ff1_1_scale_to_fp16)[name = string("input_525_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_355_cast_fp16 = silu(x = input_525_cast_fp16)[name = string("x_355_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_21_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197121856))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199219072))))[name = string("blocks_21_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_233_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_21_ff1_2_weight_to_fp16_palettized, x = x_355_cast_fp16)[name = string("linear_233_cast_fp16")];
            tensor<fp16, [1024]> blocks_21_ff1_2_scale_to_fp16 = const()[name = string("blocks_21_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199221184)))];
            tensor<fp16, [1, 188, 1024]> var_5863_cast_fp16 = mul(x = linear_233_cast_fp16, y = blocks_21_ff1_2_scale_to_fp16)[name = string("op_5863_cast_fp16")];
            fp16 var_5864_to_fp16 = const()[name = string("op_5864_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5865_cast_fp16 = mul(x = var_5863_cast_fp16, y = var_5864_to_fp16)[name = string("op_5865_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_527_cast_fp16 = add(x = input_523_cast_fp16, y = var_5865_cast_fp16)[name = string("input_527_cast_fp16")];
            tensor<int32, [1]> x_357_axes_0 = const()[name = string("x_357_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_21_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_21_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199223296)))];
            tensor<fp16, [1024]> enc_layers_21_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_21_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199225408)))];
            tensor<fp16, [1, 188, 1024]> x_357_cast_fp16 = layer_norm(axes = x_357_axes_0, beta = enc_layers_21_norm_self_att_bias_to_fp16, epsilon = var_5838_to_fp16, gamma = enc_layers_21_norm_self_att_weight_to_fp16, x = input_527_cast_fp16)[name = string("x_357_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_21_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199227520))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199751872))))[name = string("blocks_21_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_234_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_21_q_weight_to_fp16_palettized, x = x_357_cast_fp16)[name = string("linear_234_cast_fp16")];
            tensor<fp16, [1024]> blocks_21_q_scale_to_fp16 = const()[name = string("blocks_21_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199753984)))];
            tensor<fp16, [1, 188, 1024]> var_5873_cast_fp16 = mul(x = linear_234_cast_fp16, y = blocks_21_q_scale_to_fp16)[name = string("op_5873_cast_fp16")];
            tensor<int32, [4]> var_5874 = const()[name = string("op_5874"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5875_cast_fp16 = reshape(shape = var_5874, x = var_5873_cast_fp16)[name = string("op_5875_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_21_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199756096))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200280448))))[name = string("blocks_21_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_235_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_21_k_weight_to_fp16_palettized, x = x_357_cast_fp16)[name = string("linear_235_cast_fp16")];
            tensor<fp16, [1024]> blocks_21_k_scale_to_fp16 = const()[name = string("blocks_21_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200282560)))];
            tensor<fp16, [1, 188, 1024]> var_5881_cast_fp16 = mul(x = linear_235_cast_fp16, y = blocks_21_k_scale_to_fp16)[name = string("op_5881_cast_fp16")];
            tensor<int32, [4]> var_5882 = const()[name = string("op_5882"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5883_cast_fp16 = reshape(shape = var_5882, x = var_5881_cast_fp16)[name = string("op_5883_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_21_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200284672))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200809024))))[name = string("blocks_21_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_236_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_21_v_weight_to_fp16_palettized, x = x_357_cast_fp16)[name = string("linear_236_cast_fp16")];
            tensor<fp16, [1024]> blocks_21_v_scale_to_fp16 = const()[name = string("blocks_21_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200811136)))];
            tensor<fp16, [1, 188, 1024]> var_5889_cast_fp16 = mul(x = linear_236_cast_fp16, y = blocks_21_v_scale_to_fp16)[name = string("op_5889_cast_fp16")];
            tensor<int32, [4]> var_5890 = const()[name = string("op_5890"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5891_cast_fp16 = reshape(shape = var_5890, x = var_5889_cast_fp16)[name = string("op_5891_cast_fp16")];
            tensor<int32, [4]> v_43_perm_0 = const()[name = string("v_43_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_126_to_fp16 = const()[name = string("const_126_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200813248)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_43_cast_fp16 = add(x = var_5875_cast_fp16, y = const_126_to_fp16)[name = string("q_u_43_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_127_to_fp16 = const()[name = string("const_127_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200815360)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_43_cast_fp16 = add(x = var_5875_cast_fp16, y = const_127_to_fp16)[name = string("q_v_43_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_21_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200817472))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201341824))))[name = string("blocks_21_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_237_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_21_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_237_cast_fp16")];
            tensor<fp16, [1024]> blocks_21_pos_scale_to_fp16 = const()[name = string("blocks_21_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201343936)))];
            tensor<fp16, [1, 375, 1024]> var_5903_cast_fp16 = mul(x = linear_237_cast_fp16, y = blocks_21_pos_scale_to_fp16)[name = string("op_5903_cast_fp16")];
            tensor<int32, [4]> var_5904 = const()[name = string("op_5904"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_43_cast_fp16 = reshape(shape = var_5904, x = var_5903_cast_fp16)[name = string("p_43_cast_fp16")];
            bool input_529_transpose_x_0 = const()[name = string("input_529_transpose_x_0"), val = bool(false)];
            bool input_529_transpose_y_0 = const()[name = string("input_529_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_276_perm_0 = const()[name = string("transpose_276_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_277_perm_0 = const()[name = string("transpose_277_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_277 = transpose(perm = transpose_277_perm_0, x = p_43_cast_fp16)[name = string("transpose_311")];
            tensor<fp16, [1, 8, 188, 128]> transpose_276 = transpose(perm = transpose_276_perm_0, x = q_v_43_cast_fp16)[name = string("transpose_312")];
            tensor<fp16, [1, 8, 188, 375]> input_529_cast_fp16 = matmul(transpose_x = input_529_transpose_x_0, transpose_y = input_529_transpose_y_0, x = transpose_276, y = transpose_277)[name = string("input_529_cast_fp16")];
            tensor<int32, [8]> var_5910_pad_0 = const()[name = string("op_5910_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_5910_mode_0 = const()[name = string("op_5910_mode_0"), val = string("constant")];
            fp16 const_33_to_fp16 = const()[name = string("const_33_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_5910_cast_fp16 = pad(constant_val = const_33_to_fp16, mode = var_5910_mode_0, pad = var_5910_pad_0, x = input_529_cast_fp16)[name = string("op_5910_cast_fp16")];
            tensor<int32, [4]> var_5911 = const()[name = string("op_5911"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_85_cast_fp16 = reshape(shape = var_5911, x = var_5910_cast_fp16)[name = string("scores_85_cast_fp16")];
            tensor<int32, [4]> var_5915_begin_0 = const()[name = string("op_5915_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_5915_end_0 = const()[name = string("op_5915_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_5915_end_mask_0 = const()[name = string("op_5915_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_5915_cast_fp16 = slice_by_index(begin = var_5915_begin_0, end = var_5915_end_0, end_mask = var_5915_end_mask_0, x = scores_85_cast_fp16)[name = string("op_5915_cast_fp16")];
            tensor<int32, [4]> var_5916 = const()[name = string("op_5916"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_5917_cast_fp16 = reshape(shape = var_5916, x = var_5915_cast_fp16)[name = string("op_5917_cast_fp16")];
            tensor<int32, [4]> bd_43_begin_0 = const()[name = string("bd_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_43_end_0 = const()[name = string("bd_43_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_43_end_mask_0 = const()[name = string("bd_43_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_43_cast_fp16 = slice_by_index(begin = bd_43_begin_0, end = bd_43_end_0, end_mask = bd_43_end_mask_0, x = var_5917_cast_fp16)[name = string("bd_43_cast_fp16")];
            bool var_5920_transpose_x_0 = const()[name = string("op_5920_transpose_x_0"), val = bool(false)];
            bool var_5920_transpose_y_0 = const()[name = string("op_5920_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_278_perm_0 = const()[name = string("transpose_278_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_279_perm_0 = const()[name = string("transpose_279_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_279 = transpose(perm = transpose_279_perm_0, x = var_5883_cast_fp16)[name = string("transpose_309")];
            tensor<fp16, [1, 8, 188, 128]> transpose_278 = transpose(perm = transpose_278_perm_0, x = q_u_43_cast_fp16)[name = string("transpose_310")];
            tensor<fp16, [1, 8, 188, 188]> var_5920_cast_fp16 = matmul(transpose_x = var_5920_transpose_x_0, transpose_y = var_5920_transpose_y_0, x = transpose_278, y = transpose_279)[name = string("op_5920_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5921_cast_fp16 = add(x = var_5920_cast_fp16, y = bd_43_cast_fp16)[name = string("op_5921_cast_fp16")];
            fp16 var_5922_to_fp16 = const()[name = string("op_5922_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_5923_cast_fp16 = mul(x = var_5921_cast_fp16, y = var_5922_to_fp16)[name = string("op_5923_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_87_cast_fp16 = add(x = var_5923_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_87_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5925_cast_fp16 = softmax(axis = var_5831, x = scores_87_cast_fp16)[name = string("op_5925_cast_fp16")];
            bool out_43_transpose_x_0 = const()[name = string("out_43_transpose_x_0"), val = bool(false)];
            bool out_43_transpose_y_0 = const()[name = string("out_43_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_43_cast_fp16 = transpose(perm = v_43_perm_0, x = var_5891_cast_fp16)[name = string("transpose_308")];
            tensor<fp16, [1, 8, 188, 128]> out_43_cast_fp16 = matmul(transpose_x = out_43_transpose_x_0, transpose_y = out_43_transpose_y_0, x = var_5925_cast_fp16, y = v_43_cast_fp16)[name = string("out_43_cast_fp16")];
            tensor<int32, [4]> var_5927_perm_0 = const()[name = string("op_5927_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5928 = const()[name = string("op_5928"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_5927_cast_fp16 = transpose(perm = var_5927_perm_0, x = out_43_cast_fp16)[name = string("transpose_307")];
            tensor<fp16, [1, 188, 1024]> x_359_cast_fp16 = reshape(shape = var_5928, x = var_5927_cast_fp16)[name = string("x_359_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_21_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201346048))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201870400))))[name = string("blocks_21_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_238_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_21_o_weight_to_fp16_palettized, x = x_359_cast_fp16)[name = string("linear_238_cast_fp16")];
            tensor<fp16, [1024]> blocks_21_o_scale_to_fp16 = const()[name = string("blocks_21_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201872512)))];
            tensor<fp16, [1, 188, 1024]> var_5934_cast_fp16 = mul(x = linear_238_cast_fp16, y = blocks_21_o_scale_to_fp16)[name = string("op_5934_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_531_cast_fp16 = add(x = input_527_cast_fp16, y = var_5934_cast_fp16)[name = string("input_531_cast_fp16")];
            tensor<int32, [1]> x_361_axes_0 = const()[name = string("x_361_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_21_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_21_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201874624)))];
            tensor<fp16, [1024]> enc_layers_21_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_21_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201876736)))];
            tensor<fp16, [1, 188, 1024]> x_361_cast_fp16 = layer_norm(axes = x_361_axes_0, beta = enc_layers_21_norm_conv_bias_to_fp16, epsilon = var_5838_to_fp16, gamma = enc_layers_21_norm_conv_weight_to_fp16, x = input_531_cast_fp16)[name = string("x_361_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_21_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201878848))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202927488))))[name = string("blocks_21_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_239_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_21_pw1_weight_to_fp16_palettized, x = x_361_cast_fp16)[name = string("linear_239_cast_fp16")];
            tensor<fp16, [2048]> blocks_21_pw1_scale_to_fp16 = const()[name = string("blocks_21_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202931648)))];
            tensor<fp16, [1, 188, 2048]> input_533_cast_fp16 = mul(x = linear_239_cast_fp16, y = blocks_21_pw1_scale_to_fp16)[name = string("input_533_cast_fp16")];
            int32 var_5943_split_num_splits_0 = const()[name = string("op_5943_split_num_splits_0"), val = int32(2)];
            int32 var_5943_split_axis_0 = const()[name = string("op_5943_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_5943_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_5943_split_cast_fp16_1 = split(axis = var_5943_split_axis_0, num_splits = var_5943_split_num_splits_0, x = input_533_cast_fp16)[name = string("op_5943_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_5943_split_1_sigmoid_cast_fp16 = sigmoid(x = var_5943_split_cast_fp16_1)[name = string("op_5943_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_5943_cast_fp16 = mul(x = var_5943_split_cast_fp16_0, y = var_5943_split_1_sigmoid_cast_fp16)[name = string("op_5943_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_85_cast_fp16 = mul(x = var_5943_cast_fp16, y = valid_cast_fp16)[name = string("h_85_cast_fp16")];
            tensor<int32, [3]> input_535_perm_0 = const()[name = string("input_535_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_537_pad_type_0 = const()[name = string("input_537_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_537_pad_0 = const()[name = string("input_537_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_537_groups_0 = const()[name = string("input_537_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_537_strides_0 = const()[name = string("input_537_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_537_dilations_0 = const()[name = string("input_537_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_78_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202935808))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202945088))))[name = string("const_78_to_fp16_quantized")];
            tensor<fp16, [1024]> const_79_to_fp16 = const()[name = string("const_79_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202947200)))];
            tensor<fp16, [1, 1024, 188]> input_535_cast_fp16 = transpose(perm = input_535_perm_0, x = h_85_cast_fp16)[name = string("transpose_306")];
            tensor<fp16, [1, 1024, 188]> input_539_cast_fp16 = conv(bias = const_79_to_fp16, dilations = input_537_dilations_0, groups = input_537_groups_0, pad = input_537_pad_0, pad_type = input_537_pad_type_0, strides = input_537_strides_0, weight = const_78_to_fp16_quantized, x = input_535_cast_fp16)[name = string("input_539_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_87_cast_fp16 = silu(x = input_539_cast_fp16)[name = string("h_87_cast_fp16")];
            tensor<int32, [3]> x_363_perm_0 = const()[name = string("x_363_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_21_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202949312))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203473664))))[name = string("blocks_21_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_363_cast_fp16 = transpose(perm = x_363_perm_0, x = h_87_cast_fp16)[name = string("transpose_305")];
            tensor<fp16, [1, 188, 1024]> linear_240_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_21_pw2_weight_to_fp16_palettized, x = x_363_cast_fp16)[name = string("linear_240_cast_fp16")];
            tensor<fp16, [1024]> blocks_21_pw2_scale_to_fp16 = const()[name = string("blocks_21_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203475776)))];
            tensor<fp16, [1, 188, 1024]> var_5958_cast_fp16 = mul(x = linear_240_cast_fp16, y = blocks_21_pw2_scale_to_fp16)[name = string("op_5958_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_541_cast_fp16 = add(x = input_531_cast_fp16, y = var_5958_cast_fp16)[name = string("input_541_cast_fp16")];
            tensor<int32, [1]> x_365_axes_0 = const()[name = string("x_365_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_21_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_21_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203477888)))];
            tensor<fp16, [1024]> enc_layers_21_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_21_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203480000)))];
            tensor<fp16, [1, 188, 1024]> x_365_cast_fp16 = layer_norm(axes = x_365_axes_0, beta = enc_layers_21_norm_feed_forward2_bias_to_fp16, epsilon = var_5838_to_fp16, gamma = enc_layers_21_norm_feed_forward2_weight_to_fp16, x = input_541_cast_fp16)[name = string("x_365_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_21_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203482112))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205579328))))[name = string("blocks_21_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_241_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_21_ff2_1_weight_to_fp16_palettized, x = x_365_cast_fp16)[name = string("linear_241_cast_fp16")];
            tensor<fp16, [4096]> blocks_21_ff2_1_scale_to_fp16 = const()[name = string("blocks_21_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205587584)))];
            tensor<fp16, [1, 188, 4096]> input_543_cast_fp16 = mul(x = linear_241_cast_fp16, y = blocks_21_ff2_1_scale_to_fp16)[name = string("input_543_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_367_cast_fp16 = silu(x = input_543_cast_fp16)[name = string("x_367_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_21_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205595840))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207693056))))[name = string("blocks_21_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_242_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_21_ff2_2_weight_to_fp16_palettized, x = x_367_cast_fp16)[name = string("linear_242_cast_fp16")];
            tensor<fp16, [1024]> blocks_21_ff2_2_scale_to_fp16 = const()[name = string("blocks_21_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207695168)))];
            tensor<fp16, [1, 188, 1024]> var_5972_cast_fp16 = mul(x = linear_242_cast_fp16, y = blocks_21_ff2_2_scale_to_fp16)[name = string("op_5972_cast_fp16")];
            fp16 var_5973_to_fp16 = const()[name = string("op_5973_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5974_cast_fp16 = mul(x = var_5972_cast_fp16, y = var_5973_to_fp16)[name = string("op_5974_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_545_cast_fp16 = add(x = input_541_cast_fp16, y = var_5974_cast_fp16)[name = string("input_545_cast_fp16")];
            tensor<int32, [1]> input_547_axes_0 = const()[name = string("input_547_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_21_norm_out_weight_to_fp16 = const()[name = string("enc_layers_21_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207697280)))];
            tensor<fp16, [1024]> enc_layers_21_norm_out_bias_to_fp16 = const()[name = string("enc_layers_21_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207699392)))];
            tensor<fp16, [1, 188, 1024]> input_547_cast_fp16 = layer_norm(axes = input_547_axes_0, beta = enc_layers_21_norm_out_bias_to_fp16, epsilon = var_5838_to_fp16, gamma = enc_layers_21_norm_out_weight_to_fp16, x = input_545_cast_fp16)[name = string("input_547_cast_fp16")];
            int32 var_5988 = const()[name = string("op_5988"), val = int32(-1)];
            tensor<int32, [1]> x_369_axes_0 = const()[name = string("x_369_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_22_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_22_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207701504)))];
            tensor<fp16, [1024]> enc_layers_22_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_22_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207703616)))];
            fp16 var_5995_to_fp16 = const()[name = string("op_5995_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_369_cast_fp16 = layer_norm(axes = x_369_axes_0, beta = enc_layers_22_norm_feed_forward1_bias_to_fp16, epsilon = var_5995_to_fp16, gamma = enc_layers_22_norm_feed_forward1_weight_to_fp16, x = input_547_cast_fp16)[name = string("x_369_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_22_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207705728))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209802944))))[name = string("blocks_22_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_243_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_22_ff1_1_weight_to_fp16_palettized, x = x_369_cast_fp16)[name = string("linear_243_cast_fp16")];
            tensor<fp16, [4096]> blocks_22_ff1_1_scale_to_fp16 = const()[name = string("blocks_22_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209811200)))];
            tensor<fp16, [1, 188, 4096]> input_549_cast_fp16 = mul(x = linear_243_cast_fp16, y = blocks_22_ff1_1_scale_to_fp16)[name = string("input_549_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_371_cast_fp16 = silu(x = input_549_cast_fp16)[name = string("x_371_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_22_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209819456))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211916672))))[name = string("blocks_22_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_244_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_22_ff1_2_weight_to_fp16_palettized, x = x_371_cast_fp16)[name = string("linear_244_cast_fp16")];
            tensor<fp16, [1024]> blocks_22_ff1_2_scale_to_fp16 = const()[name = string("blocks_22_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211918784)))];
            tensor<fp16, [1, 188, 1024]> var_6020_cast_fp16 = mul(x = linear_244_cast_fp16, y = blocks_22_ff1_2_scale_to_fp16)[name = string("op_6020_cast_fp16")];
            fp16 var_6021_to_fp16 = const()[name = string("op_6021_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_6022_cast_fp16 = mul(x = var_6020_cast_fp16, y = var_6021_to_fp16)[name = string("op_6022_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_551_cast_fp16 = add(x = input_547_cast_fp16, y = var_6022_cast_fp16)[name = string("input_551_cast_fp16")];
            tensor<int32, [1]> x_373_axes_0 = const()[name = string("x_373_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_22_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_22_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211920896)))];
            tensor<fp16, [1024]> enc_layers_22_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_22_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211923008)))];
            tensor<fp16, [1, 188, 1024]> x_373_cast_fp16 = layer_norm(axes = x_373_axes_0, beta = enc_layers_22_norm_self_att_bias_to_fp16, epsilon = var_5995_to_fp16, gamma = enc_layers_22_norm_self_att_weight_to_fp16, x = input_551_cast_fp16)[name = string("x_373_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_22_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211925120))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212449472))))[name = string("blocks_22_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_245_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_22_q_weight_to_fp16_palettized, x = x_373_cast_fp16)[name = string("linear_245_cast_fp16")];
            tensor<fp16, [1024]> blocks_22_q_scale_to_fp16 = const()[name = string("blocks_22_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212451584)))];
            tensor<fp16, [1, 188, 1024]> var_6030_cast_fp16 = mul(x = linear_245_cast_fp16, y = blocks_22_q_scale_to_fp16)[name = string("op_6030_cast_fp16")];
            tensor<int32, [4]> var_6031 = const()[name = string("op_6031"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_6032_cast_fp16 = reshape(shape = var_6031, x = var_6030_cast_fp16)[name = string("op_6032_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_22_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212453696))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212978048))))[name = string("blocks_22_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_246_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_22_k_weight_to_fp16_palettized, x = x_373_cast_fp16)[name = string("linear_246_cast_fp16")];
            tensor<fp16, [1024]> blocks_22_k_scale_to_fp16 = const()[name = string("blocks_22_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212980160)))];
            tensor<fp16, [1, 188, 1024]> var_6038_cast_fp16 = mul(x = linear_246_cast_fp16, y = blocks_22_k_scale_to_fp16)[name = string("op_6038_cast_fp16")];
            tensor<int32, [4]> var_6039 = const()[name = string("op_6039"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_6040_cast_fp16 = reshape(shape = var_6039, x = var_6038_cast_fp16)[name = string("op_6040_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_22_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212982272))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213506624))))[name = string("blocks_22_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_247_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_22_v_weight_to_fp16_palettized, x = x_373_cast_fp16)[name = string("linear_247_cast_fp16")];
            tensor<fp16, [1024]> blocks_22_v_scale_to_fp16 = const()[name = string("blocks_22_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213508736)))];
            tensor<fp16, [1, 188, 1024]> var_6046_cast_fp16 = mul(x = linear_247_cast_fp16, y = blocks_22_v_scale_to_fp16)[name = string("op_6046_cast_fp16")];
            tensor<int32, [4]> var_6047 = const()[name = string("op_6047"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_6048_cast_fp16 = reshape(shape = var_6047, x = var_6046_cast_fp16)[name = string("op_6048_cast_fp16")];
            tensor<int32, [4]> v_45_perm_0 = const()[name = string("v_45_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_128_to_fp16 = const()[name = string("const_128_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213510848)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_45_cast_fp16 = add(x = var_6032_cast_fp16, y = const_128_to_fp16)[name = string("q_u_45_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_129_to_fp16 = const()[name = string("const_129_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213512960)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_45_cast_fp16 = add(x = var_6032_cast_fp16, y = const_129_to_fp16)[name = string("q_v_45_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_22_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213515072))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214039424))))[name = string("blocks_22_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_248_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_22_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_248_cast_fp16")];
            tensor<fp16, [1024]> blocks_22_pos_scale_to_fp16 = const()[name = string("blocks_22_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214041536)))];
            tensor<fp16, [1, 375, 1024]> var_6060_cast_fp16 = mul(x = linear_248_cast_fp16, y = blocks_22_pos_scale_to_fp16)[name = string("op_6060_cast_fp16")];
            tensor<int32, [4]> var_6061 = const()[name = string("op_6061"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_45_cast_fp16 = reshape(shape = var_6061, x = var_6060_cast_fp16)[name = string("p_45_cast_fp16")];
            bool input_553_transpose_x_0 = const()[name = string("input_553_transpose_x_0"), val = bool(false)];
            bool input_553_transpose_y_0 = const()[name = string("input_553_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_280_perm_0 = const()[name = string("transpose_280_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_281_perm_0 = const()[name = string("transpose_281_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_281 = transpose(perm = transpose_281_perm_0, x = p_45_cast_fp16)[name = string("transpose_303")];
            tensor<fp16, [1, 8, 188, 128]> transpose_280 = transpose(perm = transpose_280_perm_0, x = q_v_45_cast_fp16)[name = string("transpose_304")];
            tensor<fp16, [1, 8, 188, 375]> input_553_cast_fp16 = matmul(transpose_x = input_553_transpose_x_0, transpose_y = input_553_transpose_y_0, x = transpose_280, y = transpose_281)[name = string("input_553_cast_fp16")];
            tensor<int32, [8]> var_6067_pad_0 = const()[name = string("op_6067_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_6067_mode_0 = const()[name = string("op_6067_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]> var_6067_cast_fp16 = pad(constant_val = const_34_to_fp16, mode = var_6067_mode_0, pad = var_6067_pad_0, x = input_553_cast_fp16)[name = string("op_6067_cast_fp16")];
            tensor<int32, [4]> var_6068 = const()[name = string("op_6068"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_89_cast_fp16 = reshape(shape = var_6068, x = var_6067_cast_fp16)[name = string("scores_89_cast_fp16")];
            tensor<int32, [4]> var_6072_begin_0 = const()[name = string("op_6072_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_6072_end_0 = const()[name = string("op_6072_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_6072_end_mask_0 = const()[name = string("op_6072_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_6072_cast_fp16 = slice_by_index(begin = var_6072_begin_0, end = var_6072_end_0, end_mask = var_6072_end_mask_0, x = scores_89_cast_fp16)[name = string("op_6072_cast_fp16")];
            tensor<int32, [4]> var_6073 = const()[name = string("op_6073"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_6074_cast_fp16 = reshape(shape = var_6073, x = var_6072_cast_fp16)[name = string("op_6074_cast_fp16")];
            tensor<int32, [4]> bd_45_begin_0 = const()[name = string("bd_45_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_45_end_0 = const()[name = string("bd_45_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_45_end_mask_0 = const()[name = string("bd_45_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_45_cast_fp16 = slice_by_index(begin = bd_45_begin_0, end = bd_45_end_0, end_mask = bd_45_end_mask_0, x = var_6074_cast_fp16)[name = string("bd_45_cast_fp16")];
            bool var_6077_transpose_x_0 = const()[name = string("op_6077_transpose_x_0"), val = bool(false)];
            bool var_6077_transpose_y_0 = const()[name = string("op_6077_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_282_perm_0 = const()[name = string("transpose_282_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_283_perm_0 = const()[name = string("transpose_283_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_283 = transpose(perm = transpose_283_perm_0, x = var_6040_cast_fp16)[name = string("transpose_301")];
            tensor<fp16, [1, 8, 188, 128]> transpose_282 = transpose(perm = transpose_282_perm_0, x = q_u_45_cast_fp16)[name = string("transpose_302")];
            tensor<fp16, [1, 8, 188, 188]> var_6077_cast_fp16 = matmul(transpose_x = var_6077_transpose_x_0, transpose_y = var_6077_transpose_y_0, x = transpose_282, y = transpose_283)[name = string("op_6077_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_6078_cast_fp16 = add(x = var_6077_cast_fp16, y = bd_45_cast_fp16)[name = string("op_6078_cast_fp16")];
            fp16 var_6079_to_fp16 = const()[name = string("op_6079_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_6080_cast_fp16 = mul(x = var_6078_cast_fp16, y = var_6079_to_fp16)[name = string("op_6080_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_91_cast_fp16 = add(x = var_6080_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_91_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_6082_cast_fp16 = softmax(axis = var_5988, x = scores_91_cast_fp16)[name = string("op_6082_cast_fp16")];
            bool out_45_transpose_x_0 = const()[name = string("out_45_transpose_x_0"), val = bool(false)];
            bool out_45_transpose_y_0 = const()[name = string("out_45_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_45_cast_fp16 = transpose(perm = v_45_perm_0, x = var_6048_cast_fp16)[name = string("transpose_300")];
            tensor<fp16, [1, 8, 188, 128]> out_45_cast_fp16 = matmul(transpose_x = out_45_transpose_x_0, transpose_y = out_45_transpose_y_0, x = var_6082_cast_fp16, y = v_45_cast_fp16)[name = string("out_45_cast_fp16")];
            tensor<int32, [4]> var_6084_perm_0 = const()[name = string("op_6084_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_6085 = const()[name = string("op_6085"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_6084_cast_fp16 = transpose(perm = var_6084_perm_0, x = out_45_cast_fp16)[name = string("transpose_299")];
            tensor<fp16, [1, 188, 1024]> x_375_cast_fp16 = reshape(shape = var_6085, x = var_6084_cast_fp16)[name = string("x_375_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_22_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214043648))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214568000))))[name = string("blocks_22_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_249_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_22_o_weight_to_fp16_palettized, x = x_375_cast_fp16)[name = string("linear_249_cast_fp16")];
            tensor<fp16, [1024]> blocks_22_o_scale_to_fp16 = const()[name = string("blocks_22_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214570112)))];
            tensor<fp16, [1, 188, 1024]> var_6091_cast_fp16 = mul(x = linear_249_cast_fp16, y = blocks_22_o_scale_to_fp16)[name = string("op_6091_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_555_cast_fp16 = add(x = input_551_cast_fp16, y = var_6091_cast_fp16)[name = string("input_555_cast_fp16")];
            tensor<int32, [1]> x_377_axes_0 = const()[name = string("x_377_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_22_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_22_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214572224)))];
            tensor<fp16, [1024]> enc_layers_22_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_22_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214574336)))];
            tensor<fp16, [1, 188, 1024]> x_377_cast_fp16 = layer_norm(axes = x_377_axes_0, beta = enc_layers_22_norm_conv_bias_to_fp16, epsilon = var_5995_to_fp16, gamma = enc_layers_22_norm_conv_weight_to_fp16, x = input_555_cast_fp16)[name = string("x_377_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_22_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214576448))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215625088))))[name = string("blocks_22_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_250_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_22_pw1_weight_to_fp16_palettized, x = x_377_cast_fp16)[name = string("linear_250_cast_fp16")];
            tensor<fp16, [2048]> blocks_22_pw1_scale_to_fp16 = const()[name = string("blocks_22_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215629248)))];
            tensor<fp16, [1, 188, 2048]> input_557_cast_fp16 = mul(x = linear_250_cast_fp16, y = blocks_22_pw1_scale_to_fp16)[name = string("input_557_cast_fp16")];
            int32 var_6100_split_num_splits_0 = const()[name = string("op_6100_split_num_splits_0"), val = int32(2)];
            int32 var_6100_split_axis_0 = const()[name = string("op_6100_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_6100_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_6100_split_cast_fp16_1 = split(axis = var_6100_split_axis_0, num_splits = var_6100_split_num_splits_0, x = input_557_cast_fp16)[name = string("op_6100_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_6100_split_1_sigmoid_cast_fp16 = sigmoid(x = var_6100_split_cast_fp16_1)[name = string("op_6100_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_6100_cast_fp16 = mul(x = var_6100_split_cast_fp16_0, y = var_6100_split_1_sigmoid_cast_fp16)[name = string("op_6100_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_89_cast_fp16 = mul(x = var_6100_cast_fp16, y = valid_cast_fp16)[name = string("h_89_cast_fp16")];
            tensor<int32, [3]> input_559_perm_0 = const()[name = string("input_559_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_561_pad_type_0 = const()[name = string("input_561_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_561_pad_0 = const()[name = string("input_561_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_561_groups_0 = const()[name = string("input_561_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_561_strides_0 = const()[name = string("input_561_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_561_dilations_0 = const()[name = string("input_561_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_80_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215633408))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215642688))))[name = string("const_80_to_fp16_quantized")];
            tensor<fp16, [1024]> const_81_to_fp16 = const()[name = string("const_81_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215644800)))];
            tensor<fp16, [1, 1024, 188]> input_559_cast_fp16 = transpose(perm = input_559_perm_0, x = h_89_cast_fp16)[name = string("transpose_298")];
            tensor<fp16, [1, 1024, 188]> input_563_cast_fp16 = conv(bias = const_81_to_fp16, dilations = input_561_dilations_0, groups = input_561_groups_0, pad = input_561_pad_0, pad_type = input_561_pad_type_0, strides = input_561_strides_0, weight = const_80_to_fp16_quantized, x = input_559_cast_fp16)[name = string("input_563_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_91_cast_fp16 = silu(x = input_563_cast_fp16)[name = string("h_91_cast_fp16")];
            tensor<int32, [3]> x_379_perm_0 = const()[name = string("x_379_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_22_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215646912))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216171264))))[name = string("blocks_22_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_379_cast_fp16 = transpose(perm = x_379_perm_0, x = h_91_cast_fp16)[name = string("transpose_297")];
            tensor<fp16, [1, 188, 1024]> linear_251_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_22_pw2_weight_to_fp16_palettized, x = x_379_cast_fp16)[name = string("linear_251_cast_fp16")];
            tensor<fp16, [1024]> blocks_22_pw2_scale_to_fp16 = const()[name = string("blocks_22_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216173376)))];
            tensor<fp16, [1, 188, 1024]> var_6115_cast_fp16 = mul(x = linear_251_cast_fp16, y = blocks_22_pw2_scale_to_fp16)[name = string("op_6115_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_565_cast_fp16 = add(x = input_555_cast_fp16, y = var_6115_cast_fp16)[name = string("input_565_cast_fp16")];
            tensor<int32, [1]> x_381_axes_0 = const()[name = string("x_381_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_22_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_22_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216175488)))];
            tensor<fp16, [1024]> enc_layers_22_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_22_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216177600)))];
            tensor<fp16, [1, 188, 1024]> x_381_cast_fp16 = layer_norm(axes = x_381_axes_0, beta = enc_layers_22_norm_feed_forward2_bias_to_fp16, epsilon = var_5995_to_fp16, gamma = enc_layers_22_norm_feed_forward2_weight_to_fp16, x = input_565_cast_fp16)[name = string("x_381_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_22_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216179712))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218276928))))[name = string("blocks_22_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_252_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_22_ff2_1_weight_to_fp16_palettized, x = x_381_cast_fp16)[name = string("linear_252_cast_fp16")];
            tensor<fp16, [4096]> blocks_22_ff2_1_scale_to_fp16 = const()[name = string("blocks_22_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218285184)))];
            tensor<fp16, [1, 188, 4096]> input_567_cast_fp16 = mul(x = linear_252_cast_fp16, y = blocks_22_ff2_1_scale_to_fp16)[name = string("input_567_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_383_cast_fp16 = silu(x = input_567_cast_fp16)[name = string("x_383_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_22_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218293440))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220390656))))[name = string("blocks_22_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_253_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_22_ff2_2_weight_to_fp16_palettized, x = x_383_cast_fp16)[name = string("linear_253_cast_fp16")];
            tensor<fp16, [1024]> blocks_22_ff2_2_scale_to_fp16 = const()[name = string("blocks_22_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220392768)))];
            tensor<fp16, [1, 188, 1024]> var_6129_cast_fp16 = mul(x = linear_253_cast_fp16, y = blocks_22_ff2_2_scale_to_fp16)[name = string("op_6129_cast_fp16")];
            fp16 var_6130_to_fp16 = const()[name = string("op_6130_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_6131_cast_fp16 = mul(x = var_6129_cast_fp16, y = var_6130_to_fp16)[name = string("op_6131_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_569_cast_fp16 = add(x = input_565_cast_fp16, y = var_6131_cast_fp16)[name = string("input_569_cast_fp16")];
            tensor<int32, [1]> input_571_axes_0 = const()[name = string("input_571_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_22_norm_out_weight_to_fp16 = const()[name = string("enc_layers_22_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220394880)))];
            tensor<fp16, [1024]> enc_layers_22_norm_out_bias_to_fp16 = const()[name = string("enc_layers_22_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220396992)))];
            tensor<fp16, [1, 188, 1024]> input_571_cast_fp16 = layer_norm(axes = input_571_axes_0, beta = enc_layers_22_norm_out_bias_to_fp16, epsilon = var_5995_to_fp16, gamma = enc_layers_22_norm_out_weight_to_fp16, x = input_569_cast_fp16)[name = string("input_571_cast_fp16")];
            int32 var_6145 = const()[name = string("op_6145"), val = int32(-1)];
            tensor<int32, [1]> x_385_axes_0 = const()[name = string("x_385_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_23_norm_feed_forward1_weight_to_fp16 = const()[name = string("enc_layers_23_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220399104)))];
            tensor<fp16, [1024]> enc_layers_23_norm_feed_forward1_bias_to_fp16 = const()[name = string("enc_layers_23_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220401216)))];
            fp16 var_6152_to_fp16 = const()[name = string("op_6152_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_385_cast_fp16 = layer_norm(axes = x_385_axes_0, beta = enc_layers_23_norm_feed_forward1_bias_to_fp16, epsilon = var_6152_to_fp16, gamma = enc_layers_23_norm_feed_forward1_weight_to_fp16, x = input_571_cast_fp16)[name = string("x_385_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_23_ff1_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220403328))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222500544))))[name = string("blocks_23_ff1_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_254_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_23_ff1_1_weight_to_fp16_palettized, x = x_385_cast_fp16)[name = string("linear_254_cast_fp16")];
            tensor<fp16, [4096]> blocks_23_ff1_1_scale_to_fp16 = const()[name = string("blocks_23_ff1_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222508800)))];
            tensor<fp16, [1, 188, 4096]> input_573_cast_fp16 = mul(x = linear_254_cast_fp16, y = blocks_23_ff1_1_scale_to_fp16)[name = string("input_573_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_387_cast_fp16 = silu(x = input_573_cast_fp16)[name = string("x_387_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_23_ff1_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222517056))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224614272))))[name = string("blocks_23_ff1_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_255_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_23_ff1_2_weight_to_fp16_palettized, x = x_387_cast_fp16)[name = string("linear_255_cast_fp16")];
            tensor<fp16, [1024]> blocks_23_ff1_2_scale_to_fp16 = const()[name = string("blocks_23_ff1_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224616384)))];
            tensor<fp16, [1, 188, 1024]> var_6177_cast_fp16 = mul(x = linear_255_cast_fp16, y = blocks_23_ff1_2_scale_to_fp16)[name = string("op_6177_cast_fp16")];
            fp16 var_6178_to_fp16 = const()[name = string("op_6178_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_6179_cast_fp16 = mul(x = var_6177_cast_fp16, y = var_6178_to_fp16)[name = string("op_6179_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_575_cast_fp16 = add(x = input_571_cast_fp16, y = var_6179_cast_fp16)[name = string("input_575_cast_fp16")];
            tensor<int32, [1]> x_389_axes_0 = const()[name = string("x_389_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_23_norm_self_att_weight_to_fp16 = const()[name = string("enc_layers_23_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224618496)))];
            tensor<fp16, [1024]> enc_layers_23_norm_self_att_bias_to_fp16 = const()[name = string("enc_layers_23_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224620608)))];
            tensor<fp16, [1, 188, 1024]> x_389_cast_fp16 = layer_norm(axes = x_389_axes_0, beta = enc_layers_23_norm_self_att_bias_to_fp16, epsilon = var_6152_to_fp16, gamma = enc_layers_23_norm_self_att_weight_to_fp16, x = input_575_cast_fp16)[name = string("x_389_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_23_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224622720))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225147072))))[name = string("blocks_23_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_256_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_23_q_weight_to_fp16_palettized, x = x_389_cast_fp16)[name = string("linear_256_cast_fp16")];
            tensor<fp16, [1024]> blocks_23_q_scale_to_fp16 = const()[name = string("blocks_23_q_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225149184)))];
            tensor<fp16, [1, 188, 1024]> var_6187_cast_fp16 = mul(x = linear_256_cast_fp16, y = blocks_23_q_scale_to_fp16)[name = string("op_6187_cast_fp16")];
            tensor<int32, [4]> var_6188 = const()[name = string("op_6188"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_6189_cast_fp16 = reshape(shape = var_6188, x = var_6187_cast_fp16)[name = string("op_6189_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_23_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225151296))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225675648))))[name = string("blocks_23_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_257_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_23_k_weight_to_fp16_palettized, x = x_389_cast_fp16)[name = string("linear_257_cast_fp16")];
            tensor<fp16, [1024]> blocks_23_k_scale_to_fp16 = const()[name = string("blocks_23_k_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225677760)))];
            tensor<fp16, [1, 188, 1024]> var_6195_cast_fp16 = mul(x = linear_257_cast_fp16, y = blocks_23_k_scale_to_fp16)[name = string("op_6195_cast_fp16")];
            tensor<int32, [4]> var_6196 = const()[name = string("op_6196"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_6197_cast_fp16 = reshape(shape = var_6196, x = var_6195_cast_fp16)[name = string("op_6197_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_23_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225679872))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226204224))))[name = string("blocks_23_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_258_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_23_v_weight_to_fp16_palettized, x = x_389_cast_fp16)[name = string("linear_258_cast_fp16")];
            tensor<fp16, [1024]> blocks_23_v_scale_to_fp16 = const()[name = string("blocks_23_v_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226206336)))];
            tensor<fp16, [1, 188, 1024]> var_6203_cast_fp16 = mul(x = linear_258_cast_fp16, y = blocks_23_v_scale_to_fp16)[name = string("op_6203_cast_fp16")];
            tensor<int32, [4]> var_6204 = const()[name = string("op_6204"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_6205_cast_fp16 = reshape(shape = var_6204, x = var_6203_cast_fp16)[name = string("op_6205_cast_fp16")];
            tensor<int32, [4]> v_47_perm_0 = const()[name = string("v_47_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 8, 128]> const_130_to_fp16 = const()[name = string("const_130_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226208448)))];
            tensor<fp16, [1, 188, 8, 128]> q_u_cast_fp16 = add(x = var_6189_cast_fp16, y = const_130_to_fp16)[name = string("q_u_cast_fp16")];
            tensor<fp16, [1, 1, 8, 128]> const_131_to_fp16 = const()[name = string("const_131_to_fp16"), val = tensor<fp16, [1, 1, 8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226210560)))];
            tensor<fp16, [1, 188, 8, 128]> q_v_cast_fp16 = add(x = var_6189_cast_fp16, y = const_131_to_fp16)[name = string("q_v_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_23_pos_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226212672))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226737024))))[name = string("blocks_23_pos_weight_to_fp16_palettized")];
            tensor<fp16, [1, 375, 1024]> linear_259_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_23_pos_weight_to_fp16_palettized, x = x_21_cast_fp16)[name = string("linear_259_cast_fp16")];
            tensor<fp16, [1024]> blocks_23_pos_scale_to_fp16 = const()[name = string("blocks_23_pos_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226739136)))];
            tensor<fp16, [1, 375, 1024]> var_6217_cast_fp16 = mul(x = linear_259_cast_fp16, y = blocks_23_pos_scale_to_fp16)[name = string("op_6217_cast_fp16")];
            tensor<int32, [4]> var_6218 = const()[name = string("op_6218"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 375, 8, 128]> p_cast_fp16 = reshape(shape = var_6218, x = var_6217_cast_fp16)[name = string("p_cast_fp16")];
            bool input_577_transpose_x_0 = const()[name = string("input_577_transpose_x_0"), val = bool(false)];
            bool input_577_transpose_y_0 = const()[name = string("input_577_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_284_perm_0 = const()[name = string("transpose_284_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_285_perm_0 = const()[name = string("transpose_285_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 375]> transpose_285 = transpose(perm = transpose_285_perm_0, x = p_cast_fp16)[name = string("transpose_295")];
            tensor<fp16, [1, 8, 188, 128]> transpose_284 = transpose(perm = transpose_284_perm_0, x = q_v_cast_fp16)[name = string("transpose_296")];
            tensor<fp16, [1, 8, 188, 375]> input_577_cast_fp16 = matmul(transpose_x = input_577_transpose_x_0, transpose_y = input_577_transpose_y_0, x = transpose_284, y = transpose_285)[name = string("input_577_cast_fp16")];
            tensor<int32, [8]> var_6224_pad_0 = const()[name = string("op_6224_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_6224_mode_0 = const()[name = string("op_6224_mode_0"), val = string("constant")];
            fp16 const_35_to_fp16 = const()[name = string("const_35_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_6224_cast_fp16 = pad(constant_val = const_35_to_fp16, mode = var_6224_mode_0, pad = var_6224_pad_0, x = input_577_cast_fp16)[name = string("op_6224_cast_fp16")];
            tensor<int32, [4]> var_6225 = const()[name = string("op_6225"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<fp16, [1, 8, 376, 188]> scores_93_cast_fp16 = reshape(shape = var_6225, x = var_6224_cast_fp16)[name = string("scores_93_cast_fp16")];
            tensor<int32, [4]> var_6229_begin_0 = const()[name = string("op_6229_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_6229_end_0 = const()[name = string("op_6229_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_6229_end_mask_0 = const()[name = string("op_6229_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_6229_cast_fp16 = slice_by_index(begin = var_6229_begin_0, end = var_6229_end_0, end_mask = var_6229_end_mask_0, x = scores_93_cast_fp16)[name = string("op_6229_cast_fp16")];
            tensor<int32, [4]> var_6230 = const()[name = string("op_6230"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_6231_cast_fp16 = reshape(shape = var_6230, x = var_6229_cast_fp16)[name = string("op_6231_cast_fp16")];
            tensor<int32, [4]> bd_begin_0 = const()[name = string("bd_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_end_0 = const()[name = string("bd_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_end_mask_0 = const()[name = string("bd_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_cast_fp16 = slice_by_index(begin = bd_begin_0, end = bd_end_0, end_mask = bd_end_mask_0, x = var_6231_cast_fp16)[name = string("bd_cast_fp16")];
            bool var_6234_transpose_x_0 = const()[name = string("op_6234_transpose_x_0"), val = bool(false)];
            bool var_6234_transpose_y_0 = const()[name = string("op_6234_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_286_perm_0 = const()[name = string("transpose_286_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_287_perm_0 = const()[name = string("transpose_287_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_287 = transpose(perm = transpose_287_perm_0, x = var_6197_cast_fp16)[name = string("transpose_293")];
            tensor<fp16, [1, 8, 188, 128]> transpose_286 = transpose(perm = transpose_286_perm_0, x = q_u_cast_fp16)[name = string("transpose_294")];
            tensor<fp16, [1, 8, 188, 188]> var_6234_cast_fp16 = matmul(transpose_x = var_6234_transpose_x_0, transpose_y = var_6234_transpose_y_0, x = transpose_286, y = transpose_287)[name = string("op_6234_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_6235_cast_fp16 = add(x = var_6234_cast_fp16, y = bd_cast_fp16)[name = string("op_6235_cast_fp16")];
            fp16 var_6236_to_fp16 = const()[name = string("op_6236_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_6237_cast_fp16 = mul(x = var_6235_cast_fp16, y = var_6236_to_fp16)[name = string("op_6237_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_cast_fp16 = add(x = var_6237_cast_fp16, y = key_bias_cast_fp16)[name = string("scores_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_6239_cast_fp16 = softmax(axis = var_6145, x = scores_cast_fp16)[name = string("op_6239_cast_fp16")];
            bool out_transpose_x_0 = const()[name = string("out_transpose_x_0"), val = bool(false)];
            bool out_transpose_y_0 = const()[name = string("out_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_47_cast_fp16 = transpose(perm = v_47_perm_0, x = var_6205_cast_fp16)[name = string("transpose_292")];
            tensor<fp16, [1, 8, 188, 128]> out_cast_fp16 = matmul(transpose_x = out_transpose_x_0, transpose_y = out_transpose_y_0, x = var_6239_cast_fp16, y = v_47_cast_fp16)[name = string("out_cast_fp16")];
            tensor<int32, [4]> var_6241_perm_0 = const()[name = string("op_6241_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_6242 = const()[name = string("op_6242"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> var_6241_cast_fp16 = transpose(perm = var_6241_perm_0, x = out_cast_fp16)[name = string("transpose_291")];
            tensor<fp16, [1, 188, 1024]> x_391_cast_fp16 = reshape(shape = var_6242, x = var_6241_cast_fp16)[name = string("x_391_cast_fp16")];
            tensor<fp16, [1024, 1024]> blocks_23_o_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226741248))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227265600))))[name = string("blocks_23_o_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_260_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_23_o_weight_to_fp16_palettized, x = x_391_cast_fp16)[name = string("linear_260_cast_fp16")];
            tensor<fp16, [1024]> blocks_23_o_scale_to_fp16 = const()[name = string("blocks_23_o_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227267712)))];
            tensor<fp16, [1, 188, 1024]> var_6248_cast_fp16 = mul(x = linear_260_cast_fp16, y = blocks_23_o_scale_to_fp16)[name = string("op_6248_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_579_cast_fp16 = add(x = input_575_cast_fp16, y = var_6248_cast_fp16)[name = string("input_579_cast_fp16")];
            tensor<int32, [1]> x_393_axes_0 = const()[name = string("x_393_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_23_norm_conv_weight_to_fp16 = const()[name = string("enc_layers_23_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227269824)))];
            tensor<fp16, [1024]> enc_layers_23_norm_conv_bias_to_fp16 = const()[name = string("enc_layers_23_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227271936)))];
            tensor<fp16, [1, 188, 1024]> x_393_cast_fp16 = layer_norm(axes = x_393_axes_0, beta = enc_layers_23_norm_conv_bias_to_fp16, epsilon = var_6152_to_fp16, gamma = enc_layers_23_norm_conv_weight_to_fp16, x = input_579_cast_fp16)[name = string("x_393_cast_fp16")];
            tensor<fp16, [2048, 1024]> blocks_23_pw1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227274048))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228322688))))[name = string("blocks_23_pw1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 2048]> linear_261_cast_fp16 = linear(bias = blocks_0_pw1_bias_to_fp16, weight = blocks_23_pw1_weight_to_fp16_palettized, x = x_393_cast_fp16)[name = string("linear_261_cast_fp16")];
            tensor<fp16, [2048]> blocks_23_pw1_scale_to_fp16 = const()[name = string("blocks_23_pw1_scale_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228326848)))];
            tensor<fp16, [1, 188, 2048]> input_581_cast_fp16 = mul(x = linear_261_cast_fp16, y = blocks_23_pw1_scale_to_fp16)[name = string("input_581_cast_fp16")];
            int32 var_6257_split_num_splits_0 = const()[name = string("op_6257_split_num_splits_0"), val = int32(2)];
            int32 var_6257_split_axis_0 = const()[name = string("op_6257_split_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 188, 1024]> var_6257_split_cast_fp16_0, tensor<fp16, [1, 188, 1024]> var_6257_split_cast_fp16_1 = split(axis = var_6257_split_axis_0, num_splits = var_6257_split_num_splits_0, x = input_581_cast_fp16)[name = string("op_6257_split_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_6257_split_1_sigmoid_cast_fp16 = sigmoid(x = var_6257_split_cast_fp16_1)[name = string("op_6257_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> var_6257_cast_fp16 = mul(x = var_6257_split_cast_fp16_0, y = var_6257_split_1_sigmoid_cast_fp16)[name = string("op_6257_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> h_93_cast_fp16 = mul(x = var_6257_cast_fp16, y = valid_cast_fp16)[name = string("h_93_cast_fp16")];
            tensor<int32, [3]> input_583_perm_0 = const()[name = string("input_583_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_585_pad_type_0 = const()[name = string("input_585_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_585_pad_0 = const()[name = string("input_585_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_585_groups_0 = const()[name = string("input_585_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_585_strides_0 = const()[name = string("input_585_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_585_dilations_0 = const()[name = string("input_585_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_82_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228331008))), scale = tensor<fp16, [1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228340288))))[name = string("const_82_to_fp16_quantized")];
            tensor<fp16, [1024]> const_83_to_fp16 = const()[name = string("const_83_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228342400)))];
            tensor<fp16, [1, 1024, 188]> input_583_cast_fp16 = transpose(perm = input_583_perm_0, x = h_93_cast_fp16)[name = string("transpose_290")];
            tensor<fp16, [1, 1024, 188]> input_587_cast_fp16 = conv(bias = const_83_to_fp16, dilations = input_585_dilations_0, groups = input_585_groups_0, pad = input_585_pad_0, pad_type = input_585_pad_type_0, strides = input_585_strides_0, weight = const_82_to_fp16_quantized, x = input_583_cast_fp16)[name = string("input_587_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> h_cast_fp16 = silu(x = input_587_cast_fp16)[name = string("h_cast_fp16")];
            tensor<int32, [3]> x_395_perm_0 = const()[name = string("x_395_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> blocks_23_pw2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228344512))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228868864))))[name = string("blocks_23_pw2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> x_395_cast_fp16 = transpose(perm = x_395_perm_0, x = h_cast_fp16)[name = string("transpose_289")];
            tensor<fp16, [1, 188, 1024]> linear_262_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_23_pw2_weight_to_fp16_palettized, x = x_395_cast_fp16)[name = string("linear_262_cast_fp16")];
            tensor<fp16, [1024]> blocks_23_pw2_scale_to_fp16 = const()[name = string("blocks_23_pw2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228870976)))];
            tensor<fp16, [1, 188, 1024]> var_6272_cast_fp16 = mul(x = linear_262_cast_fp16, y = blocks_23_pw2_scale_to_fp16)[name = string("op_6272_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_589_cast_fp16 = add(x = input_579_cast_fp16, y = var_6272_cast_fp16)[name = string("input_589_cast_fp16")];
            tensor<int32, [1]> x_397_axes_0 = const()[name = string("x_397_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_23_norm_feed_forward2_weight_to_fp16 = const()[name = string("enc_layers_23_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228873088)))];
            tensor<fp16, [1024]> enc_layers_23_norm_feed_forward2_bias_to_fp16 = const()[name = string("enc_layers_23_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228875200)))];
            tensor<fp16, [1, 188, 1024]> x_397_cast_fp16 = layer_norm(axes = x_397_axes_0, beta = enc_layers_23_norm_feed_forward2_bias_to_fp16, epsilon = var_6152_to_fp16, gamma = enc_layers_23_norm_feed_forward2_weight_to_fp16, x = input_589_cast_fp16)[name = string("x_397_cast_fp16")];
            tensor<fp16, [4096, 1024]> blocks_23_ff2_1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228877312))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230974528))))[name = string("blocks_23_ff2_1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_263_cast_fp16 = linear(bias = blocks_0_ff1_1_bias_to_fp16, weight = blocks_23_ff2_1_weight_to_fp16_palettized, x = x_397_cast_fp16)[name = string("linear_263_cast_fp16")];
            tensor<fp16, [4096]> blocks_23_ff2_1_scale_to_fp16 = const()[name = string("blocks_23_ff2_1_scale_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230982784)))];
            tensor<fp16, [1, 188, 4096]> input_591_cast_fp16 = mul(x = linear_263_cast_fp16, y = blocks_23_ff2_1_scale_to_fp16)[name = string("input_591_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> x_399_cast_fp16 = silu(x = input_591_cast_fp16)[name = string("x_399_cast_fp16")];
            tensor<fp16, [1024, 4096]> blocks_23_ff2_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230991040))), lut = tensor<fp16, [64, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233088256))))[name = string("blocks_23_ff2_2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_264_cast_fp16 = linear(bias = blocks_0_ff1_2_bias_to_fp16, weight = blocks_23_ff2_2_weight_to_fp16_palettized, x = x_399_cast_fp16)[name = string("linear_264_cast_fp16")];
            tensor<fp16, [1024]> blocks_23_ff2_2_scale_to_fp16 = const()[name = string("blocks_23_ff2_2_scale_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233090368)))];
            tensor<fp16, [1, 188, 1024]> var_6286_cast_fp16 = mul(x = linear_264_cast_fp16, y = blocks_23_ff2_2_scale_to_fp16)[name = string("op_6286_cast_fp16")];
            fp16 var_6287_to_fp16 = const()[name = string("op_6287_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_6288_cast_fp16 = mul(x = var_6286_cast_fp16, y = var_6287_to_fp16)[name = string("op_6288_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_cast_fp16 = add(x = input_589_cast_fp16, y = var_6288_cast_fp16)[name = string("input_cast_fp16")];
            tensor<int32, [1]> x_axes_0 = const()[name = string("x_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> enc_layers_23_norm_out_weight_to_fp16 = const()[name = string("enc_layers_23_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233092480)))];
            tensor<fp16, [1024]> enc_layers_23_norm_out_bias_to_fp16 = const()[name = string("enc_layers_23_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233094592)))];
            tensor<fp16, [1, 188, 1024]> x_cast_fp16 = layer_norm(axes = x_axes_0, beta = enc_layers_23_norm_out_bias_to_fp16, epsilon = var_6152_to_fp16, gamma = enc_layers_23_norm_out_weight_to_fp16, x = input_cast_fp16)[name = string("x_cast_fp16")];
            tensor<int32, [3]> var_6294_perm_0 = const()[name = string("op_6294_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string var_6294_cast_fp16_to_fp32_dtype_0 = const()[name = string("op_6294_cast_fp16_to_fp32_dtype_0"), val = string("fp32")];
            string cast_236_dtype_0 = const()[name = string("cast_236_dtype_0"), val = string("int32")];
            tensor<int32, [1]> encoder_length = cast(dtype = cast_236_dtype_0, x = length_cast_fp16)[name = string("cast_0")];
            tensor<fp16, [1, 1024, 188]> var_6294_cast_fp16 = transpose(perm = var_6294_perm_0, x = x_cast_fp16)[name = string("transpose_288")];
            tensor<fp32, [1, 1024, 188]> encoder = cast(dtype = var_6294_cast_fp16_to_fp32_dtype_0, x = var_6294_cast_fp16)[name = string("cast_1")];
        } -> (encoder, encoder_length);
}