program(1.3)
[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3600.16.1"}, {"coremlc-version", "3600.25.2"}})]
{
    func main<ios18>(tensor<int32, [1, 1501]> attention_mask, tensor<fp32, [1, 1501, 128]> input_features) {
            tensor<int32, [1]> unsqueeze_axes_0 = const()[name = string("unsqueeze_axes_0"), val = tensor<int32, [1]>([1])];
            string input_features_to_fp16_dtype_0 = const()[name = string("input_features_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 1501, 128]> input_features_to_fp16 = cast(dtype = input_features_to_fp16_dtype_0, x = input_features)[name = string("cast_8")];
            tensor<fp16, [1, 1, 1501, 128]> unsqueeze_cast_fp16 = expand_dims(axes = unsqueeze_axes_0, x = input_features_to_fp16)[name = string("unsqueeze_cast_fp16")];
            tensor<int32, [1]> sum_1_axes_0 = const()[name = string("sum_1_axes_0"), val = tensor<int32, [1]>([-1])];
            bool sum_1_keep_dims_0 = const()[name = string("sum_1_keep_dims_0"), val = bool(false)];
            tensor<int32, [1]> sum_1 = reduce_sum(axes = sum_1_axes_0, keep_dims = sum_1_keep_dims_0, x = attention_mask)[name = string("sum_1")];
            string conv2d_pad_type_0 = const()[name = string("conv2d_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> conv2d_pad_0 = const()[name = string("conv2d_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> conv2d_strides_0 = const()[name = string("conv2d_strides_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> conv2d_dilations_0 = const()[name = string("conv2d_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 conv2d_groups_0 = const()[name = string("conv2d_groups_0"), val = int32(1)];
            tensor<fp16, [256, 1, 3, 3]> p_encoder_subsampling_layers_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64))), lut = tensor<fp16, [16, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1856))))[name = string("p_encoder_subsampling_layers_0_weight_to_fp16_palettized")];
            tensor<fp16, [256]> p_encoder_subsampling_layers_0_bias_to_fp16 = const()[name = string("p_encoder_subsampling_layers_0_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3968)))];
            tensor<fp16, [1, 256, 751, 64]> conv2d_cast_fp16 = conv(bias = p_encoder_subsampling_layers_0_bias_to_fp16, dilations = conv2d_dilations_0, groups = conv2d_groups_0, pad = conv2d_pad_0, pad_type = conv2d_pad_type_0, strides = conv2d_strides_0, weight = p_encoder_subsampling_layers_0_weight_to_fp16_palettized, x = unsqueeze_cast_fp16)[name = string("conv2d_cast_fp16")];
            int32 const_4 = const()[name = string("const_4"), val = int32(1)];
            tensor<int32, [1]> add = add(x = sum_1, y = const_4)[name = string("add")];
            int32 const_5 = const()[name = string("const_5"), val = int32(1)];
            tensor<int32, [1]> add_1 = add(x = add, y = const_5)[name = string("add_1")];
            int32 const_6 = const()[name = string("const_6"), val = int32(3)];
            tensor<int32, [1]> sub = sub(x = add_1, y = const_6)[name = string("sub")];
            int32 const_7 = const()[name = string("const_7"), val = int32(2)];
            tensor<int32, [1]> floor_div_0 = floor_div(x = sub, y = const_7)[name = string("floor_div_0")];
            string floor_divide_to_fp16_dtype_0 = const()[name = string("floor_divide_to_fp16_dtype_0"), val = string("fp16")];
            fp16 const_8_promoted_to_fp16 = const()[name = string("const_8_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> floor_div_0_to_fp16 = cast(dtype = floor_divide_to_fp16_dtype_0, x = floor_div_0)[name = string("cast_7")];
            tensor<fp16, [1]> add_2_cast_fp16 = add(x = floor_div_0_to_fp16, y = const_8_promoted_to_fp16)[name = string("add_2_cast_fp16")];
            tensor<int32, [1]> unsqueeze_1_axes_0 = const()[name = string("unsqueeze_1_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1]> unsqueeze_1_cast_fp16 = expand_dims(axes = unsqueeze_1_axes_0, x = add_2_cast_fp16)[name = string("unsqueeze_1_cast_fp16")];
            tensor<fp16, [751]> arange_promoted_to_fp16 = const()[name = string("arange_promoted_to_fp16"), val = tensor<fp16, [751]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4544)))];
            tensor<bool, [1, 751]> lt_cast_fp16 = less(x = arange_promoted_to_fp16, y = unsqueeze_1_cast_fp16)[name = string("lt_cast_fp16")];
            tensor<int32, [1]> unsqueeze_2_axes_0 = const()[name = string("unsqueeze_2_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<bool, [1, 1, 751]> unsqueeze_2 = expand_dims(axes = unsqueeze_2_axes_0, x = lt_cast_fp16)[name = string("unsqueeze_2")];
            tensor<int32, [1]> unsqueeze_3_axes_0 = const()[name = string("unsqueeze_3_axes_0"), val = tensor<int32, [1]>([3])];
            tensor<bool, [1, 1, 751, 1]> unsqueeze_3 = expand_dims(axes = unsqueeze_3_axes_0, x = unsqueeze_2)[name = string("unsqueeze_3")];
            string unsqueeze_3_promoted_to_fp16_dtype_0 = const()[name = string("unsqueeze_3_promoted_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 1, 751, 1]> unsqueeze_3_to_fp16 = cast(dtype = unsqueeze_3_promoted_to_fp16_dtype_0, x = unsqueeze_3)[name = string("cast_6")];
            tensor<fp16, [1, 256, 751, 64]> mul_cast_fp16 = mul(x = conv2d_cast_fp16, y = unsqueeze_3_to_fp16)[name = string("mul_cast_fp16")];
            tensor<fp16, [1, 256, 751, 64]> relu_cast_fp16 = relu(x = mul_cast_fp16)[name = string("relu_cast_fp16")];
            string conv2d_1_pad_type_0 = const()[name = string("conv2d_1_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> conv2d_1_pad_0 = const()[name = string("conv2d_1_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> conv2d_1_strides_0 = const()[name = string("conv2d_1_strides_0"), val = tensor<int32, [2]>([2, 2])];
            int32 conv2d_1_groups_0 = const()[name = string("conv2d_1_groups_0"), val = int32(256)];
            tensor<int32, [2]> conv2d_1_dilations_0 = const()[name = string("conv2d_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [256, 1, 3, 3]> p_encoder_subsampling_layers_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6144))), lut = tensor<fp16, [16, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7936))))[name = string("p_encoder_subsampling_layers_2_weight_to_fp16_palettized")];
            tensor<fp16, [256]> p_encoder_subsampling_layers_2_bias_to_fp16 = const()[name = string("p_encoder_subsampling_layers_2_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10048)))];
            tensor<fp16, [1, 256, 376, 32]> conv2d_1_cast_fp16 = conv(bias = p_encoder_subsampling_layers_2_bias_to_fp16, dilations = conv2d_1_dilations_0, groups = conv2d_1_groups_0, pad = conv2d_1_pad_0, pad_type = conv2d_1_pad_type_0, strides = conv2d_1_strides_0, weight = p_encoder_subsampling_layers_2_weight_to_fp16_palettized, x = relu_cast_fp16)[name = string("conv2d_1_cast_fp16")];
            fp16 const_19_promoted_to_fp16 = const()[name = string("const_19_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> add_3_cast_fp16 = add(x = add_2_cast_fp16, y = const_19_promoted_to_fp16)[name = string("add_3_cast_fp16")];
            fp16 const_20_promoted_to_fp16 = const()[name = string("const_20_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> add_4_cast_fp16 = add(x = add_3_cast_fp16, y = const_20_promoted_to_fp16)[name = string("add_4_cast_fp16")];
            fp16 const_21_promoted_to_fp16 = const()[name = string("const_21_promoted_to_fp16"), val = fp16(0x1.8p+1)];
            tensor<fp16, [1]> sub_1_cast_fp16 = sub(x = add_4_cast_fp16, y = const_21_promoted_to_fp16)[name = string("sub_1_cast_fp16")];
            fp16 const_22_promoted_to_fp16 = const()[name = string("const_22_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1]> floor_div_1_cast_fp16 = floor_div(x = sub_1_cast_fp16, y = const_22_promoted_to_fp16)[name = string("floor_div_1_cast_fp16")];
            fp16 const_23_promoted_to_fp16 = const()[name = string("const_23_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> add_5_cast_fp16 = add(x = floor_div_1_cast_fp16, y = const_23_promoted_to_fp16)[name = string("add_5_cast_fp16")];
            tensor<int32, [1]> unsqueeze_4_axes_0 = const()[name = string("unsqueeze_4_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1]> unsqueeze_4_cast_fp16 = expand_dims(axes = unsqueeze_4_axes_0, x = add_5_cast_fp16)[name = string("unsqueeze_4_cast_fp16")];
            tensor<fp16, [376]> arange_1_promoted_to_fp16 = const()[name = string("arange_1_promoted_to_fp16"), val = tensor<fp16, [376]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10624)))];
            tensor<bool, [1, 376]> lt_1_cast_fp16 = less(x = arange_1_promoted_to_fp16, y = unsqueeze_4_cast_fp16)[name = string("lt_1_cast_fp16")];
            tensor<int32, [1]> unsqueeze_5_axes_0 = const()[name = string("unsqueeze_5_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<bool, [1, 1, 376]> unsqueeze_5 = expand_dims(axes = unsqueeze_5_axes_0, x = lt_1_cast_fp16)[name = string("unsqueeze_5")];
            tensor<int32, [1]> unsqueeze_6_axes_0 = const()[name = string("unsqueeze_6_axes_0"), val = tensor<int32, [1]>([3])];
            tensor<bool, [1, 1, 376, 1]> unsqueeze_6 = expand_dims(axes = unsqueeze_6_axes_0, x = unsqueeze_5)[name = string("unsqueeze_6")];
            string unsqueeze_6_promoted_to_fp16_dtype_0 = const()[name = string("unsqueeze_6_promoted_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 1, 376, 1]> unsqueeze_6_to_fp16 = cast(dtype = unsqueeze_6_promoted_to_fp16_dtype_0, x = unsqueeze_6)[name = string("cast_5")];
            tensor<fp16, [1, 256, 376, 32]> mul_1_cast_fp16 = mul(x = conv2d_1_cast_fp16, y = unsqueeze_6_to_fp16)[name = string("mul_1_cast_fp16")];
            string conv2d_2_pad_type_0 = const()[name = string("conv2d_2_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> conv2d_2_strides_0 = const()[name = string("conv2d_2_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> conv2d_2_pad_0 = const()[name = string("conv2d_2_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> conv2d_2_dilations_0 = const()[name = string("conv2d_2_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 conv2d_2_groups_0 = const()[name = string("conv2d_2_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1, 1]> p_encoder_subsampling_layers_3_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11456))), lut = tensor<fp16, [16, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60672))))[name = string("p_encoder_subsampling_layers_3_weight_to_fp16_palettized")];
            tensor<fp16, [256]> p_encoder_subsampling_layers_3_bias_to_fp16 = const()[name = string("p_encoder_subsampling_layers_3_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62784)))];
            tensor<fp16, [1, 256, 376, 32]> conv2d_2_cast_fp16 = conv(bias = p_encoder_subsampling_layers_3_bias_to_fp16, dilations = conv2d_2_dilations_0, groups = conv2d_2_groups_0, pad = conv2d_2_pad_0, pad_type = conv2d_2_pad_type_0, strides = conv2d_2_strides_0, weight = p_encoder_subsampling_layers_3_weight_to_fp16_palettized, x = mul_1_cast_fp16)[name = string("conv2d_2_cast_fp16")];
            tensor<fp16, [1, 256, 376, 32]> mul_2_cast_fp16 = mul(x = conv2d_2_cast_fp16, y = unsqueeze_6_to_fp16)[name = string("mul_2_cast_fp16")];
            tensor<fp16, [1, 256, 376, 32]> relu_1_cast_fp16 = relu(x = mul_2_cast_fp16)[name = string("relu_1_cast_fp16")];
            string conv2d_3_pad_type_0 = const()[name = string("conv2d_3_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> conv2d_3_pad_0 = const()[name = string("conv2d_3_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> conv2d_3_strides_0 = const()[name = string("conv2d_3_strides_0"), val = tensor<int32, [2]>([2, 2])];
            int32 conv2d_3_groups_0 = const()[name = string("conv2d_3_groups_0"), val = int32(256)];
            tensor<int32, [2]> conv2d_3_dilations_0 = const()[name = string("conv2d_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [256, 1, 3, 3]> p_encoder_subsampling_layers_5_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63360))), lut = tensor<fp16, [16, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65152))))[name = string("p_encoder_subsampling_layers_5_weight_to_fp16_palettized")];
            tensor<fp16, [256]> p_encoder_subsampling_layers_5_bias_to_fp16 = const()[name = string("p_encoder_subsampling_layers_5_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67264)))];
            tensor<fp16, [1, 256, 188, 16]> conv2d_3_cast_fp16 = conv(bias = p_encoder_subsampling_layers_5_bias_to_fp16, dilations = conv2d_3_dilations_0, groups = conv2d_3_groups_0, pad = conv2d_3_pad_0, pad_type = conv2d_3_pad_type_0, strides = conv2d_3_strides_0, weight = p_encoder_subsampling_layers_5_weight_to_fp16_palettized, x = relu_1_cast_fp16)[name = string("conv2d_3_cast_fp16")];
            fp16 const_40_promoted_to_fp16 = const()[name = string("const_40_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> add_6_cast_fp16 = add(x = add_5_cast_fp16, y = const_40_promoted_to_fp16)[name = string("add_6_cast_fp16")];
            fp16 const_41_promoted_to_fp16 = const()[name = string("const_41_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> add_7_cast_fp16 = add(x = add_6_cast_fp16, y = const_41_promoted_to_fp16)[name = string("add_7_cast_fp16")];
            fp16 const_42_promoted_to_fp16 = const()[name = string("const_42_promoted_to_fp16"), val = fp16(0x1.8p+1)];
            tensor<fp16, [1]> sub_2_cast_fp16 = sub(x = add_7_cast_fp16, y = const_42_promoted_to_fp16)[name = string("sub_2_cast_fp16")];
            fp16 const_43_promoted_to_fp16 = const()[name = string("const_43_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1]> floor_div_2_cast_fp16 = floor_div(x = sub_2_cast_fp16, y = const_43_promoted_to_fp16)[name = string("floor_div_2_cast_fp16")];
            fp16 const_44_promoted_to_fp16 = const()[name = string("const_44_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> add_8_cast_fp16 = add(x = floor_div_2_cast_fp16, y = const_44_promoted_to_fp16)[name = string("add_8_cast_fp16")];
            tensor<int32, [1]> unsqueeze_10_axes_0 = const()[name = string("unsqueeze_10_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1]> unsqueeze_10_cast_fp16 = expand_dims(axes = unsqueeze_10_axes_0, x = add_8_cast_fp16)[name = string("unsqueeze_10_cast_fp16")];
            tensor<fp16, [188]> arange_3_promoted_to_fp16 = const()[name = string("arange_3_promoted_to_fp16"), val = tensor<fp16, [188]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67840)))];
            tensor<bool, [1, 188]> lt_3_cast_fp16 = less(x = arange_3_promoted_to_fp16, y = unsqueeze_10_cast_fp16)[name = string("lt_3_cast_fp16")];
            tensor<int32, [1]> unsqueeze_11_axes_0 = const()[name = string("unsqueeze_11_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<bool, [1, 1, 188]> unsqueeze_11 = expand_dims(axes = unsqueeze_11_axes_0, x = lt_3_cast_fp16)[name = string("unsqueeze_11")];
            tensor<int32, [1]> unsqueeze_12_axes_0 = const()[name = string("unsqueeze_12_axes_0"), val = tensor<int32, [1]>([3])];
            tensor<bool, [1, 1, 188, 1]> unsqueeze_12 = expand_dims(axes = unsqueeze_12_axes_0, x = unsqueeze_11)[name = string("unsqueeze_12")];
            string unsqueeze_12_promoted_to_fp16_dtype_0 = const()[name = string("unsqueeze_12_promoted_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 1, 188, 1]> unsqueeze_12_to_fp16 = cast(dtype = unsqueeze_12_promoted_to_fp16_dtype_0, x = unsqueeze_12)[name = string("cast_4")];
            tensor<fp16, [1, 256, 188, 16]> mul_3_cast_fp16 = mul(x = conv2d_3_cast_fp16, y = unsqueeze_12_to_fp16)[name = string("mul_3_cast_fp16")];
            string conv2d_4_pad_type_0 = const()[name = string("conv2d_4_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> conv2d_4_strides_0 = const()[name = string("conv2d_4_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> conv2d_4_pad_0 = const()[name = string("conv2d_4_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> conv2d_4_dilations_0 = const()[name = string("conv2d_4_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 conv2d_4_groups_0 = const()[name = string("conv2d_4_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1, 1]> p_encoder_subsampling_layers_6_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68288))), lut = tensor<fp16, [16, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117504))))[name = string("p_encoder_subsampling_layers_6_weight_to_fp16_palettized")];
            tensor<fp16, [256]> p_encoder_subsampling_layers_6_bias_to_fp16 = const()[name = string("p_encoder_subsampling_layers_6_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119616)))];
            tensor<fp16, [1, 256, 188, 16]> conv2d_4_cast_fp16 = conv(bias = p_encoder_subsampling_layers_6_bias_to_fp16, dilations = conv2d_4_dilations_0, groups = conv2d_4_groups_0, pad = conv2d_4_pad_0, pad_type = conv2d_4_pad_type_0, strides = conv2d_4_strides_0, weight = p_encoder_subsampling_layers_6_weight_to_fp16_palettized, x = mul_3_cast_fp16)[name = string("conv2d_4_cast_fp16")];
            tensor<fp16, [1, 256, 188, 16]> mul_4_cast_fp16 = mul(x = conv2d_4_cast_fp16, y = unsqueeze_12_to_fp16)[name = string("mul_4_cast_fp16")];
            tensor<fp16, [1, 256, 188, 16]> relu_2_cast_fp16 = relu(x = mul_4_cast_fp16)[name = string("relu_2_cast_fp16")];
            tensor<int32, [4]> transpose_perm_0 = const()[name = string("transpose_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_59 = const()[name = string("const_59"), val = tensor<int32, [3]>([1, 188, 4096])];
            tensor<fp16, [1, 188, 256, 16]> transpose_cast_fp16 = transpose(perm = transpose_perm_0, x = relu_2_cast_fp16)[name = string("transpose_145")];
            tensor<fp16, [1, 188, 4096]> _unsafe_view_cast_fp16 = reshape(shape = const_59, x = transpose_cast_fp16)[name = string("_unsafe_view_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_subsampling_linear_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120192))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3265984))))[name = string("p_encoder_subsampling_linear_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> p_encoder_subsampling_linear_bias_to_fp16 = const()[name = string("p_encoder_subsampling_linear_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3274240)))];
            tensor<fp16, [1, 188, 1024]> linear_0_cast_fp16 = linear(bias = p_encoder_subsampling_linear_bias_to_fp16, weight = p_encoder_subsampling_linear_weight_to_fp16_palettized, x = _unsafe_view_cast_fp16)[name = string("linear_0_cast_fp16")];
            string _to_copy_1_to_fp16_dtype_0 = const()[name = string("_to_copy_1_to_fp16_dtype_0"), val = string("fp16")];
            fp16 const_84_promoted_to_fp16 = const()[name = string("const_84_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1]> sum_1_to_fp16 = cast(dtype = _to_copy_1_to_fp16_dtype_0, x = sum_1)[name = string("cast_3")];
            tensor<fp16, [1]> add_9_cast_fp16 = add(x = sum_1_to_fp16, y = const_84_promoted_to_fp16)[name = string("add_9_cast_fp16")];
            fp16 _inversed_div_y_0_to_fp16 = const()[name = string("_inversed_div_y_0_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1]> _inversed_div_cast_fp16 = mul(x = add_9_cast_fp16, y = _inversed_div_y_0_to_fp16)[name = string("_inversed_div_cast_fp16")];
            fp16 const_86_to_fp16 = const()[name = string("const_86_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> add_10_cast_fp16 = add(x = _inversed_div_cast_fp16, y = const_86_to_fp16)[name = string("add_10_cast_fp16")];
            tensor<fp16, [1]> floor_cast_fp16 = floor(x = add_10_cast_fp16)[name = string("floor_cast_fp16")];
            fp16 const_88_promoted_to_fp16 = const()[name = string("const_88_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1]> add_11_cast_fp16 = add(x = floor_cast_fp16, y = const_88_promoted_to_fp16)[name = string("add_11_cast_fp16")];
            fp16 _inversed_div_1_y_0_to_fp16 = const()[name = string("_inversed_div_1_y_0_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1]> _inversed_div_1_cast_fp16 = mul(x = add_11_cast_fp16, y = _inversed_div_1_y_0_to_fp16)[name = string("_inversed_div_1_cast_fp16")];
            fp16 const_90_to_fp16 = const()[name = string("const_90_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> add_12_cast_fp16 = add(x = _inversed_div_1_cast_fp16, y = const_90_to_fp16)[name = string("add_12_cast_fp16")];
            tensor<fp16, [1]> floor_1_cast_fp16 = floor(x = add_12_cast_fp16)[name = string("floor_1_cast_fp16")];
            fp16 const_92_promoted_to_fp16 = const()[name = string("const_92_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1]> add_13_cast_fp16 = add(x = floor_1_cast_fp16, y = const_92_promoted_to_fp16)[name = string("add_13_cast_fp16")];
            fp16 _inversed_div_2_y_0_to_fp16 = const()[name = string("_inversed_div_2_y_0_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1]> _inversed_div_2_cast_fp16 = mul(x = add_13_cast_fp16, y = _inversed_div_2_y_0_to_fp16)[name = string("_inversed_div_2_cast_fp16")];
            fp16 const_94_to_fp16 = const()[name = string("const_94_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> add_14_cast_fp16 = add(x = _inversed_div_2_cast_fp16, y = const_94_to_fp16)[name = string("add_14_cast_fp16")];
            tensor<fp16, [1]> floor_2_cast_fp16 = floor(x = add_14_cast_fp16)[name = string("floor_2_cast_fp16")];
            string _to_copy_2_dtype_0 = const()[name = string("_to_copy_2_dtype_0"), val = string("int32")];
            tensor<int32, [188]> arange_6 = const()[name = string("arange_6"), val = tensor<int32, [188]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3276352)))];
            tensor<int32, [1]> unsqueeze_20_axes_0 = const()[name = string("unsqueeze_20_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> floor_2_cast_fp16_to_int32 = cast(dtype = _to_copy_2_dtype_0, x = floor_2_cast_fp16)[name = string("cast_2")];
            tensor<int32, [1, 1]> unsqueeze_20 = expand_dims(axes = unsqueeze_20_axes_0, x = floor_2_cast_fp16_to_int32)[name = string("unsqueeze_20")];
            tensor<bool, [1, 188]> lt_5 = less(x = arange_6, y = unsqueeze_20)[name = string("lt_5")];
            tensor<int32, [1]> unsqueeze_21_axes_0 = const()[name = string("unsqueeze_21_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<bool, [1, 1, 188]> unsqueeze_21 = expand_dims(axes = unsqueeze_21_axes_0, x = lt_5)[name = string("unsqueeze_21")];
            tensor<int32, [3]> expand_1_reps_0 = const()[name = string("expand_1_reps_0"), val = tensor<int32, [3]>([1, 188, 1])];
            tensor<bool, [1, 188, 188]> expand_1 = tile(reps = expand_1_reps_0, x = unsqueeze_21)[name = string("expand_1")];
            tensor<int32, [3]> transpose_2_perm_0 = const()[name = string("transpose_2_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<bool, [1, 188, 188]> transpose_2 = transpose(perm = transpose_2_perm_0, x = expand_1)[name = string("transpose_144")];
            tensor<bool, [1, 188, 188]> and_1 = logical_and(x = expand_1, y = transpose_2)[name = string("and_1")];
            tensor<int32, [1]> unsqueeze_22_axes_0 = const()[name = string("unsqueeze_22_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<bool, [1, 1, 188, 188]> unsqueeze_22 = expand_dims(axes = unsqueeze_22_axes_0, x = and_1)[name = string("unsqueeze_22")];
            tensor<int32, [1]> layer_norm_axes_0 = const()[name = string("layer_norm_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_0_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_0_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3277184)))];
            tensor<fp16, [1024]> p_encoder_layers_0_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_0_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3279296)))];
            fp16 const_107_to_fp16 = const()[name = string("const_107_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_cast_fp16 = layer_norm(axes = layer_norm_axes_0, beta = p_encoder_layers_0_norm_feed_forward1_bias_to_fp16, epsilon = const_107_to_fp16, gamma = p_encoder_layers_0_norm_feed_forward1_weight_to_fp16, x = linear_0_cast_fp16)[name = string("layer_norm_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_0_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3281408))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6427200))))[name = string("p_encoder_layers_0_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6460032)))];
            tensor<fp16, [1, 188, 4096]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_0_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_cast_fp16)[name = string("linear_1_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_cast_fp16 = silu(x = linear_1_cast_fp16)[name = string("silu_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_0_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6468288))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9614080))))[name = string("p_encoder_layers_0_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> linear_2_bias_0_to_fp16 = const()[name = string("linear_2_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9622336)))];
            tensor<fp16, [1, 188, 1024]> linear_2_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_0_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_cast_fp16)[name = string("linear_2_cast_fp16")];
            fp16 const_109_to_fp16 = const()[name = string("const_109_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_6_cast_fp16 = mul(x = linear_2_cast_fp16, y = const_109_to_fp16)[name = string("mul_6_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_15_cast_fp16 = add(x = linear_0_cast_fp16, y = mul_6_cast_fp16)[name = string("add_15_cast_fp16")];
            tensor<int32, [1]> layer_norm_1_axes_0 = const()[name = string("layer_norm_1_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_0_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_0_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9624448)))];
            tensor<fp16, [1024]> p_encoder_layers_0_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_0_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9626560)))];
            fp16 const_111_to_fp16 = const()[name = string("const_111_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_1_cast_fp16 = layer_norm(axes = layer_norm_1_axes_0, beta = p_encoder_layers_0_norm_self_att_bias_to_fp16, epsilon = const_111_to_fp16, gamma = p_encoder_layers_0_norm_self_att_weight_to_fp16, x = add_15_cast_fp16)[name = string("layer_norm_1_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9628672))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10415168))))[name = string("p_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_3_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_1_cast_fp16)[name = string("linear_3_cast_fp16")];
            tensor<int32, [4]> const_113 = const()[name = string("const_113"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_1_cast_fp16 = reshape(shape = const_113, x = linear_3_cast_fp16)[name = string("view_1_cast_fp16")];
            tensor<int32, [4]> transpose_3_perm_0 = const()[name = string("transpose_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10423424))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11209920))))[name = string("p_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_4_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_1_cast_fp16)[name = string("linear_4_cast_fp16")];
            tensor<int32, [4]> const_116 = const()[name = string("const_116"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_2_cast_fp16 = reshape(shape = const_116, x = linear_4_cast_fp16)[name = string("view_2_cast_fp16")];
            tensor<int32, [4]> transpose_4_perm_0 = const()[name = string("transpose_4_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11218176))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12004672))))[name = string("p_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_5_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_1_cast_fp16)[name = string("linear_5_cast_fp16")];
            tensor<int32, [4]> const_119 = const()[name = string("const_119"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_3_cast_fp16 = reshape(shape = const_119, x = linear_5_cast_fp16)[name = string("view_3_cast_fp16")];
            tensor<int32, [4]> transpose_5_perm_0 = const()[name = string("transpose_5_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_4_to_fp16 = const()[name = string("view_4_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12012928)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_3_cast_fp16 = transpose(perm = transpose_3_perm_0, x = view_1_cast_fp16)[name = string("transpose_143")];
            tensor<fp16, [1, 8, 188, 128]> add_16_cast_fp16 = add(x = transpose_3_cast_fp16, y = view_4_to_fp16)[name = string("add_16_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_5_to_fp16 = const()[name = string("view_5_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12015040)))];
            tensor<fp16, [1, 8, 188, 128]> add_17_cast_fp16 = add(x = transpose_3_cast_fp16, y = view_5_to_fp16)[name = string("add_17_cast_fp16")];
            bool matmul_1_transpose_x_0 = const()[name = string("matmul_1_transpose_x_0"), val = bool(false)];
            bool matmul_1_transpose_y_0 = const()[name = string("matmul_1_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_to_fp16 = const()[name = string("permute_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12017152)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_1_cast_fp16 = matmul(transpose_x = matmul_1_transpose_x_0, transpose_y = matmul_1_transpose_y_0, x = add_17_cast_fp16, y = permute_to_fp16)[name = string("matmul_1_cast_fp16")];
            tensor<int32, [8]> pad_pad_0 = const()[name = string("pad_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_mode_0 = const()[name = string("pad_mode_0"), val = string("constant")];
            fp16 const_127_to_fp16 = const()[name = string("const_127_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_cast_fp16 = pad(constant_val = const_127_to_fp16, mode = pad_mode_0, pad = pad_pad_0, x = matmul_1_cast_fp16)[name = string("pad_cast_fp16")];
            tensor<int32, [4]> const_128 = const()[name = string("const_128"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_7_cast_fp16 = reshape(shape = const_128, x = pad_cast_fp16)[name = string("view_7_cast_fp16")];
            tensor<int32, [4]> slice_1_begin_0 = const()[name = string("slice_1_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_1_end_0 = const()[name = string("slice_1_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_1_end_mask_0 = const()[name = string("slice_1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_1_cast_fp16 = slice_by_index(begin = slice_1_begin_0, end = slice_1_end_0, end_mask = slice_1_end_mask_0, x = view_7_cast_fp16)[name = string("slice_1_cast_fp16")];
            tensor<int32, [4]> const_132 = const()[name = string("const_132"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_8_cast_fp16 = reshape(shape = const_132, x = slice_1_cast_fp16)[name = string("view_8_cast_fp16")];
            tensor<int32, [4]> slice_2_begin_0 = const()[name = string("slice_2_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_2_end_0 = const()[name = string("slice_2_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_2_end_mask_0 = const()[name = string("slice_2_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_2_cast_fp16 = slice_by_index(begin = slice_2_begin_0, end = slice_2_end_0, end_mask = slice_2_end_mask_0, x = view_8_cast_fp16)[name = string("slice_2_cast_fp16")];
            fp16 const_136_to_fp16 = const()[name = string("const_136_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_7_cast_fp16 = mul(x = slice_2_cast_fp16, y = const_136_to_fp16)[name = string("mul_7_cast_fp16")];
            tensor<bool, [1, 1, 188, 188]> logical_not = logical_not(x = unsqueeze_22)[name = string("logical_not")];
            fp16 const_137_to_fp16 = const()[name = string("const_137_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_cast_fp16 = select(a = const_137_to_fp16, b = mul_7_cast_fp16, cond = logical_not)[name = string("masked_fill_cast_fp16")];
            fp16 const_138_to_fp16 = const()[name = string("const_138_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_0_cast_fp16 = mul(x = add_16_cast_fp16, y = const_138_to_fp16)[name = string("mul_0_cast_fp16")];
            bool matmul_0_transpose_y_0 = const()[name = string("matmul_0_transpose_y_0"), val = bool(true)];
            bool matmul_0_transpose_x_0 = const()[name = string("matmul_0_transpose_x_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_4_cast_fp16 = transpose(perm = transpose_4_perm_0, x = view_2_cast_fp16)[name = string("transpose_142")];
            tensor<fp16, [1, 8, 188, 188]> matmul_0_cast_fp16 = matmul(transpose_x = matmul_0_transpose_x_0, transpose_y = matmul_0_transpose_y_0, x = mul_0_cast_fp16, y = transpose_4_cast_fp16)[name = string("matmul_0_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_0_cast_fp16 = add(x = matmul_0_cast_fp16, y = masked_fill_cast_fp16)[name = string("add_0_cast_fp16")];
            int32 softmax_0_axis_0 = const()[name = string("softmax_0_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_0_cast_fp16 = softmax(axis = softmax_0_axis_0, x = add_0_cast_fp16)[name = string("softmax_0_cast_fp16")];
            bool scaled_dot_product_attention_transpose_x_0 = const()[name = string("scaled_dot_product_attention_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_transpose_y_0 = const()[name = string("scaled_dot_product_attention_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_5_cast_fp16 = transpose(perm = transpose_5_perm_0, x = view_3_cast_fp16)[name = string("transpose_141")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_transpose_x_0, transpose_y = scaled_dot_product_attention_transpose_y_0, x = softmax_0_cast_fp16, y = transpose_5_cast_fp16)[name = string("scaled_dot_product_attention_cast_fp16")];
            tensor<int32, [4]> transpose_6_perm_0 = const()[name = string("transpose_6_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_141 = const()[name = string("const_141"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_6_cast_fp16 = transpose(perm = transpose_6_perm_0, x = scaled_dot_product_attention_cast_fp16)[name = string("transpose_140")];
            tensor<fp16, [1, 188, 1024]> view_9_cast_fp16 = reshape(shape = const_141, x = transpose_6_cast_fp16)[name = string("view_9_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_0_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12785216))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13571712))))[name = string("p_encoder_layers_0_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_7_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_0_self_attn_o_proj_weight_to_fp16_palettized, x = view_9_cast_fp16)[name = string("linear_7_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_18_cast_fp16 = add(x = add_15_cast_fp16, y = linear_7_cast_fp16)[name = string("add_18_cast_fp16")];
            tensor<int32, [1]> layer_norm_2_axes_0 = const()[name = string("layer_norm_2_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_0_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_0_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13579968)))];
            tensor<fp16, [1024]> p_encoder_layers_0_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_0_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13582080)))];
            fp16 const_143_to_fp16 = const()[name = string("const_143_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_2_cast_fp16 = layer_norm(axes = layer_norm_2_axes_0, beta = p_encoder_layers_0_norm_conv_bias_to_fp16, epsilon = const_143_to_fp16, gamma = p_encoder_layers_0_norm_conv_weight_to_fp16, x = add_18_cast_fp16)[name = string("layer_norm_2_cast_fp16")];
            tensor<int32, [3]> transpose_7_perm_0 = const()[name = string("transpose_7_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_pad_type_0 = const()[name = string("conv1d_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_strides_0 = const()[name = string("conv1d_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_pad_0 = const()[name = string("conv1d_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_dilations_0 = const()[name = string("conv1d_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_groups_0 = const()[name = string("conv1d_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_0_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13584192))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15157120))))[name = string("p_encoder_layers_0_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_7_cast_fp16 = transpose(perm = transpose_7_perm_0, x = layer_norm_2_cast_fp16)[name = string("transpose_139")];
            tensor<fp16, [1, 2048, 188]> conv1d_cast_fp16 = conv(dilations = conv1d_dilations_0, groups = conv1d_groups_0, pad = conv1d_pad_0, pad_type = conv1d_pad_type_0, strides = conv1d_strides_0, weight = p_encoder_layers_0_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_7_cast_fp16)[name = string("conv1d_cast_fp16")];
            int32 glu_split_num_splits_0 = const()[name = string("glu_split_num_splits_0"), val = int32(2)];
            int32 glu_split_axis_0 = const()[name = string("glu_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_split_cast_fp16_1 = split(axis = glu_split_axis_0, num_splits = glu_split_num_splits_0, x = conv1d_cast_fp16)[name = string("glu_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_split_cast_fp16_1)[name = string("glu_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_cast_fp16 = mul(x = glu_split_cast_fp16_0, y = glu_split_1_sigmoid_cast_fp16)[name = string("glu_cast_fp16")];
            tensor<int32, [1]> const_148_list = const()[name = string("const_148_list"), val = tensor<int32, [1]>([2])];
            string cast_19_dtype_0 = const()[name = string("cast_19_dtype_0"), val = string("int32")];
            bool reduce_min_0_keep_dims_0 = const()[name = string("reduce_min_0_keep_dims_0"), val = bool(false)];
            tensor<int32, [1, 1, 188, 188]> cast_19 = cast(dtype = cast_19_dtype_0, x = logical_not)[name = string("cast_1")];
            tensor<int32, [1, 1, 188]> reduce_min_0 = reduce_min(axes = const_148_list, keep_dims = reduce_min_0_keep_dims_0, x = cast_19)[name = string("reduce_min_0")];
            string all_1_dtype_0 = const()[name = string("all_1_dtype_0"), val = string("bool")];
            fp16 const_149_to_fp16 = const()[name = string("const_149_to_fp16"), val = fp16(0x0p+0)];
            tensor<bool, [1, 1, 188]> all_1 = cast(dtype = all_1_dtype_0, x = reduce_min_0)[name = string("cast_0")];
            tensor<fp16, [1, 1024, 188]> masked_fill_1_cast_fp16 = select(a = const_149_to_fp16, b = glu_cast_fp16, cond = all_1)[name = string("masked_fill_1_cast_fp16")];
            string conv1d_1_pad_type_0 = const()[name = string("conv1d_1_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_1_pad_0 = const()[name = string("conv1d_1_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_1_groups_0 = const()[name = string("conv1d_1_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_1_strides_0 = const()[name = string("conv1d_1_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_1_dilations_0 = const()[name = string("conv1d_1_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1547_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15173568))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15180544))))[name = string("const_1547_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1548_to_fp16 = const()[name = string("const_1548_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15188800)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_cast_fp16 = conv(bias = const_1548_to_fp16, dilations = conv1d_1_dilations_0, groups = conv1d_1_groups_0, pad = conv1d_1_pad_0, pad_type = conv1d_1_pad_type_0, strides = conv1d_1_strides_0, weight = const_1547_to_fp16_palettized, x = masked_fill_1_cast_fp16)[name = string("_native_batch_norm_legit_no_training_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_1_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_cast_fp16)[name = string("silu_1_cast_fp16")];
            string conv1d_2_pad_type_0 = const()[name = string("conv1d_2_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_2_strides_0 = const()[name = string("conv1d_2_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_2_pad_0 = const()[name = string("conv1d_2_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_2_dilations_0 = const()[name = string("conv1d_2_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_2_groups_0 = const()[name = string("conv1d_2_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_0_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15190912))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15977408))))[name = string("p_encoder_layers_0_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_2_cast_fp16 = conv(dilations = conv1d_2_dilations_0, groups = conv1d_2_groups_0, pad = conv1d_2_pad_0, pad_type = conv1d_2_pad_type_0, strides = conv1d_2_strides_0, weight = p_encoder_layers_0_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_1_cast_fp16)[name = string("conv1d_2_cast_fp16")];
            tensor<int32, [3]> transpose_8_perm_0 = const()[name = string("transpose_8_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_8_cast_fp16 = transpose(perm = transpose_8_perm_0, x = conv1d_2_cast_fp16)[name = string("transpose_138")];
            tensor<fp16, [1, 188, 1024]> add_19_cast_fp16 = add(x = add_18_cast_fp16, y = transpose_8_cast_fp16)[name = string("add_19_cast_fp16")];
            tensor<int32, [1]> layer_norm_3_axes_0 = const()[name = string("layer_norm_3_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_0_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_0_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15985664)))];
            tensor<fp16, [1024]> p_encoder_layers_0_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_0_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15987776)))];
            fp16 const_160_to_fp16 = const()[name = string("const_160_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_3_cast_fp16 = layer_norm(axes = layer_norm_3_axes_0, beta = p_encoder_layers_0_norm_feed_forward2_bias_to_fp16, epsilon = const_160_to_fp16, gamma = p_encoder_layers_0_norm_feed_forward2_weight_to_fp16, x = add_19_cast_fp16)[name = string("layer_norm_3_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_0_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15989888))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19135680))))[name = string("p_encoder_layers_0_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_8_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_0_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_3_cast_fp16)[name = string("linear_8_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_2_cast_fp16 = silu(x = linear_8_cast_fp16)[name = string("silu_2_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_0_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19168512))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22314304))))[name = string("p_encoder_layers_0_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_9_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_0_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_2_cast_fp16)[name = string("linear_9_cast_fp16")];
            fp16 const_162_to_fp16 = const()[name = string("const_162_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_8_cast_fp16 = mul(x = linear_9_cast_fp16, y = const_162_to_fp16)[name = string("mul_8_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_20_cast_fp16 = add(x = add_19_cast_fp16, y = mul_8_cast_fp16)[name = string("add_20_cast_fp16")];
            tensor<int32, [1]> layer_norm_4_axes_0 = const()[name = string("layer_norm_4_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_0_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_0_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22322560)))];
            tensor<fp16, [1024]> p_encoder_layers_0_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_0_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22324672)))];
            fp16 const_164_to_fp16 = const()[name = string("const_164_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_4_cast_fp16 = layer_norm(axes = layer_norm_4_axes_0, beta = p_encoder_layers_0_norm_out_bias_to_fp16, epsilon = const_164_to_fp16, gamma = p_encoder_layers_0_norm_out_weight_to_fp16, x = add_20_cast_fp16)[name = string("layer_norm_4_cast_fp16")];
            tensor<int32, [1]> layer_norm_5_axes_0 = const()[name = string("layer_norm_5_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_1_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_1_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22326784)))];
            tensor<fp16, [1024]> p_encoder_layers_1_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_1_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22328896)))];
            fp16 const_167_to_fp16 = const()[name = string("const_167_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_5_cast_fp16 = layer_norm(axes = layer_norm_5_axes_0, beta = p_encoder_layers_1_norm_feed_forward1_bias_to_fp16, epsilon = const_167_to_fp16, gamma = p_encoder_layers_1_norm_feed_forward1_weight_to_fp16, x = layer_norm_4_cast_fp16)[name = string("layer_norm_5_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_1_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22331008))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25476800))))[name = string("p_encoder_layers_1_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_10_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_1_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_5_cast_fp16)[name = string("linear_10_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_3_cast_fp16 = silu(x = linear_10_cast_fp16)[name = string("silu_3_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_1_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25509632))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28655424))))[name = string("p_encoder_layers_1_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_11_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_1_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_3_cast_fp16)[name = string("linear_11_cast_fp16")];
            fp16 const_169_to_fp16 = const()[name = string("const_169_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_9_cast_fp16 = mul(x = linear_11_cast_fp16, y = const_169_to_fp16)[name = string("mul_9_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_21_cast_fp16 = add(x = layer_norm_4_cast_fp16, y = mul_9_cast_fp16)[name = string("add_21_cast_fp16")];
            tensor<int32, [1]> layer_norm_6_axes_0 = const()[name = string("layer_norm_6_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_1_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_1_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28663680)))];
            tensor<fp16, [1024]> p_encoder_layers_1_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_1_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28665792)))];
            fp16 const_171_to_fp16 = const()[name = string("const_171_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_6_cast_fp16 = layer_norm(axes = layer_norm_6_axes_0, beta = p_encoder_layers_1_norm_self_att_bias_to_fp16, epsilon = const_171_to_fp16, gamma = p_encoder_layers_1_norm_self_att_weight_to_fp16, x = add_21_cast_fp16)[name = string("layer_norm_6_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28667904))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29454400))))[name = string("p_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_12_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_6_cast_fp16)[name = string("linear_12_cast_fp16")];
            tensor<int32, [4]> const_173 = const()[name = string("const_173"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_10_cast_fp16 = reshape(shape = const_173, x = linear_12_cast_fp16)[name = string("view_10_cast_fp16")];
            tensor<int32, [4]> transpose_9_perm_0 = const()[name = string("transpose_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29462656))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30249152))))[name = string("p_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_13_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_6_cast_fp16)[name = string("linear_13_cast_fp16")];
            tensor<int32, [4]> const_176 = const()[name = string("const_176"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_11_cast_fp16 = reshape(shape = const_176, x = linear_13_cast_fp16)[name = string("view_11_cast_fp16")];
            tensor<int32, [4]> transpose_10_perm_0 = const()[name = string("transpose_10_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30257408))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31043904))))[name = string("p_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_14_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_6_cast_fp16)[name = string("linear_14_cast_fp16")];
            tensor<int32, [4]> const_179 = const()[name = string("const_179"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_12_cast_fp16 = reshape(shape = const_179, x = linear_14_cast_fp16)[name = string("view_12_cast_fp16")];
            tensor<int32, [4]> transpose_11_perm_0 = const()[name = string("transpose_11_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_13_to_fp16 = const()[name = string("view_13_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31052160)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_9_cast_fp16 = transpose(perm = transpose_9_perm_0, x = view_10_cast_fp16)[name = string("transpose_137")];
            tensor<fp16, [1, 8, 188, 128]> add_22_cast_fp16 = add(x = transpose_9_cast_fp16, y = view_13_to_fp16)[name = string("add_22_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_14_to_fp16 = const()[name = string("view_14_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31054272)))];
            tensor<fp16, [1, 8, 188, 128]> add_23_cast_fp16 = add(x = transpose_9_cast_fp16, y = view_14_to_fp16)[name = string("add_23_cast_fp16")];
            bool matmul_2_transpose_x_0 = const()[name = string("matmul_2_transpose_x_0"), val = bool(false)];
            bool matmul_2_transpose_y_0 = const()[name = string("matmul_2_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_1_to_fp16 = const()[name = string("permute_1_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31056384)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_2_cast_fp16 = matmul(transpose_x = matmul_2_transpose_x_0, transpose_y = matmul_2_transpose_y_0, x = add_23_cast_fp16, y = permute_1_to_fp16)[name = string("matmul_2_cast_fp16")];
            tensor<int32, [8]> pad_1_pad_0 = const()[name = string("pad_1_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_1_mode_0 = const()[name = string("pad_1_mode_0"), val = string("constant")];
            fp16 const_187_to_fp16 = const()[name = string("const_187_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_1_cast_fp16 = pad(constant_val = const_187_to_fp16, mode = pad_1_mode_0, pad = pad_1_pad_0, x = matmul_2_cast_fp16)[name = string("pad_1_cast_fp16")];
            tensor<int32, [4]> const_188 = const()[name = string("const_188"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_16_cast_fp16 = reshape(shape = const_188, x = pad_1_cast_fp16)[name = string("view_16_cast_fp16")];
            tensor<int32, [4]> slice_3_begin_0 = const()[name = string("slice_3_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_3_end_0 = const()[name = string("slice_3_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_3_end_mask_0 = const()[name = string("slice_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_3_cast_fp16 = slice_by_index(begin = slice_3_begin_0, end = slice_3_end_0, end_mask = slice_3_end_mask_0, x = view_16_cast_fp16)[name = string("slice_3_cast_fp16")];
            tensor<int32, [4]> const_192 = const()[name = string("const_192"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_17_cast_fp16 = reshape(shape = const_192, x = slice_3_cast_fp16)[name = string("view_17_cast_fp16")];
            tensor<int32, [4]> slice_4_begin_0 = const()[name = string("slice_4_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_4_end_0 = const()[name = string("slice_4_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_4_end_mask_0 = const()[name = string("slice_4_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_4_cast_fp16 = slice_by_index(begin = slice_4_begin_0, end = slice_4_end_0, end_mask = slice_4_end_mask_0, x = view_17_cast_fp16)[name = string("slice_4_cast_fp16")];
            fp16 const_196_to_fp16 = const()[name = string("const_196_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_10_cast_fp16 = mul(x = slice_4_cast_fp16, y = const_196_to_fp16)[name = string("mul_10_cast_fp16")];
            fp16 const_197_to_fp16 = const()[name = string("const_197_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_2_cast_fp16 = select(a = const_197_to_fp16, b = mul_10_cast_fp16, cond = logical_not)[name = string("masked_fill_2_cast_fp16")];
            fp16 const_198_to_fp16 = const()[name = string("const_198_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_1_1_cast_fp16 = mul(x = add_22_cast_fp16, y = const_198_to_fp16)[name = string("mul_1_1_cast_fp16")];
            bool matmul_1_transpose_y_1 = const()[name = string("matmul_1_transpose_y_1"), val = bool(true)];
            bool matmul_1_transpose_x_1 = const()[name = string("matmul_1_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_10_cast_fp16 = transpose(perm = transpose_10_perm_0, x = view_11_cast_fp16)[name = string("transpose_136")];
            tensor<fp16, [1, 8, 188, 188]> matmul_1_1_cast_fp16 = matmul(transpose_x = matmul_1_transpose_x_1, transpose_y = matmul_1_transpose_y_1, x = mul_1_1_cast_fp16, y = transpose_10_cast_fp16)[name = string("matmul_1_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_1_1_cast_fp16 = add(x = matmul_1_1_cast_fp16, y = masked_fill_2_cast_fp16)[name = string("add_1_1_cast_fp16")];
            int32 softmax_1_axis_0 = const()[name = string("softmax_1_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_1_cast_fp16 = softmax(axis = softmax_1_axis_0, x = add_1_1_cast_fp16)[name = string("softmax_1_cast_fp16")];
            bool scaled_dot_product_attention_1_transpose_x_0 = const()[name = string("scaled_dot_product_attention_1_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_1_transpose_y_0 = const()[name = string("scaled_dot_product_attention_1_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_11_cast_fp16 = transpose(perm = transpose_11_perm_0, x = view_12_cast_fp16)[name = string("transpose_135")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_1_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_1_transpose_x_0, transpose_y = scaled_dot_product_attention_1_transpose_y_0, x = softmax_1_cast_fp16, y = transpose_11_cast_fp16)[name = string("scaled_dot_product_attention_1_cast_fp16")];
            tensor<int32, [4]> transpose_12_perm_0 = const()[name = string("transpose_12_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_201 = const()[name = string("const_201"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_12_cast_fp16 = transpose(perm = transpose_12_perm_0, x = scaled_dot_product_attention_1_cast_fp16)[name = string("transpose_134")];
            tensor<fp16, [1, 188, 1024]> view_18_cast_fp16 = reshape(shape = const_201, x = transpose_12_cast_fp16)[name = string("view_18_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_1_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31824448))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32610944))))[name = string("p_encoder_layers_1_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_16_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_1_self_attn_o_proj_weight_to_fp16_palettized, x = view_18_cast_fp16)[name = string("linear_16_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_24_cast_fp16 = add(x = add_21_cast_fp16, y = linear_16_cast_fp16)[name = string("add_24_cast_fp16")];
            tensor<int32, [1]> layer_norm_7_axes_0 = const()[name = string("layer_norm_7_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_1_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_1_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32619200)))];
            tensor<fp16, [1024]> p_encoder_layers_1_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_1_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32621312)))];
            fp16 const_203_to_fp16 = const()[name = string("const_203_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_7_cast_fp16 = layer_norm(axes = layer_norm_7_axes_0, beta = p_encoder_layers_1_norm_conv_bias_to_fp16, epsilon = const_203_to_fp16, gamma = p_encoder_layers_1_norm_conv_weight_to_fp16, x = add_24_cast_fp16)[name = string("layer_norm_7_cast_fp16")];
            tensor<int32, [3]> transpose_13_perm_0 = const()[name = string("transpose_13_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_3_pad_type_0 = const()[name = string("conv1d_3_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_3_strides_0 = const()[name = string("conv1d_3_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_3_pad_0 = const()[name = string("conv1d_3_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_3_dilations_0 = const()[name = string("conv1d_3_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_3_groups_0 = const()[name = string("conv1d_3_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_1_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32623424))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34196352))))[name = string("p_encoder_layers_1_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_13_cast_fp16 = transpose(perm = transpose_13_perm_0, x = layer_norm_7_cast_fp16)[name = string("transpose_133")];
            tensor<fp16, [1, 2048, 188]> conv1d_3_cast_fp16 = conv(dilations = conv1d_3_dilations_0, groups = conv1d_3_groups_0, pad = conv1d_3_pad_0, pad_type = conv1d_3_pad_type_0, strides = conv1d_3_strides_0, weight = p_encoder_layers_1_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_13_cast_fp16)[name = string("conv1d_3_cast_fp16")];
            int32 glu_1_split_num_splits_0 = const()[name = string("glu_1_split_num_splits_0"), val = int32(2)];
            int32 glu_1_split_axis_0 = const()[name = string("glu_1_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_1_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_1_split_cast_fp16_1 = split(axis = glu_1_split_axis_0, num_splits = glu_1_split_num_splits_0, x = conv1d_3_cast_fp16)[name = string("glu_1_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_1_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_1_split_cast_fp16_1)[name = string("glu_1_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_1_cast_fp16 = mul(x = glu_1_split_cast_fp16_0, y = glu_1_split_1_sigmoid_cast_fp16)[name = string("glu_1_cast_fp16")];
            fp16 const_209_to_fp16 = const()[name = string("const_209_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_3_cast_fp16 = select(a = const_209_to_fp16, b = glu_1_cast_fp16, cond = all_1)[name = string("masked_fill_3_cast_fp16")];
            string conv1d_4_pad_type_0 = const()[name = string("conv1d_4_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_4_pad_0 = const()[name = string("conv1d_4_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_4_groups_0 = const()[name = string("conv1d_4_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_4_strides_0 = const()[name = string("conv1d_4_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_4_dilations_0 = const()[name = string("conv1d_4_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1549_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34212800))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34219776))))[name = string("const_1549_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1550_to_fp16 = const()[name = string("const_1550_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34228032)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_1_cast_fp16 = conv(bias = const_1550_to_fp16, dilations = conv1d_4_dilations_0, groups = conv1d_4_groups_0, pad = conv1d_4_pad_0, pad_type = conv1d_4_pad_type_0, strides = conv1d_4_strides_0, weight = const_1549_to_fp16_palettized, x = masked_fill_3_cast_fp16)[name = string("_native_batch_norm_legit_no_training_1_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_4_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_1_cast_fp16)[name = string("silu_4_cast_fp16")];
            string conv1d_5_pad_type_0 = const()[name = string("conv1d_5_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_5_strides_0 = const()[name = string("conv1d_5_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_5_pad_0 = const()[name = string("conv1d_5_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_5_dilations_0 = const()[name = string("conv1d_5_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_5_groups_0 = const()[name = string("conv1d_5_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_1_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34230144))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35016640))))[name = string("p_encoder_layers_1_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_5_cast_fp16 = conv(dilations = conv1d_5_dilations_0, groups = conv1d_5_groups_0, pad = conv1d_5_pad_0, pad_type = conv1d_5_pad_type_0, strides = conv1d_5_strides_0, weight = p_encoder_layers_1_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_4_cast_fp16)[name = string("conv1d_5_cast_fp16")];
            tensor<int32, [3]> transpose_14_perm_0 = const()[name = string("transpose_14_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_14_cast_fp16 = transpose(perm = transpose_14_perm_0, x = conv1d_5_cast_fp16)[name = string("transpose_132")];
            tensor<fp16, [1, 188, 1024]> add_25_cast_fp16 = add(x = add_24_cast_fp16, y = transpose_14_cast_fp16)[name = string("add_25_cast_fp16")];
            tensor<int32, [1]> layer_norm_8_axes_0 = const()[name = string("layer_norm_8_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_1_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_1_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35024896)))];
            tensor<fp16, [1024]> p_encoder_layers_1_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_1_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35027008)))];
            fp16 const_220_to_fp16 = const()[name = string("const_220_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_8_cast_fp16 = layer_norm(axes = layer_norm_8_axes_0, beta = p_encoder_layers_1_norm_feed_forward2_bias_to_fp16, epsilon = const_220_to_fp16, gamma = p_encoder_layers_1_norm_feed_forward2_weight_to_fp16, x = add_25_cast_fp16)[name = string("layer_norm_8_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_1_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35029120))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38174912))))[name = string("p_encoder_layers_1_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_17_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_1_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_8_cast_fp16)[name = string("linear_17_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_5_cast_fp16 = silu(x = linear_17_cast_fp16)[name = string("silu_5_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_1_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38207744))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41353536))))[name = string("p_encoder_layers_1_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_18_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_1_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_5_cast_fp16)[name = string("linear_18_cast_fp16")];
            fp16 const_222_to_fp16 = const()[name = string("const_222_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_11_cast_fp16 = mul(x = linear_18_cast_fp16, y = const_222_to_fp16)[name = string("mul_11_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_26_cast_fp16 = add(x = add_25_cast_fp16, y = mul_11_cast_fp16)[name = string("add_26_cast_fp16")];
            tensor<int32, [1]> layer_norm_9_axes_0 = const()[name = string("layer_norm_9_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_1_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_1_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41361792)))];
            tensor<fp16, [1024]> p_encoder_layers_1_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_1_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41363904)))];
            fp16 const_224_to_fp16 = const()[name = string("const_224_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_9_cast_fp16 = layer_norm(axes = layer_norm_9_axes_0, beta = p_encoder_layers_1_norm_out_bias_to_fp16, epsilon = const_224_to_fp16, gamma = p_encoder_layers_1_norm_out_weight_to_fp16, x = add_26_cast_fp16)[name = string("layer_norm_9_cast_fp16")];
            tensor<int32, [1]> layer_norm_10_axes_0 = const()[name = string("layer_norm_10_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_2_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_2_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41366016)))];
            tensor<fp16, [1024]> p_encoder_layers_2_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_2_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41368128)))];
            fp16 const_227_to_fp16 = const()[name = string("const_227_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_10_cast_fp16 = layer_norm(axes = layer_norm_10_axes_0, beta = p_encoder_layers_2_norm_feed_forward1_bias_to_fp16, epsilon = const_227_to_fp16, gamma = p_encoder_layers_2_norm_feed_forward1_weight_to_fp16, x = layer_norm_9_cast_fp16)[name = string("layer_norm_10_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_2_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41370240))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44516032))))[name = string("p_encoder_layers_2_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_2_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_10_cast_fp16)[name = string("linear_19_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_6_cast_fp16 = silu(x = linear_19_cast_fp16)[name = string("silu_6_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_2_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44548864))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47694656))))[name = string("p_encoder_layers_2_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_20_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_2_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_6_cast_fp16)[name = string("linear_20_cast_fp16")];
            fp16 const_229_to_fp16 = const()[name = string("const_229_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_12_cast_fp16 = mul(x = linear_20_cast_fp16, y = const_229_to_fp16)[name = string("mul_12_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_27_cast_fp16 = add(x = layer_norm_9_cast_fp16, y = mul_12_cast_fp16)[name = string("add_27_cast_fp16")];
            tensor<int32, [1]> layer_norm_11_axes_0 = const()[name = string("layer_norm_11_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_2_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_2_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47702912)))];
            tensor<fp16, [1024]> p_encoder_layers_2_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_2_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47705024)))];
            fp16 const_231_to_fp16 = const()[name = string("const_231_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_11_cast_fp16 = layer_norm(axes = layer_norm_11_axes_0, beta = p_encoder_layers_2_norm_self_att_bias_to_fp16, epsilon = const_231_to_fp16, gamma = p_encoder_layers_2_norm_self_att_weight_to_fp16, x = add_27_cast_fp16)[name = string("layer_norm_11_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47707136))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48493632))))[name = string("p_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_21_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_11_cast_fp16)[name = string("linear_21_cast_fp16")];
            tensor<int32, [4]> const_233 = const()[name = string("const_233"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_19_cast_fp16 = reshape(shape = const_233, x = linear_21_cast_fp16)[name = string("view_19_cast_fp16")];
            tensor<int32, [4]> transpose_15_perm_0 = const()[name = string("transpose_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48501888))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49288384))))[name = string("p_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_22_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_11_cast_fp16)[name = string("linear_22_cast_fp16")];
            tensor<int32, [4]> const_236 = const()[name = string("const_236"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_20_cast_fp16 = reshape(shape = const_236, x = linear_22_cast_fp16)[name = string("view_20_cast_fp16")];
            tensor<int32, [4]> transpose_16_perm_0 = const()[name = string("transpose_16_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49296640))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50083136))))[name = string("p_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_23_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_11_cast_fp16)[name = string("linear_23_cast_fp16")];
            tensor<int32, [4]> const_239 = const()[name = string("const_239"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_21_cast_fp16 = reshape(shape = const_239, x = linear_23_cast_fp16)[name = string("view_21_cast_fp16")];
            tensor<int32, [4]> transpose_17_perm_0 = const()[name = string("transpose_17_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_22_to_fp16 = const()[name = string("view_22_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50091392)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_15_cast_fp16 = transpose(perm = transpose_15_perm_0, x = view_19_cast_fp16)[name = string("transpose_131")];
            tensor<fp16, [1, 8, 188, 128]> add_28_cast_fp16 = add(x = transpose_15_cast_fp16, y = view_22_to_fp16)[name = string("add_28_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_23_to_fp16 = const()[name = string("view_23_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50093504)))];
            tensor<fp16, [1, 8, 188, 128]> add_29_cast_fp16 = add(x = transpose_15_cast_fp16, y = view_23_to_fp16)[name = string("add_29_cast_fp16")];
            bool matmul_3_transpose_x_0 = const()[name = string("matmul_3_transpose_x_0"), val = bool(false)];
            bool matmul_3_transpose_y_0 = const()[name = string("matmul_3_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_2_to_fp16 = const()[name = string("permute_2_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50095616)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_3_cast_fp16 = matmul(transpose_x = matmul_3_transpose_x_0, transpose_y = matmul_3_transpose_y_0, x = add_29_cast_fp16, y = permute_2_to_fp16)[name = string("matmul_3_cast_fp16")];
            tensor<int32, [8]> pad_2_pad_0 = const()[name = string("pad_2_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_2_mode_0 = const()[name = string("pad_2_mode_0"), val = string("constant")];
            fp16 const_247_to_fp16 = const()[name = string("const_247_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_2_cast_fp16 = pad(constant_val = const_247_to_fp16, mode = pad_2_mode_0, pad = pad_2_pad_0, x = matmul_3_cast_fp16)[name = string("pad_2_cast_fp16")];
            tensor<int32, [4]> const_248 = const()[name = string("const_248"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_25_cast_fp16 = reshape(shape = const_248, x = pad_2_cast_fp16)[name = string("view_25_cast_fp16")];
            tensor<int32, [4]> slice_5_begin_0 = const()[name = string("slice_5_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_5_end_0 = const()[name = string("slice_5_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_5_end_mask_0 = const()[name = string("slice_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_5_cast_fp16 = slice_by_index(begin = slice_5_begin_0, end = slice_5_end_0, end_mask = slice_5_end_mask_0, x = view_25_cast_fp16)[name = string("slice_5_cast_fp16")];
            tensor<int32, [4]> const_252 = const()[name = string("const_252"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_26_cast_fp16 = reshape(shape = const_252, x = slice_5_cast_fp16)[name = string("view_26_cast_fp16")];
            tensor<int32, [4]> slice_6_begin_0 = const()[name = string("slice_6_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_6_end_0 = const()[name = string("slice_6_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_6_end_mask_0 = const()[name = string("slice_6_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_6_cast_fp16 = slice_by_index(begin = slice_6_begin_0, end = slice_6_end_0, end_mask = slice_6_end_mask_0, x = view_26_cast_fp16)[name = string("slice_6_cast_fp16")];
            fp16 const_256_to_fp16 = const()[name = string("const_256_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_13_cast_fp16 = mul(x = slice_6_cast_fp16, y = const_256_to_fp16)[name = string("mul_13_cast_fp16")];
            fp16 const_257_to_fp16 = const()[name = string("const_257_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_4_cast_fp16 = select(a = const_257_to_fp16, b = mul_13_cast_fp16, cond = logical_not)[name = string("masked_fill_4_cast_fp16")];
            fp16 const_258_to_fp16 = const()[name = string("const_258_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_2_1_cast_fp16 = mul(x = add_28_cast_fp16, y = const_258_to_fp16)[name = string("mul_2_1_cast_fp16")];
            bool matmul_2_transpose_y_1 = const()[name = string("matmul_2_transpose_y_1"), val = bool(true)];
            bool matmul_2_transpose_x_1 = const()[name = string("matmul_2_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_16_cast_fp16 = transpose(perm = transpose_16_perm_0, x = view_20_cast_fp16)[name = string("transpose_130")];
            tensor<fp16, [1, 8, 188, 188]> matmul_2_1_cast_fp16 = matmul(transpose_x = matmul_2_transpose_x_1, transpose_y = matmul_2_transpose_y_1, x = mul_2_1_cast_fp16, y = transpose_16_cast_fp16)[name = string("matmul_2_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_2_1_cast_fp16 = add(x = matmul_2_1_cast_fp16, y = masked_fill_4_cast_fp16)[name = string("add_2_1_cast_fp16")];
            int32 softmax_2_axis_0 = const()[name = string("softmax_2_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_2_cast_fp16 = softmax(axis = softmax_2_axis_0, x = add_2_1_cast_fp16)[name = string("softmax_2_cast_fp16")];
            bool scaled_dot_product_attention_2_transpose_x_0 = const()[name = string("scaled_dot_product_attention_2_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_2_transpose_y_0 = const()[name = string("scaled_dot_product_attention_2_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_17_cast_fp16 = transpose(perm = transpose_17_perm_0, x = view_21_cast_fp16)[name = string("transpose_129")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_2_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_2_transpose_x_0, transpose_y = scaled_dot_product_attention_2_transpose_y_0, x = softmax_2_cast_fp16, y = transpose_17_cast_fp16)[name = string("scaled_dot_product_attention_2_cast_fp16")];
            tensor<int32, [4]> transpose_18_perm_0 = const()[name = string("transpose_18_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_261 = const()[name = string("const_261"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_18_cast_fp16 = transpose(perm = transpose_18_perm_0, x = scaled_dot_product_attention_2_cast_fp16)[name = string("transpose_128")];
            tensor<fp16, [1, 188, 1024]> view_27_cast_fp16 = reshape(shape = const_261, x = transpose_18_cast_fp16)[name = string("view_27_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_2_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50863680))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51650176))))[name = string("p_encoder_layers_2_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_25_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_2_self_attn_o_proj_weight_to_fp16_palettized, x = view_27_cast_fp16)[name = string("linear_25_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_30_cast_fp16 = add(x = add_27_cast_fp16, y = linear_25_cast_fp16)[name = string("add_30_cast_fp16")];
            tensor<int32, [1]> layer_norm_12_axes_0 = const()[name = string("layer_norm_12_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_2_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_2_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51658432)))];
            tensor<fp16, [1024]> p_encoder_layers_2_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_2_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51660544)))];
            fp16 const_263_to_fp16 = const()[name = string("const_263_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_12_cast_fp16 = layer_norm(axes = layer_norm_12_axes_0, beta = p_encoder_layers_2_norm_conv_bias_to_fp16, epsilon = const_263_to_fp16, gamma = p_encoder_layers_2_norm_conv_weight_to_fp16, x = add_30_cast_fp16)[name = string("layer_norm_12_cast_fp16")];
            tensor<int32, [3]> transpose_19_perm_0 = const()[name = string("transpose_19_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_6_pad_type_0 = const()[name = string("conv1d_6_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_6_strides_0 = const()[name = string("conv1d_6_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_6_pad_0 = const()[name = string("conv1d_6_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_6_dilations_0 = const()[name = string("conv1d_6_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_6_groups_0 = const()[name = string("conv1d_6_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_2_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51662656))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53235584))))[name = string("p_encoder_layers_2_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_19_cast_fp16 = transpose(perm = transpose_19_perm_0, x = layer_norm_12_cast_fp16)[name = string("transpose_127")];
            tensor<fp16, [1, 2048, 188]> conv1d_6_cast_fp16 = conv(dilations = conv1d_6_dilations_0, groups = conv1d_6_groups_0, pad = conv1d_6_pad_0, pad_type = conv1d_6_pad_type_0, strides = conv1d_6_strides_0, weight = p_encoder_layers_2_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_19_cast_fp16)[name = string("conv1d_6_cast_fp16")];
            int32 glu_2_split_num_splits_0 = const()[name = string("glu_2_split_num_splits_0"), val = int32(2)];
            int32 glu_2_split_axis_0 = const()[name = string("glu_2_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_2_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_2_split_cast_fp16_1 = split(axis = glu_2_split_axis_0, num_splits = glu_2_split_num_splits_0, x = conv1d_6_cast_fp16)[name = string("glu_2_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_2_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_2_split_cast_fp16_1)[name = string("glu_2_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_2_cast_fp16 = mul(x = glu_2_split_cast_fp16_0, y = glu_2_split_1_sigmoid_cast_fp16)[name = string("glu_2_cast_fp16")];
            fp16 const_269_to_fp16 = const()[name = string("const_269_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_5_cast_fp16 = select(a = const_269_to_fp16, b = glu_2_cast_fp16, cond = all_1)[name = string("masked_fill_5_cast_fp16")];
            string conv1d_7_pad_type_0 = const()[name = string("conv1d_7_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_7_pad_0 = const()[name = string("conv1d_7_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_7_groups_0 = const()[name = string("conv1d_7_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_7_strides_0 = const()[name = string("conv1d_7_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_7_dilations_0 = const()[name = string("conv1d_7_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1551_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53252032))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53259008))))[name = string("const_1551_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1552_to_fp16 = const()[name = string("const_1552_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53267264)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_2_cast_fp16 = conv(bias = const_1552_to_fp16, dilations = conv1d_7_dilations_0, groups = conv1d_7_groups_0, pad = conv1d_7_pad_0, pad_type = conv1d_7_pad_type_0, strides = conv1d_7_strides_0, weight = const_1551_to_fp16_palettized, x = masked_fill_5_cast_fp16)[name = string("_native_batch_norm_legit_no_training_2_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_7_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_2_cast_fp16)[name = string("silu_7_cast_fp16")];
            string conv1d_8_pad_type_0 = const()[name = string("conv1d_8_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_8_strides_0 = const()[name = string("conv1d_8_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_8_pad_0 = const()[name = string("conv1d_8_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_8_dilations_0 = const()[name = string("conv1d_8_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_8_groups_0 = const()[name = string("conv1d_8_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_2_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53269376))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54055872))))[name = string("p_encoder_layers_2_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_8_cast_fp16 = conv(dilations = conv1d_8_dilations_0, groups = conv1d_8_groups_0, pad = conv1d_8_pad_0, pad_type = conv1d_8_pad_type_0, strides = conv1d_8_strides_0, weight = p_encoder_layers_2_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_7_cast_fp16)[name = string("conv1d_8_cast_fp16")];
            tensor<int32, [3]> transpose_20_perm_0 = const()[name = string("transpose_20_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_20_cast_fp16 = transpose(perm = transpose_20_perm_0, x = conv1d_8_cast_fp16)[name = string("transpose_126")];
            tensor<fp16, [1, 188, 1024]> add_31_cast_fp16 = add(x = add_30_cast_fp16, y = transpose_20_cast_fp16)[name = string("add_31_cast_fp16")];
            tensor<int32, [1]> layer_norm_13_axes_0 = const()[name = string("layer_norm_13_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_2_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_2_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54064128)))];
            tensor<fp16, [1024]> p_encoder_layers_2_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_2_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54066240)))];
            fp16 const_280_to_fp16 = const()[name = string("const_280_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_13_cast_fp16 = layer_norm(axes = layer_norm_13_axes_0, beta = p_encoder_layers_2_norm_feed_forward2_bias_to_fp16, epsilon = const_280_to_fp16, gamma = p_encoder_layers_2_norm_feed_forward2_weight_to_fp16, x = add_31_cast_fp16)[name = string("layer_norm_13_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_2_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54068352))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57214144))))[name = string("p_encoder_layers_2_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_26_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_2_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_13_cast_fp16)[name = string("linear_26_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_8_cast_fp16 = silu(x = linear_26_cast_fp16)[name = string("silu_8_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_2_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57246976))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60392768))))[name = string("p_encoder_layers_2_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_27_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_2_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_8_cast_fp16)[name = string("linear_27_cast_fp16")];
            fp16 const_282_to_fp16 = const()[name = string("const_282_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_14_cast_fp16 = mul(x = linear_27_cast_fp16, y = const_282_to_fp16)[name = string("mul_14_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_32_cast_fp16 = add(x = add_31_cast_fp16, y = mul_14_cast_fp16)[name = string("add_32_cast_fp16")];
            tensor<int32, [1]> layer_norm_14_axes_0 = const()[name = string("layer_norm_14_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_2_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_2_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60401024)))];
            tensor<fp16, [1024]> p_encoder_layers_2_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_2_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60403136)))];
            fp16 const_284_to_fp16 = const()[name = string("const_284_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_14_cast_fp16 = layer_norm(axes = layer_norm_14_axes_0, beta = p_encoder_layers_2_norm_out_bias_to_fp16, epsilon = const_284_to_fp16, gamma = p_encoder_layers_2_norm_out_weight_to_fp16, x = add_32_cast_fp16)[name = string("layer_norm_14_cast_fp16")];
            tensor<int32, [1]> layer_norm_15_axes_0 = const()[name = string("layer_norm_15_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_3_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_3_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60405248)))];
            tensor<fp16, [1024]> p_encoder_layers_3_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_3_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60407360)))];
            fp16 const_287_to_fp16 = const()[name = string("const_287_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_15_cast_fp16 = layer_norm(axes = layer_norm_15_axes_0, beta = p_encoder_layers_3_norm_feed_forward1_bias_to_fp16, epsilon = const_287_to_fp16, gamma = p_encoder_layers_3_norm_feed_forward1_weight_to_fp16, x = layer_norm_14_cast_fp16)[name = string("layer_norm_15_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_3_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60409472))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63555264))))[name = string("p_encoder_layers_3_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_28_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_3_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_15_cast_fp16)[name = string("linear_28_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_9_cast_fp16 = silu(x = linear_28_cast_fp16)[name = string("silu_9_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_3_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63588096))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66733888))))[name = string("p_encoder_layers_3_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_29_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_3_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_9_cast_fp16)[name = string("linear_29_cast_fp16")];
            fp16 const_289_to_fp16 = const()[name = string("const_289_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_15_cast_fp16 = mul(x = linear_29_cast_fp16, y = const_289_to_fp16)[name = string("mul_15_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_33_cast_fp16 = add(x = layer_norm_14_cast_fp16, y = mul_15_cast_fp16)[name = string("add_33_cast_fp16")];
            tensor<int32, [1]> layer_norm_16_axes_0 = const()[name = string("layer_norm_16_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_3_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_3_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66742144)))];
            tensor<fp16, [1024]> p_encoder_layers_3_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_3_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66744256)))];
            fp16 const_291_to_fp16 = const()[name = string("const_291_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_16_cast_fp16 = layer_norm(axes = layer_norm_16_axes_0, beta = p_encoder_layers_3_norm_self_att_bias_to_fp16, epsilon = const_291_to_fp16, gamma = p_encoder_layers_3_norm_self_att_weight_to_fp16, x = add_33_cast_fp16)[name = string("layer_norm_16_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66746368))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67532864))))[name = string("p_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_30_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_16_cast_fp16)[name = string("linear_30_cast_fp16")];
            tensor<int32, [4]> const_293 = const()[name = string("const_293"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_28_cast_fp16 = reshape(shape = const_293, x = linear_30_cast_fp16)[name = string("view_28_cast_fp16")];
            tensor<int32, [4]> transpose_21_perm_0 = const()[name = string("transpose_21_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67541120))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68327616))))[name = string("p_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_31_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_16_cast_fp16)[name = string("linear_31_cast_fp16")];
            tensor<int32, [4]> const_296 = const()[name = string("const_296"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_29_cast_fp16 = reshape(shape = const_296, x = linear_31_cast_fp16)[name = string("view_29_cast_fp16")];
            tensor<int32, [4]> transpose_22_perm_0 = const()[name = string("transpose_22_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68335872))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69122368))))[name = string("p_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_32_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_16_cast_fp16)[name = string("linear_32_cast_fp16")];
            tensor<int32, [4]> const_299 = const()[name = string("const_299"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_30_cast_fp16 = reshape(shape = const_299, x = linear_32_cast_fp16)[name = string("view_30_cast_fp16")];
            tensor<int32, [4]> transpose_23_perm_0 = const()[name = string("transpose_23_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_31_to_fp16 = const()[name = string("view_31_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69130624)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_21_cast_fp16 = transpose(perm = transpose_21_perm_0, x = view_28_cast_fp16)[name = string("transpose_125")];
            tensor<fp16, [1, 8, 188, 128]> add_34_cast_fp16 = add(x = transpose_21_cast_fp16, y = view_31_to_fp16)[name = string("add_34_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_32_to_fp16 = const()[name = string("view_32_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69132736)))];
            tensor<fp16, [1, 8, 188, 128]> add_35_cast_fp16 = add(x = transpose_21_cast_fp16, y = view_32_to_fp16)[name = string("add_35_cast_fp16")];
            bool matmul_4_transpose_x_0 = const()[name = string("matmul_4_transpose_x_0"), val = bool(false)];
            bool matmul_4_transpose_y_0 = const()[name = string("matmul_4_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_3_to_fp16 = const()[name = string("permute_3_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69134848)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_4_cast_fp16 = matmul(transpose_x = matmul_4_transpose_x_0, transpose_y = matmul_4_transpose_y_0, x = add_35_cast_fp16, y = permute_3_to_fp16)[name = string("matmul_4_cast_fp16")];
            tensor<int32, [8]> pad_3_pad_0 = const()[name = string("pad_3_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_3_mode_0 = const()[name = string("pad_3_mode_0"), val = string("constant")];
            fp16 const_307_to_fp16 = const()[name = string("const_307_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_3_cast_fp16 = pad(constant_val = const_307_to_fp16, mode = pad_3_mode_0, pad = pad_3_pad_0, x = matmul_4_cast_fp16)[name = string("pad_3_cast_fp16")];
            tensor<int32, [4]> const_308 = const()[name = string("const_308"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_34_cast_fp16 = reshape(shape = const_308, x = pad_3_cast_fp16)[name = string("view_34_cast_fp16")];
            tensor<int32, [4]> slice_7_begin_0 = const()[name = string("slice_7_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_7_end_0 = const()[name = string("slice_7_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_7_end_mask_0 = const()[name = string("slice_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_7_cast_fp16 = slice_by_index(begin = slice_7_begin_0, end = slice_7_end_0, end_mask = slice_7_end_mask_0, x = view_34_cast_fp16)[name = string("slice_7_cast_fp16")];
            tensor<int32, [4]> const_312 = const()[name = string("const_312"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_35_cast_fp16 = reshape(shape = const_312, x = slice_7_cast_fp16)[name = string("view_35_cast_fp16")];
            tensor<int32, [4]> slice_8_begin_0 = const()[name = string("slice_8_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_8_end_0 = const()[name = string("slice_8_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_8_end_mask_0 = const()[name = string("slice_8_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_8_cast_fp16 = slice_by_index(begin = slice_8_begin_0, end = slice_8_end_0, end_mask = slice_8_end_mask_0, x = view_35_cast_fp16)[name = string("slice_8_cast_fp16")];
            fp16 const_316_to_fp16 = const()[name = string("const_316_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_16_cast_fp16 = mul(x = slice_8_cast_fp16, y = const_316_to_fp16)[name = string("mul_16_cast_fp16")];
            fp16 const_317_to_fp16 = const()[name = string("const_317_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_6_cast_fp16 = select(a = const_317_to_fp16, b = mul_16_cast_fp16, cond = logical_not)[name = string("masked_fill_6_cast_fp16")];
            fp16 const_318_to_fp16 = const()[name = string("const_318_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_3_1_cast_fp16 = mul(x = add_34_cast_fp16, y = const_318_to_fp16)[name = string("mul_3_1_cast_fp16")];
            bool matmul_3_transpose_y_1 = const()[name = string("matmul_3_transpose_y_1"), val = bool(true)];
            bool matmul_3_transpose_x_1 = const()[name = string("matmul_3_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_22_cast_fp16 = transpose(perm = transpose_22_perm_0, x = view_29_cast_fp16)[name = string("transpose_124")];
            tensor<fp16, [1, 8, 188, 188]> matmul_3_1_cast_fp16 = matmul(transpose_x = matmul_3_transpose_x_1, transpose_y = matmul_3_transpose_y_1, x = mul_3_1_cast_fp16, y = transpose_22_cast_fp16)[name = string("matmul_3_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_3_1_cast_fp16 = add(x = matmul_3_1_cast_fp16, y = masked_fill_6_cast_fp16)[name = string("add_3_1_cast_fp16")];
            int32 softmax_3_axis_0 = const()[name = string("softmax_3_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_3_cast_fp16 = softmax(axis = softmax_3_axis_0, x = add_3_1_cast_fp16)[name = string("softmax_3_cast_fp16")];
            bool scaled_dot_product_attention_3_transpose_x_0 = const()[name = string("scaled_dot_product_attention_3_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_3_transpose_y_0 = const()[name = string("scaled_dot_product_attention_3_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_23_cast_fp16 = transpose(perm = transpose_23_perm_0, x = view_30_cast_fp16)[name = string("transpose_123")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_3_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_3_transpose_x_0, transpose_y = scaled_dot_product_attention_3_transpose_y_0, x = softmax_3_cast_fp16, y = transpose_23_cast_fp16)[name = string("scaled_dot_product_attention_3_cast_fp16")];
            tensor<int32, [4]> transpose_24_perm_0 = const()[name = string("transpose_24_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_321 = const()[name = string("const_321"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_24_cast_fp16 = transpose(perm = transpose_24_perm_0, x = scaled_dot_product_attention_3_cast_fp16)[name = string("transpose_122")];
            tensor<fp16, [1, 188, 1024]> view_36_cast_fp16 = reshape(shape = const_321, x = transpose_24_cast_fp16)[name = string("view_36_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_3_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69902912))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70689408))))[name = string("p_encoder_layers_3_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_34_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_3_self_attn_o_proj_weight_to_fp16_palettized, x = view_36_cast_fp16)[name = string("linear_34_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_36_cast_fp16 = add(x = add_33_cast_fp16, y = linear_34_cast_fp16)[name = string("add_36_cast_fp16")];
            tensor<int32, [1]> layer_norm_17_axes_0 = const()[name = string("layer_norm_17_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_3_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_3_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70697664)))];
            tensor<fp16, [1024]> p_encoder_layers_3_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_3_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70699776)))];
            fp16 const_323_to_fp16 = const()[name = string("const_323_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_17_cast_fp16 = layer_norm(axes = layer_norm_17_axes_0, beta = p_encoder_layers_3_norm_conv_bias_to_fp16, epsilon = const_323_to_fp16, gamma = p_encoder_layers_3_norm_conv_weight_to_fp16, x = add_36_cast_fp16)[name = string("layer_norm_17_cast_fp16")];
            tensor<int32, [3]> transpose_25_perm_0 = const()[name = string("transpose_25_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_9_pad_type_0 = const()[name = string("conv1d_9_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_9_strides_0 = const()[name = string("conv1d_9_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_9_pad_0 = const()[name = string("conv1d_9_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_9_dilations_0 = const()[name = string("conv1d_9_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_9_groups_0 = const()[name = string("conv1d_9_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_3_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70701888))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72274816))))[name = string("p_encoder_layers_3_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_25_cast_fp16 = transpose(perm = transpose_25_perm_0, x = layer_norm_17_cast_fp16)[name = string("transpose_121")];
            tensor<fp16, [1, 2048, 188]> conv1d_9_cast_fp16 = conv(dilations = conv1d_9_dilations_0, groups = conv1d_9_groups_0, pad = conv1d_9_pad_0, pad_type = conv1d_9_pad_type_0, strides = conv1d_9_strides_0, weight = p_encoder_layers_3_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_25_cast_fp16)[name = string("conv1d_9_cast_fp16")];
            int32 glu_3_split_num_splits_0 = const()[name = string("glu_3_split_num_splits_0"), val = int32(2)];
            int32 glu_3_split_axis_0 = const()[name = string("glu_3_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_3_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_3_split_cast_fp16_1 = split(axis = glu_3_split_axis_0, num_splits = glu_3_split_num_splits_0, x = conv1d_9_cast_fp16)[name = string("glu_3_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_3_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_3_split_cast_fp16_1)[name = string("glu_3_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_3_cast_fp16 = mul(x = glu_3_split_cast_fp16_0, y = glu_3_split_1_sigmoid_cast_fp16)[name = string("glu_3_cast_fp16")];
            fp16 const_329_to_fp16 = const()[name = string("const_329_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_7_cast_fp16 = select(a = const_329_to_fp16, b = glu_3_cast_fp16, cond = all_1)[name = string("masked_fill_7_cast_fp16")];
            string conv1d_10_pad_type_0 = const()[name = string("conv1d_10_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_10_pad_0 = const()[name = string("conv1d_10_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_10_groups_0 = const()[name = string("conv1d_10_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_10_strides_0 = const()[name = string("conv1d_10_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_10_dilations_0 = const()[name = string("conv1d_10_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1553_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72291264))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72298240))))[name = string("const_1553_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1554_to_fp16 = const()[name = string("const_1554_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72306496)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_3_cast_fp16 = conv(bias = const_1554_to_fp16, dilations = conv1d_10_dilations_0, groups = conv1d_10_groups_0, pad = conv1d_10_pad_0, pad_type = conv1d_10_pad_type_0, strides = conv1d_10_strides_0, weight = const_1553_to_fp16_palettized, x = masked_fill_7_cast_fp16)[name = string("_native_batch_norm_legit_no_training_3_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_10_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_3_cast_fp16)[name = string("silu_10_cast_fp16")];
            string conv1d_11_pad_type_0 = const()[name = string("conv1d_11_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_11_strides_0 = const()[name = string("conv1d_11_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_11_pad_0 = const()[name = string("conv1d_11_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_11_dilations_0 = const()[name = string("conv1d_11_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_11_groups_0 = const()[name = string("conv1d_11_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_3_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72308608))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73095104))))[name = string("p_encoder_layers_3_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_11_cast_fp16 = conv(dilations = conv1d_11_dilations_0, groups = conv1d_11_groups_0, pad = conv1d_11_pad_0, pad_type = conv1d_11_pad_type_0, strides = conv1d_11_strides_0, weight = p_encoder_layers_3_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_10_cast_fp16)[name = string("conv1d_11_cast_fp16")];
            tensor<int32, [3]> transpose_26_perm_0 = const()[name = string("transpose_26_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_26_cast_fp16 = transpose(perm = transpose_26_perm_0, x = conv1d_11_cast_fp16)[name = string("transpose_120")];
            tensor<fp16, [1, 188, 1024]> add_37_cast_fp16 = add(x = add_36_cast_fp16, y = transpose_26_cast_fp16)[name = string("add_37_cast_fp16")];
            tensor<int32, [1]> layer_norm_18_axes_0 = const()[name = string("layer_norm_18_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_3_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_3_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73103360)))];
            tensor<fp16, [1024]> p_encoder_layers_3_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_3_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73105472)))];
            fp16 const_340_to_fp16 = const()[name = string("const_340_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_18_cast_fp16 = layer_norm(axes = layer_norm_18_axes_0, beta = p_encoder_layers_3_norm_feed_forward2_bias_to_fp16, epsilon = const_340_to_fp16, gamma = p_encoder_layers_3_norm_feed_forward2_weight_to_fp16, x = add_37_cast_fp16)[name = string("layer_norm_18_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_3_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73107584))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76253376))))[name = string("p_encoder_layers_3_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_35_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_3_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_18_cast_fp16)[name = string("linear_35_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_11_cast_fp16 = silu(x = linear_35_cast_fp16)[name = string("silu_11_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_3_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76286208))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79432000))))[name = string("p_encoder_layers_3_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_36_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_3_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_11_cast_fp16)[name = string("linear_36_cast_fp16")];
            fp16 const_342_to_fp16 = const()[name = string("const_342_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_17_cast_fp16 = mul(x = linear_36_cast_fp16, y = const_342_to_fp16)[name = string("mul_17_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_38_cast_fp16 = add(x = add_37_cast_fp16, y = mul_17_cast_fp16)[name = string("add_38_cast_fp16")];
            tensor<int32, [1]> layer_norm_19_axes_0 = const()[name = string("layer_norm_19_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_3_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_3_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79440256)))];
            tensor<fp16, [1024]> p_encoder_layers_3_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_3_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79442368)))];
            fp16 const_344_to_fp16 = const()[name = string("const_344_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_19_cast_fp16 = layer_norm(axes = layer_norm_19_axes_0, beta = p_encoder_layers_3_norm_out_bias_to_fp16, epsilon = const_344_to_fp16, gamma = p_encoder_layers_3_norm_out_weight_to_fp16, x = add_38_cast_fp16)[name = string("layer_norm_19_cast_fp16")];
            tensor<int32, [1]> layer_norm_20_axes_0 = const()[name = string("layer_norm_20_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_4_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_4_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79444480)))];
            tensor<fp16, [1024]> p_encoder_layers_4_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_4_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79446592)))];
            fp16 const_347_to_fp16 = const()[name = string("const_347_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_20_cast_fp16 = layer_norm(axes = layer_norm_20_axes_0, beta = p_encoder_layers_4_norm_feed_forward1_bias_to_fp16, epsilon = const_347_to_fp16, gamma = p_encoder_layers_4_norm_feed_forward1_weight_to_fp16, x = layer_norm_19_cast_fp16)[name = string("layer_norm_20_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_4_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79448704))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82594496))))[name = string("p_encoder_layers_4_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_37_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_4_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_20_cast_fp16)[name = string("linear_37_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_12_cast_fp16 = silu(x = linear_37_cast_fp16)[name = string("silu_12_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_4_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82627328))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85773120))))[name = string("p_encoder_layers_4_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_38_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_4_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_12_cast_fp16)[name = string("linear_38_cast_fp16")];
            fp16 const_349_to_fp16 = const()[name = string("const_349_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_18_cast_fp16 = mul(x = linear_38_cast_fp16, y = const_349_to_fp16)[name = string("mul_18_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_39_cast_fp16 = add(x = layer_norm_19_cast_fp16, y = mul_18_cast_fp16)[name = string("add_39_cast_fp16")];
            tensor<int32, [1]> layer_norm_21_axes_0 = const()[name = string("layer_norm_21_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_4_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_4_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85781376)))];
            tensor<fp16, [1024]> p_encoder_layers_4_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_4_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85783488)))];
            fp16 const_351_to_fp16 = const()[name = string("const_351_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_21_cast_fp16 = layer_norm(axes = layer_norm_21_axes_0, beta = p_encoder_layers_4_norm_self_att_bias_to_fp16, epsilon = const_351_to_fp16, gamma = p_encoder_layers_4_norm_self_att_weight_to_fp16, x = add_39_cast_fp16)[name = string("layer_norm_21_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85785600))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86572096))))[name = string("p_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_39_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_21_cast_fp16)[name = string("linear_39_cast_fp16")];
            tensor<int32, [4]> const_353 = const()[name = string("const_353"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_37_cast_fp16 = reshape(shape = const_353, x = linear_39_cast_fp16)[name = string("view_37_cast_fp16")];
            tensor<int32, [4]> transpose_27_perm_0 = const()[name = string("transpose_27_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86580352))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87366848))))[name = string("p_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_40_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_21_cast_fp16)[name = string("linear_40_cast_fp16")];
            tensor<int32, [4]> const_356 = const()[name = string("const_356"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_38_cast_fp16 = reshape(shape = const_356, x = linear_40_cast_fp16)[name = string("view_38_cast_fp16")];
            tensor<int32, [4]> transpose_28_perm_0 = const()[name = string("transpose_28_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87375104))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88161600))))[name = string("p_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_41_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_21_cast_fp16)[name = string("linear_41_cast_fp16")];
            tensor<int32, [4]> const_359 = const()[name = string("const_359"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_39_cast_fp16 = reshape(shape = const_359, x = linear_41_cast_fp16)[name = string("view_39_cast_fp16")];
            tensor<int32, [4]> transpose_29_perm_0 = const()[name = string("transpose_29_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_40_to_fp16 = const()[name = string("view_40_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88169856)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_27_cast_fp16 = transpose(perm = transpose_27_perm_0, x = view_37_cast_fp16)[name = string("transpose_119")];
            tensor<fp16, [1, 8, 188, 128]> add_40_cast_fp16 = add(x = transpose_27_cast_fp16, y = view_40_to_fp16)[name = string("add_40_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_41_to_fp16 = const()[name = string("view_41_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88171968)))];
            tensor<fp16, [1, 8, 188, 128]> add_41_cast_fp16 = add(x = transpose_27_cast_fp16, y = view_41_to_fp16)[name = string("add_41_cast_fp16")];
            bool matmul_5_transpose_x_0 = const()[name = string("matmul_5_transpose_x_0"), val = bool(false)];
            bool matmul_5_transpose_y_0 = const()[name = string("matmul_5_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_4_to_fp16 = const()[name = string("permute_4_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88174080)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_5_cast_fp16 = matmul(transpose_x = matmul_5_transpose_x_0, transpose_y = matmul_5_transpose_y_0, x = add_41_cast_fp16, y = permute_4_to_fp16)[name = string("matmul_5_cast_fp16")];
            tensor<int32, [8]> pad_4_pad_0 = const()[name = string("pad_4_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_4_mode_0 = const()[name = string("pad_4_mode_0"), val = string("constant")];
            fp16 const_367_to_fp16 = const()[name = string("const_367_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_4_cast_fp16 = pad(constant_val = const_367_to_fp16, mode = pad_4_mode_0, pad = pad_4_pad_0, x = matmul_5_cast_fp16)[name = string("pad_4_cast_fp16")];
            tensor<int32, [4]> const_368 = const()[name = string("const_368"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_43_cast_fp16 = reshape(shape = const_368, x = pad_4_cast_fp16)[name = string("view_43_cast_fp16")];
            tensor<int32, [4]> slice_9_begin_0 = const()[name = string("slice_9_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_9_end_0 = const()[name = string("slice_9_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_9_end_mask_0 = const()[name = string("slice_9_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_9_cast_fp16 = slice_by_index(begin = slice_9_begin_0, end = slice_9_end_0, end_mask = slice_9_end_mask_0, x = view_43_cast_fp16)[name = string("slice_9_cast_fp16")];
            tensor<int32, [4]> const_372 = const()[name = string("const_372"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_44_cast_fp16 = reshape(shape = const_372, x = slice_9_cast_fp16)[name = string("view_44_cast_fp16")];
            tensor<int32, [4]> slice_10_begin_0 = const()[name = string("slice_10_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_10_end_0 = const()[name = string("slice_10_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_10_end_mask_0 = const()[name = string("slice_10_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_10_cast_fp16 = slice_by_index(begin = slice_10_begin_0, end = slice_10_end_0, end_mask = slice_10_end_mask_0, x = view_44_cast_fp16)[name = string("slice_10_cast_fp16")];
            fp16 const_376_to_fp16 = const()[name = string("const_376_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_19_cast_fp16 = mul(x = slice_10_cast_fp16, y = const_376_to_fp16)[name = string("mul_19_cast_fp16")];
            fp16 const_377_to_fp16 = const()[name = string("const_377_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_8_cast_fp16 = select(a = const_377_to_fp16, b = mul_19_cast_fp16, cond = logical_not)[name = string("masked_fill_8_cast_fp16")];
            fp16 const_378_to_fp16 = const()[name = string("const_378_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_4_1_cast_fp16 = mul(x = add_40_cast_fp16, y = const_378_to_fp16)[name = string("mul_4_1_cast_fp16")];
            bool matmul_4_transpose_y_1 = const()[name = string("matmul_4_transpose_y_1"), val = bool(true)];
            bool matmul_4_transpose_x_1 = const()[name = string("matmul_4_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_28_cast_fp16 = transpose(perm = transpose_28_perm_0, x = view_38_cast_fp16)[name = string("transpose_118")];
            tensor<fp16, [1, 8, 188, 188]> matmul_4_1_cast_fp16 = matmul(transpose_x = matmul_4_transpose_x_1, transpose_y = matmul_4_transpose_y_1, x = mul_4_1_cast_fp16, y = transpose_28_cast_fp16)[name = string("matmul_4_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_4_1_cast_fp16 = add(x = matmul_4_1_cast_fp16, y = masked_fill_8_cast_fp16)[name = string("add_4_1_cast_fp16")];
            int32 softmax_4_axis_0 = const()[name = string("softmax_4_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_4_cast_fp16 = softmax(axis = softmax_4_axis_0, x = add_4_1_cast_fp16)[name = string("softmax_4_cast_fp16")];
            bool scaled_dot_product_attention_4_transpose_x_0 = const()[name = string("scaled_dot_product_attention_4_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_4_transpose_y_0 = const()[name = string("scaled_dot_product_attention_4_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_29_cast_fp16 = transpose(perm = transpose_29_perm_0, x = view_39_cast_fp16)[name = string("transpose_117")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_4_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_4_transpose_x_0, transpose_y = scaled_dot_product_attention_4_transpose_y_0, x = softmax_4_cast_fp16, y = transpose_29_cast_fp16)[name = string("scaled_dot_product_attention_4_cast_fp16")];
            tensor<int32, [4]> transpose_30_perm_0 = const()[name = string("transpose_30_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_381 = const()[name = string("const_381"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_30_cast_fp16 = transpose(perm = transpose_30_perm_0, x = scaled_dot_product_attention_4_cast_fp16)[name = string("transpose_116")];
            tensor<fp16, [1, 188, 1024]> view_45_cast_fp16 = reshape(shape = const_381, x = transpose_30_cast_fp16)[name = string("view_45_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_4_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88942144))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89728640))))[name = string("p_encoder_layers_4_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_43_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_4_self_attn_o_proj_weight_to_fp16_palettized, x = view_45_cast_fp16)[name = string("linear_43_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_42_cast_fp16 = add(x = add_39_cast_fp16, y = linear_43_cast_fp16)[name = string("add_42_cast_fp16")];
            tensor<int32, [1]> layer_norm_22_axes_0 = const()[name = string("layer_norm_22_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_4_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_4_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89736896)))];
            tensor<fp16, [1024]> p_encoder_layers_4_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_4_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89739008)))];
            fp16 const_383_to_fp16 = const()[name = string("const_383_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_22_cast_fp16 = layer_norm(axes = layer_norm_22_axes_0, beta = p_encoder_layers_4_norm_conv_bias_to_fp16, epsilon = const_383_to_fp16, gamma = p_encoder_layers_4_norm_conv_weight_to_fp16, x = add_42_cast_fp16)[name = string("layer_norm_22_cast_fp16")];
            tensor<int32, [3]> transpose_31_perm_0 = const()[name = string("transpose_31_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_12_pad_type_0 = const()[name = string("conv1d_12_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_12_strides_0 = const()[name = string("conv1d_12_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_12_pad_0 = const()[name = string("conv1d_12_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_12_dilations_0 = const()[name = string("conv1d_12_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_12_groups_0 = const()[name = string("conv1d_12_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_4_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89741120))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91314048))))[name = string("p_encoder_layers_4_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_31_cast_fp16 = transpose(perm = transpose_31_perm_0, x = layer_norm_22_cast_fp16)[name = string("transpose_115")];
            tensor<fp16, [1, 2048, 188]> conv1d_12_cast_fp16 = conv(dilations = conv1d_12_dilations_0, groups = conv1d_12_groups_0, pad = conv1d_12_pad_0, pad_type = conv1d_12_pad_type_0, strides = conv1d_12_strides_0, weight = p_encoder_layers_4_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_31_cast_fp16)[name = string("conv1d_12_cast_fp16")];
            int32 glu_4_split_num_splits_0 = const()[name = string("glu_4_split_num_splits_0"), val = int32(2)];
            int32 glu_4_split_axis_0 = const()[name = string("glu_4_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_4_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_4_split_cast_fp16_1 = split(axis = glu_4_split_axis_0, num_splits = glu_4_split_num_splits_0, x = conv1d_12_cast_fp16)[name = string("glu_4_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_4_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_4_split_cast_fp16_1)[name = string("glu_4_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_4_cast_fp16 = mul(x = glu_4_split_cast_fp16_0, y = glu_4_split_1_sigmoid_cast_fp16)[name = string("glu_4_cast_fp16")];
            fp16 const_389_to_fp16 = const()[name = string("const_389_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_9_cast_fp16 = select(a = const_389_to_fp16, b = glu_4_cast_fp16, cond = all_1)[name = string("masked_fill_9_cast_fp16")];
            string conv1d_13_pad_type_0 = const()[name = string("conv1d_13_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_13_pad_0 = const()[name = string("conv1d_13_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_13_groups_0 = const()[name = string("conv1d_13_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_13_strides_0 = const()[name = string("conv1d_13_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_13_dilations_0 = const()[name = string("conv1d_13_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1555_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91330496))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91337472))))[name = string("const_1555_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1556_to_fp16 = const()[name = string("const_1556_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91345728)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_4_cast_fp16 = conv(bias = const_1556_to_fp16, dilations = conv1d_13_dilations_0, groups = conv1d_13_groups_0, pad = conv1d_13_pad_0, pad_type = conv1d_13_pad_type_0, strides = conv1d_13_strides_0, weight = const_1555_to_fp16_palettized, x = masked_fill_9_cast_fp16)[name = string("_native_batch_norm_legit_no_training_4_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_13_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_4_cast_fp16)[name = string("silu_13_cast_fp16")];
            string conv1d_14_pad_type_0 = const()[name = string("conv1d_14_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_14_strides_0 = const()[name = string("conv1d_14_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_14_pad_0 = const()[name = string("conv1d_14_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_14_dilations_0 = const()[name = string("conv1d_14_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_14_groups_0 = const()[name = string("conv1d_14_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_4_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91347840))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92134336))))[name = string("p_encoder_layers_4_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_14_cast_fp16 = conv(dilations = conv1d_14_dilations_0, groups = conv1d_14_groups_0, pad = conv1d_14_pad_0, pad_type = conv1d_14_pad_type_0, strides = conv1d_14_strides_0, weight = p_encoder_layers_4_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_13_cast_fp16)[name = string("conv1d_14_cast_fp16")];
            tensor<int32, [3]> transpose_32_perm_0 = const()[name = string("transpose_32_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_32_cast_fp16 = transpose(perm = transpose_32_perm_0, x = conv1d_14_cast_fp16)[name = string("transpose_114")];
            tensor<fp16, [1, 188, 1024]> add_43_cast_fp16 = add(x = add_42_cast_fp16, y = transpose_32_cast_fp16)[name = string("add_43_cast_fp16")];
            tensor<int32, [1]> layer_norm_23_axes_0 = const()[name = string("layer_norm_23_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_4_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_4_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92142592)))];
            tensor<fp16, [1024]> p_encoder_layers_4_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_4_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92144704)))];
            fp16 const_400_to_fp16 = const()[name = string("const_400_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_23_cast_fp16 = layer_norm(axes = layer_norm_23_axes_0, beta = p_encoder_layers_4_norm_feed_forward2_bias_to_fp16, epsilon = const_400_to_fp16, gamma = p_encoder_layers_4_norm_feed_forward2_weight_to_fp16, x = add_43_cast_fp16)[name = string("layer_norm_23_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_4_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92146816))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95292608))))[name = string("p_encoder_layers_4_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_44_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_4_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_23_cast_fp16)[name = string("linear_44_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_14_cast_fp16 = silu(x = linear_44_cast_fp16)[name = string("silu_14_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_4_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95325440))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98471232))))[name = string("p_encoder_layers_4_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_45_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_4_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_14_cast_fp16)[name = string("linear_45_cast_fp16")];
            fp16 const_402_to_fp16 = const()[name = string("const_402_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_20_cast_fp16 = mul(x = linear_45_cast_fp16, y = const_402_to_fp16)[name = string("mul_20_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_44_cast_fp16 = add(x = add_43_cast_fp16, y = mul_20_cast_fp16)[name = string("add_44_cast_fp16")];
            tensor<int32, [1]> layer_norm_24_axes_0 = const()[name = string("layer_norm_24_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_4_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_4_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98479488)))];
            tensor<fp16, [1024]> p_encoder_layers_4_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_4_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98481600)))];
            fp16 const_404_to_fp16 = const()[name = string("const_404_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_24_cast_fp16 = layer_norm(axes = layer_norm_24_axes_0, beta = p_encoder_layers_4_norm_out_bias_to_fp16, epsilon = const_404_to_fp16, gamma = p_encoder_layers_4_norm_out_weight_to_fp16, x = add_44_cast_fp16)[name = string("layer_norm_24_cast_fp16")];
            tensor<int32, [1]> layer_norm_25_axes_0 = const()[name = string("layer_norm_25_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_5_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_5_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98483712)))];
            tensor<fp16, [1024]> p_encoder_layers_5_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_5_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98485824)))];
            fp16 const_407_to_fp16 = const()[name = string("const_407_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_25_cast_fp16 = layer_norm(axes = layer_norm_25_axes_0, beta = p_encoder_layers_5_norm_feed_forward1_bias_to_fp16, epsilon = const_407_to_fp16, gamma = p_encoder_layers_5_norm_feed_forward1_weight_to_fp16, x = layer_norm_24_cast_fp16)[name = string("layer_norm_25_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_5_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98487936))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101633728))))[name = string("p_encoder_layers_5_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_46_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_5_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_25_cast_fp16)[name = string("linear_46_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_15_cast_fp16 = silu(x = linear_46_cast_fp16)[name = string("silu_15_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_5_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101666560))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104812352))))[name = string("p_encoder_layers_5_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_47_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_5_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_15_cast_fp16)[name = string("linear_47_cast_fp16")];
            fp16 const_409_to_fp16 = const()[name = string("const_409_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_21_cast_fp16 = mul(x = linear_47_cast_fp16, y = const_409_to_fp16)[name = string("mul_21_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_45_cast_fp16 = add(x = layer_norm_24_cast_fp16, y = mul_21_cast_fp16)[name = string("add_45_cast_fp16")];
            tensor<int32, [1]> layer_norm_26_axes_0 = const()[name = string("layer_norm_26_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_5_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_5_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104820608)))];
            tensor<fp16, [1024]> p_encoder_layers_5_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_5_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104822720)))];
            fp16 const_411_to_fp16 = const()[name = string("const_411_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_26_cast_fp16 = layer_norm(axes = layer_norm_26_axes_0, beta = p_encoder_layers_5_norm_self_att_bias_to_fp16, epsilon = const_411_to_fp16, gamma = p_encoder_layers_5_norm_self_att_weight_to_fp16, x = add_45_cast_fp16)[name = string("layer_norm_26_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104824832))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105611328))))[name = string("p_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_48_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_26_cast_fp16)[name = string("linear_48_cast_fp16")];
            tensor<int32, [4]> const_413 = const()[name = string("const_413"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_46_cast_fp16 = reshape(shape = const_413, x = linear_48_cast_fp16)[name = string("view_46_cast_fp16")];
            tensor<int32, [4]> transpose_33_perm_0 = const()[name = string("transpose_33_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105619584))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106406080))))[name = string("p_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_49_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_26_cast_fp16)[name = string("linear_49_cast_fp16")];
            tensor<int32, [4]> const_416 = const()[name = string("const_416"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_47_cast_fp16 = reshape(shape = const_416, x = linear_49_cast_fp16)[name = string("view_47_cast_fp16")];
            tensor<int32, [4]> transpose_34_perm_0 = const()[name = string("transpose_34_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106414336))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107200832))))[name = string("p_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_50_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_26_cast_fp16)[name = string("linear_50_cast_fp16")];
            tensor<int32, [4]> const_419 = const()[name = string("const_419"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_48_cast_fp16 = reshape(shape = const_419, x = linear_50_cast_fp16)[name = string("view_48_cast_fp16")];
            tensor<int32, [4]> transpose_35_perm_0 = const()[name = string("transpose_35_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_49_to_fp16 = const()[name = string("view_49_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107209088)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_33_cast_fp16 = transpose(perm = transpose_33_perm_0, x = view_46_cast_fp16)[name = string("transpose_113")];
            tensor<fp16, [1, 8, 188, 128]> add_46_cast_fp16 = add(x = transpose_33_cast_fp16, y = view_49_to_fp16)[name = string("add_46_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_50_to_fp16 = const()[name = string("view_50_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107211200)))];
            tensor<fp16, [1, 8, 188, 128]> add_47_cast_fp16 = add(x = transpose_33_cast_fp16, y = view_50_to_fp16)[name = string("add_47_cast_fp16")];
            bool matmul_6_transpose_x_0 = const()[name = string("matmul_6_transpose_x_0"), val = bool(false)];
            bool matmul_6_transpose_y_0 = const()[name = string("matmul_6_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_5_to_fp16 = const()[name = string("permute_5_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107213312)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_6_cast_fp16 = matmul(transpose_x = matmul_6_transpose_x_0, transpose_y = matmul_6_transpose_y_0, x = add_47_cast_fp16, y = permute_5_to_fp16)[name = string("matmul_6_cast_fp16")];
            tensor<int32, [8]> pad_5_pad_0 = const()[name = string("pad_5_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_5_mode_0 = const()[name = string("pad_5_mode_0"), val = string("constant")];
            fp16 const_427_to_fp16 = const()[name = string("const_427_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_5_cast_fp16 = pad(constant_val = const_427_to_fp16, mode = pad_5_mode_0, pad = pad_5_pad_0, x = matmul_6_cast_fp16)[name = string("pad_5_cast_fp16")];
            tensor<int32, [4]> const_428 = const()[name = string("const_428"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_52_cast_fp16 = reshape(shape = const_428, x = pad_5_cast_fp16)[name = string("view_52_cast_fp16")];
            tensor<int32, [4]> slice_11_begin_0 = const()[name = string("slice_11_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_11_end_0 = const()[name = string("slice_11_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_11_end_mask_0 = const()[name = string("slice_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_11_cast_fp16 = slice_by_index(begin = slice_11_begin_0, end = slice_11_end_0, end_mask = slice_11_end_mask_0, x = view_52_cast_fp16)[name = string("slice_11_cast_fp16")];
            tensor<int32, [4]> const_432 = const()[name = string("const_432"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_53_cast_fp16 = reshape(shape = const_432, x = slice_11_cast_fp16)[name = string("view_53_cast_fp16")];
            tensor<int32, [4]> slice_12_begin_0 = const()[name = string("slice_12_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_12_end_0 = const()[name = string("slice_12_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_12_end_mask_0 = const()[name = string("slice_12_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_12_cast_fp16 = slice_by_index(begin = slice_12_begin_0, end = slice_12_end_0, end_mask = slice_12_end_mask_0, x = view_53_cast_fp16)[name = string("slice_12_cast_fp16")];
            fp16 const_436_to_fp16 = const()[name = string("const_436_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_22_cast_fp16 = mul(x = slice_12_cast_fp16, y = const_436_to_fp16)[name = string("mul_22_cast_fp16")];
            fp16 const_437_to_fp16 = const()[name = string("const_437_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_10_cast_fp16 = select(a = const_437_to_fp16, b = mul_22_cast_fp16, cond = logical_not)[name = string("masked_fill_10_cast_fp16")];
            fp16 const_438_to_fp16 = const()[name = string("const_438_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_5_cast_fp16 = mul(x = add_46_cast_fp16, y = const_438_to_fp16)[name = string("mul_5_cast_fp16")];
            bool matmul_5_transpose_y_1 = const()[name = string("matmul_5_transpose_y_1"), val = bool(true)];
            bool matmul_5_transpose_x_1 = const()[name = string("matmul_5_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_34_cast_fp16 = transpose(perm = transpose_34_perm_0, x = view_47_cast_fp16)[name = string("transpose_112")];
            tensor<fp16, [1, 8, 188, 188]> matmul_5_1_cast_fp16 = matmul(transpose_x = matmul_5_transpose_x_1, transpose_y = matmul_5_transpose_y_1, x = mul_5_cast_fp16, y = transpose_34_cast_fp16)[name = string("matmul_5_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_5_1_cast_fp16 = add(x = matmul_5_1_cast_fp16, y = masked_fill_10_cast_fp16)[name = string("add_5_1_cast_fp16")];
            int32 softmax_5_axis_0 = const()[name = string("softmax_5_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_5_cast_fp16 = softmax(axis = softmax_5_axis_0, x = add_5_1_cast_fp16)[name = string("softmax_5_cast_fp16")];
            bool scaled_dot_product_attention_5_transpose_x_0 = const()[name = string("scaled_dot_product_attention_5_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_5_transpose_y_0 = const()[name = string("scaled_dot_product_attention_5_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_35_cast_fp16 = transpose(perm = transpose_35_perm_0, x = view_48_cast_fp16)[name = string("transpose_111")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_5_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_5_transpose_x_0, transpose_y = scaled_dot_product_attention_5_transpose_y_0, x = softmax_5_cast_fp16, y = transpose_35_cast_fp16)[name = string("scaled_dot_product_attention_5_cast_fp16")];
            tensor<int32, [4]> transpose_36_perm_0 = const()[name = string("transpose_36_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_441 = const()[name = string("const_441"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_36_cast_fp16 = transpose(perm = transpose_36_perm_0, x = scaled_dot_product_attention_5_cast_fp16)[name = string("transpose_110")];
            tensor<fp16, [1, 188, 1024]> view_54_cast_fp16 = reshape(shape = const_441, x = transpose_36_cast_fp16)[name = string("view_54_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_5_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107981376))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108767872))))[name = string("p_encoder_layers_5_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_52_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_5_self_attn_o_proj_weight_to_fp16_palettized, x = view_54_cast_fp16)[name = string("linear_52_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_48_cast_fp16 = add(x = add_45_cast_fp16, y = linear_52_cast_fp16)[name = string("add_48_cast_fp16")];
            tensor<int32, [1]> layer_norm_27_axes_0 = const()[name = string("layer_norm_27_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_5_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_5_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108776128)))];
            tensor<fp16, [1024]> p_encoder_layers_5_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_5_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108778240)))];
            fp16 const_443_to_fp16 = const()[name = string("const_443_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_27_cast_fp16 = layer_norm(axes = layer_norm_27_axes_0, beta = p_encoder_layers_5_norm_conv_bias_to_fp16, epsilon = const_443_to_fp16, gamma = p_encoder_layers_5_norm_conv_weight_to_fp16, x = add_48_cast_fp16)[name = string("layer_norm_27_cast_fp16")];
            tensor<int32, [3]> transpose_37_perm_0 = const()[name = string("transpose_37_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_15_pad_type_0 = const()[name = string("conv1d_15_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_15_strides_0 = const()[name = string("conv1d_15_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_15_pad_0 = const()[name = string("conv1d_15_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_15_dilations_0 = const()[name = string("conv1d_15_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_15_groups_0 = const()[name = string("conv1d_15_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_5_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108780352))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110353280))))[name = string("p_encoder_layers_5_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_37_cast_fp16 = transpose(perm = transpose_37_perm_0, x = layer_norm_27_cast_fp16)[name = string("transpose_109")];
            tensor<fp16, [1, 2048, 188]> conv1d_15_cast_fp16 = conv(dilations = conv1d_15_dilations_0, groups = conv1d_15_groups_0, pad = conv1d_15_pad_0, pad_type = conv1d_15_pad_type_0, strides = conv1d_15_strides_0, weight = p_encoder_layers_5_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_37_cast_fp16)[name = string("conv1d_15_cast_fp16")];
            int32 glu_5_split_num_splits_0 = const()[name = string("glu_5_split_num_splits_0"), val = int32(2)];
            int32 glu_5_split_axis_0 = const()[name = string("glu_5_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_5_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_5_split_cast_fp16_1 = split(axis = glu_5_split_axis_0, num_splits = glu_5_split_num_splits_0, x = conv1d_15_cast_fp16)[name = string("glu_5_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_5_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_5_split_cast_fp16_1)[name = string("glu_5_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_5_cast_fp16 = mul(x = glu_5_split_cast_fp16_0, y = glu_5_split_1_sigmoid_cast_fp16)[name = string("glu_5_cast_fp16")];
            fp16 const_449_to_fp16 = const()[name = string("const_449_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_11_cast_fp16 = select(a = const_449_to_fp16, b = glu_5_cast_fp16, cond = all_1)[name = string("masked_fill_11_cast_fp16")];
            string conv1d_16_pad_type_0 = const()[name = string("conv1d_16_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_16_pad_0 = const()[name = string("conv1d_16_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_16_groups_0 = const()[name = string("conv1d_16_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_16_strides_0 = const()[name = string("conv1d_16_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_16_dilations_0 = const()[name = string("conv1d_16_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1557_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110369728))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110376704))))[name = string("const_1557_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1558_to_fp16 = const()[name = string("const_1558_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110384960)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_5_cast_fp16 = conv(bias = const_1558_to_fp16, dilations = conv1d_16_dilations_0, groups = conv1d_16_groups_0, pad = conv1d_16_pad_0, pad_type = conv1d_16_pad_type_0, strides = conv1d_16_strides_0, weight = const_1557_to_fp16_palettized, x = masked_fill_11_cast_fp16)[name = string("_native_batch_norm_legit_no_training_5_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_16_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_5_cast_fp16)[name = string("silu_16_cast_fp16")];
            string conv1d_17_pad_type_0 = const()[name = string("conv1d_17_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_17_strides_0 = const()[name = string("conv1d_17_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_17_pad_0 = const()[name = string("conv1d_17_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_17_dilations_0 = const()[name = string("conv1d_17_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_17_groups_0 = const()[name = string("conv1d_17_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_5_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110387072))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111173568))))[name = string("p_encoder_layers_5_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_17_cast_fp16 = conv(dilations = conv1d_17_dilations_0, groups = conv1d_17_groups_0, pad = conv1d_17_pad_0, pad_type = conv1d_17_pad_type_0, strides = conv1d_17_strides_0, weight = p_encoder_layers_5_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_16_cast_fp16)[name = string("conv1d_17_cast_fp16")];
            tensor<int32, [3]> transpose_38_perm_0 = const()[name = string("transpose_38_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_38_cast_fp16 = transpose(perm = transpose_38_perm_0, x = conv1d_17_cast_fp16)[name = string("transpose_108")];
            tensor<fp16, [1, 188, 1024]> add_49_cast_fp16 = add(x = add_48_cast_fp16, y = transpose_38_cast_fp16)[name = string("add_49_cast_fp16")];
            tensor<int32, [1]> layer_norm_28_axes_0 = const()[name = string("layer_norm_28_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_5_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_5_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111181824)))];
            tensor<fp16, [1024]> p_encoder_layers_5_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_5_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111183936)))];
            fp16 const_460_to_fp16 = const()[name = string("const_460_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_28_cast_fp16 = layer_norm(axes = layer_norm_28_axes_0, beta = p_encoder_layers_5_norm_feed_forward2_bias_to_fp16, epsilon = const_460_to_fp16, gamma = p_encoder_layers_5_norm_feed_forward2_weight_to_fp16, x = add_49_cast_fp16)[name = string("layer_norm_28_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_5_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111186048))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114331840))))[name = string("p_encoder_layers_5_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_53_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_5_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_28_cast_fp16)[name = string("linear_53_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_17_cast_fp16 = silu(x = linear_53_cast_fp16)[name = string("silu_17_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_5_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114364672))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117510464))))[name = string("p_encoder_layers_5_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_54_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_5_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_17_cast_fp16)[name = string("linear_54_cast_fp16")];
            fp16 const_462_to_fp16 = const()[name = string("const_462_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_23_cast_fp16 = mul(x = linear_54_cast_fp16, y = const_462_to_fp16)[name = string("mul_23_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_50_cast_fp16 = add(x = add_49_cast_fp16, y = mul_23_cast_fp16)[name = string("add_50_cast_fp16")];
            tensor<int32, [1]> layer_norm_29_axes_0 = const()[name = string("layer_norm_29_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_5_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_5_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117518720)))];
            tensor<fp16, [1024]> p_encoder_layers_5_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_5_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117520832)))];
            fp16 const_464_to_fp16 = const()[name = string("const_464_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_29_cast_fp16 = layer_norm(axes = layer_norm_29_axes_0, beta = p_encoder_layers_5_norm_out_bias_to_fp16, epsilon = const_464_to_fp16, gamma = p_encoder_layers_5_norm_out_weight_to_fp16, x = add_50_cast_fp16)[name = string("layer_norm_29_cast_fp16")];
            tensor<int32, [1]> layer_norm_30_axes_0 = const()[name = string("layer_norm_30_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_6_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_6_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117522944)))];
            tensor<fp16, [1024]> p_encoder_layers_6_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_6_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117525056)))];
            fp16 const_467_to_fp16 = const()[name = string("const_467_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_30_cast_fp16 = layer_norm(axes = layer_norm_30_axes_0, beta = p_encoder_layers_6_norm_feed_forward1_bias_to_fp16, epsilon = const_467_to_fp16, gamma = p_encoder_layers_6_norm_feed_forward1_weight_to_fp16, x = layer_norm_29_cast_fp16)[name = string("layer_norm_30_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_6_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117527168))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120672960))))[name = string("p_encoder_layers_6_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_6_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_30_cast_fp16)[name = string("linear_55_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_18_cast_fp16 = silu(x = linear_55_cast_fp16)[name = string("silu_18_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_6_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120705792))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123851584))))[name = string("p_encoder_layers_6_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_56_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_6_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_18_cast_fp16)[name = string("linear_56_cast_fp16")];
            fp16 const_469_to_fp16 = const()[name = string("const_469_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_24_cast_fp16 = mul(x = linear_56_cast_fp16, y = const_469_to_fp16)[name = string("mul_24_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_51_cast_fp16 = add(x = layer_norm_29_cast_fp16, y = mul_24_cast_fp16)[name = string("add_51_cast_fp16")];
            tensor<int32, [1]> layer_norm_31_axes_0 = const()[name = string("layer_norm_31_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_6_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_6_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123859840)))];
            tensor<fp16, [1024]> p_encoder_layers_6_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_6_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123861952)))];
            fp16 const_471_to_fp16 = const()[name = string("const_471_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_31_cast_fp16 = layer_norm(axes = layer_norm_31_axes_0, beta = p_encoder_layers_6_norm_self_att_bias_to_fp16, epsilon = const_471_to_fp16, gamma = p_encoder_layers_6_norm_self_att_weight_to_fp16, x = add_51_cast_fp16)[name = string("layer_norm_31_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123864064))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124650560))))[name = string("p_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_57_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_31_cast_fp16)[name = string("linear_57_cast_fp16")];
            tensor<int32, [4]> const_473 = const()[name = string("const_473"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_55_cast_fp16 = reshape(shape = const_473, x = linear_57_cast_fp16)[name = string("view_55_cast_fp16")];
            tensor<int32, [4]> transpose_39_perm_0 = const()[name = string("transpose_39_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124658816))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125445312))))[name = string("p_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_58_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_31_cast_fp16)[name = string("linear_58_cast_fp16")];
            tensor<int32, [4]> const_476 = const()[name = string("const_476"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_56_cast_fp16 = reshape(shape = const_476, x = linear_58_cast_fp16)[name = string("view_56_cast_fp16")];
            tensor<int32, [4]> transpose_40_perm_0 = const()[name = string("transpose_40_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125453568))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126240064))))[name = string("p_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_59_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_31_cast_fp16)[name = string("linear_59_cast_fp16")];
            tensor<int32, [4]> const_479 = const()[name = string("const_479"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_57_cast_fp16 = reshape(shape = const_479, x = linear_59_cast_fp16)[name = string("view_57_cast_fp16")];
            tensor<int32, [4]> transpose_41_perm_0 = const()[name = string("transpose_41_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_58_to_fp16 = const()[name = string("view_58_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126248320)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_39_cast_fp16 = transpose(perm = transpose_39_perm_0, x = view_55_cast_fp16)[name = string("transpose_107")];
            tensor<fp16, [1, 8, 188, 128]> add_52_cast_fp16 = add(x = transpose_39_cast_fp16, y = view_58_to_fp16)[name = string("add_52_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_59_to_fp16 = const()[name = string("view_59_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126250432)))];
            tensor<fp16, [1, 8, 188, 128]> add_53_cast_fp16 = add(x = transpose_39_cast_fp16, y = view_59_to_fp16)[name = string("add_53_cast_fp16")];
            bool matmul_7_transpose_x_0 = const()[name = string("matmul_7_transpose_x_0"), val = bool(false)];
            bool matmul_7_transpose_y_0 = const()[name = string("matmul_7_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_6_to_fp16 = const()[name = string("permute_6_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126252544)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_7_cast_fp16 = matmul(transpose_x = matmul_7_transpose_x_0, transpose_y = matmul_7_transpose_y_0, x = add_53_cast_fp16, y = permute_6_to_fp16)[name = string("matmul_7_cast_fp16")];
            tensor<int32, [8]> pad_6_pad_0 = const()[name = string("pad_6_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_6_mode_0 = const()[name = string("pad_6_mode_0"), val = string("constant")];
            fp16 const_487_to_fp16 = const()[name = string("const_487_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_6_cast_fp16 = pad(constant_val = const_487_to_fp16, mode = pad_6_mode_0, pad = pad_6_pad_0, x = matmul_7_cast_fp16)[name = string("pad_6_cast_fp16")];
            tensor<int32, [4]> const_488 = const()[name = string("const_488"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_61_cast_fp16 = reshape(shape = const_488, x = pad_6_cast_fp16)[name = string("view_61_cast_fp16")];
            tensor<int32, [4]> slice_13_begin_0 = const()[name = string("slice_13_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_13_end_0 = const()[name = string("slice_13_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_13_end_mask_0 = const()[name = string("slice_13_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_13_cast_fp16 = slice_by_index(begin = slice_13_begin_0, end = slice_13_end_0, end_mask = slice_13_end_mask_0, x = view_61_cast_fp16)[name = string("slice_13_cast_fp16")];
            tensor<int32, [4]> const_492 = const()[name = string("const_492"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_62_cast_fp16 = reshape(shape = const_492, x = slice_13_cast_fp16)[name = string("view_62_cast_fp16")];
            tensor<int32, [4]> slice_14_begin_0 = const()[name = string("slice_14_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_14_end_0 = const()[name = string("slice_14_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_14_end_mask_0 = const()[name = string("slice_14_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_14_cast_fp16 = slice_by_index(begin = slice_14_begin_0, end = slice_14_end_0, end_mask = slice_14_end_mask_0, x = view_62_cast_fp16)[name = string("slice_14_cast_fp16")];
            fp16 const_496_to_fp16 = const()[name = string("const_496_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_25_cast_fp16 = mul(x = slice_14_cast_fp16, y = const_496_to_fp16)[name = string("mul_25_cast_fp16")];
            fp16 const_497_to_fp16 = const()[name = string("const_497_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_12_cast_fp16 = select(a = const_497_to_fp16, b = mul_25_cast_fp16, cond = logical_not)[name = string("masked_fill_12_cast_fp16")];
            fp16 const_498_to_fp16 = const()[name = string("const_498_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_6_1_cast_fp16 = mul(x = add_52_cast_fp16, y = const_498_to_fp16)[name = string("mul_6_1_cast_fp16")];
            bool matmul_6_transpose_y_1 = const()[name = string("matmul_6_transpose_y_1"), val = bool(true)];
            bool matmul_6_transpose_x_1 = const()[name = string("matmul_6_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_40_cast_fp16 = transpose(perm = transpose_40_perm_0, x = view_56_cast_fp16)[name = string("transpose_106")];
            tensor<fp16, [1, 8, 188, 188]> matmul_6_1_cast_fp16 = matmul(transpose_x = matmul_6_transpose_x_1, transpose_y = matmul_6_transpose_y_1, x = mul_6_1_cast_fp16, y = transpose_40_cast_fp16)[name = string("matmul_6_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_6_1_cast_fp16 = add(x = matmul_6_1_cast_fp16, y = masked_fill_12_cast_fp16)[name = string("add_6_1_cast_fp16")];
            int32 softmax_6_axis_0 = const()[name = string("softmax_6_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_6_cast_fp16 = softmax(axis = softmax_6_axis_0, x = add_6_1_cast_fp16)[name = string("softmax_6_cast_fp16")];
            bool scaled_dot_product_attention_6_transpose_x_0 = const()[name = string("scaled_dot_product_attention_6_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_6_transpose_y_0 = const()[name = string("scaled_dot_product_attention_6_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_41_cast_fp16 = transpose(perm = transpose_41_perm_0, x = view_57_cast_fp16)[name = string("transpose_105")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_6_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_6_transpose_x_0, transpose_y = scaled_dot_product_attention_6_transpose_y_0, x = softmax_6_cast_fp16, y = transpose_41_cast_fp16)[name = string("scaled_dot_product_attention_6_cast_fp16")];
            tensor<int32, [4]> transpose_42_perm_0 = const()[name = string("transpose_42_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_501 = const()[name = string("const_501"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_42_cast_fp16 = transpose(perm = transpose_42_perm_0, x = scaled_dot_product_attention_6_cast_fp16)[name = string("transpose_104")];
            tensor<fp16, [1, 188, 1024]> view_63_cast_fp16 = reshape(shape = const_501, x = transpose_42_cast_fp16)[name = string("view_63_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_6_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127020608))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127807104))))[name = string("p_encoder_layers_6_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_61_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_6_self_attn_o_proj_weight_to_fp16_palettized, x = view_63_cast_fp16)[name = string("linear_61_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_54_cast_fp16 = add(x = add_51_cast_fp16, y = linear_61_cast_fp16)[name = string("add_54_cast_fp16")];
            tensor<int32, [1]> layer_norm_32_axes_0 = const()[name = string("layer_norm_32_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_6_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_6_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127815360)))];
            tensor<fp16, [1024]> p_encoder_layers_6_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_6_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127817472)))];
            fp16 const_503_to_fp16 = const()[name = string("const_503_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_32_cast_fp16 = layer_norm(axes = layer_norm_32_axes_0, beta = p_encoder_layers_6_norm_conv_bias_to_fp16, epsilon = const_503_to_fp16, gamma = p_encoder_layers_6_norm_conv_weight_to_fp16, x = add_54_cast_fp16)[name = string("layer_norm_32_cast_fp16")];
            tensor<int32, [3]> transpose_43_perm_0 = const()[name = string("transpose_43_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_18_pad_type_0 = const()[name = string("conv1d_18_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_18_strides_0 = const()[name = string("conv1d_18_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_18_pad_0 = const()[name = string("conv1d_18_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_18_dilations_0 = const()[name = string("conv1d_18_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_18_groups_0 = const()[name = string("conv1d_18_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_6_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127819584))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129392512))))[name = string("p_encoder_layers_6_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_43_cast_fp16 = transpose(perm = transpose_43_perm_0, x = layer_norm_32_cast_fp16)[name = string("transpose_103")];
            tensor<fp16, [1, 2048, 188]> conv1d_18_cast_fp16 = conv(dilations = conv1d_18_dilations_0, groups = conv1d_18_groups_0, pad = conv1d_18_pad_0, pad_type = conv1d_18_pad_type_0, strides = conv1d_18_strides_0, weight = p_encoder_layers_6_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_43_cast_fp16)[name = string("conv1d_18_cast_fp16")];
            int32 glu_6_split_num_splits_0 = const()[name = string("glu_6_split_num_splits_0"), val = int32(2)];
            int32 glu_6_split_axis_0 = const()[name = string("glu_6_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_6_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_6_split_cast_fp16_1 = split(axis = glu_6_split_axis_0, num_splits = glu_6_split_num_splits_0, x = conv1d_18_cast_fp16)[name = string("glu_6_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_6_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_6_split_cast_fp16_1)[name = string("glu_6_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_6_cast_fp16 = mul(x = glu_6_split_cast_fp16_0, y = glu_6_split_1_sigmoid_cast_fp16)[name = string("glu_6_cast_fp16")];
            fp16 const_509_to_fp16 = const()[name = string("const_509_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_13_cast_fp16 = select(a = const_509_to_fp16, b = glu_6_cast_fp16, cond = all_1)[name = string("masked_fill_13_cast_fp16")];
            string conv1d_19_pad_type_0 = const()[name = string("conv1d_19_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_19_pad_0 = const()[name = string("conv1d_19_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_19_groups_0 = const()[name = string("conv1d_19_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_19_strides_0 = const()[name = string("conv1d_19_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_19_dilations_0 = const()[name = string("conv1d_19_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1559_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129408960))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129415936))))[name = string("const_1559_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1560_to_fp16 = const()[name = string("const_1560_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129424192)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_6_cast_fp16 = conv(bias = const_1560_to_fp16, dilations = conv1d_19_dilations_0, groups = conv1d_19_groups_0, pad = conv1d_19_pad_0, pad_type = conv1d_19_pad_type_0, strides = conv1d_19_strides_0, weight = const_1559_to_fp16_palettized, x = masked_fill_13_cast_fp16)[name = string("_native_batch_norm_legit_no_training_6_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_19_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_6_cast_fp16)[name = string("silu_19_cast_fp16")];
            string conv1d_20_pad_type_0 = const()[name = string("conv1d_20_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_20_strides_0 = const()[name = string("conv1d_20_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_20_pad_0 = const()[name = string("conv1d_20_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_20_dilations_0 = const()[name = string("conv1d_20_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_20_groups_0 = const()[name = string("conv1d_20_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_6_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129426304))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130212800))))[name = string("p_encoder_layers_6_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_20_cast_fp16 = conv(dilations = conv1d_20_dilations_0, groups = conv1d_20_groups_0, pad = conv1d_20_pad_0, pad_type = conv1d_20_pad_type_0, strides = conv1d_20_strides_0, weight = p_encoder_layers_6_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_19_cast_fp16)[name = string("conv1d_20_cast_fp16")];
            tensor<int32, [3]> transpose_44_perm_0 = const()[name = string("transpose_44_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_44_cast_fp16 = transpose(perm = transpose_44_perm_0, x = conv1d_20_cast_fp16)[name = string("transpose_102")];
            tensor<fp16, [1, 188, 1024]> add_55_cast_fp16 = add(x = add_54_cast_fp16, y = transpose_44_cast_fp16)[name = string("add_55_cast_fp16")];
            tensor<int32, [1]> layer_norm_33_axes_0 = const()[name = string("layer_norm_33_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_6_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_6_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130221056)))];
            tensor<fp16, [1024]> p_encoder_layers_6_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_6_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130223168)))];
            fp16 const_520_to_fp16 = const()[name = string("const_520_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_33_cast_fp16 = layer_norm(axes = layer_norm_33_axes_0, beta = p_encoder_layers_6_norm_feed_forward2_bias_to_fp16, epsilon = const_520_to_fp16, gamma = p_encoder_layers_6_norm_feed_forward2_weight_to_fp16, x = add_55_cast_fp16)[name = string("layer_norm_33_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_6_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130225280))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133371072))))[name = string("p_encoder_layers_6_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_62_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_6_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_33_cast_fp16)[name = string("linear_62_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_20_cast_fp16 = silu(x = linear_62_cast_fp16)[name = string("silu_20_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_6_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133403904))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136549696))))[name = string("p_encoder_layers_6_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_63_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_6_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_20_cast_fp16)[name = string("linear_63_cast_fp16")];
            fp16 const_522_to_fp16 = const()[name = string("const_522_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_26_cast_fp16 = mul(x = linear_63_cast_fp16, y = const_522_to_fp16)[name = string("mul_26_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_56_cast_fp16 = add(x = add_55_cast_fp16, y = mul_26_cast_fp16)[name = string("add_56_cast_fp16")];
            tensor<int32, [1]> layer_norm_34_axes_0 = const()[name = string("layer_norm_34_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_6_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_6_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136557952)))];
            tensor<fp16, [1024]> p_encoder_layers_6_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_6_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136560064)))];
            fp16 const_524_to_fp16 = const()[name = string("const_524_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_34_cast_fp16 = layer_norm(axes = layer_norm_34_axes_0, beta = p_encoder_layers_6_norm_out_bias_to_fp16, epsilon = const_524_to_fp16, gamma = p_encoder_layers_6_norm_out_weight_to_fp16, x = add_56_cast_fp16)[name = string("layer_norm_34_cast_fp16")];
            tensor<int32, [1]> layer_norm_35_axes_0 = const()[name = string("layer_norm_35_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_7_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_7_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136562176)))];
            tensor<fp16, [1024]> p_encoder_layers_7_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_7_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136564288)))];
            fp16 const_527_to_fp16 = const()[name = string("const_527_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_35_cast_fp16 = layer_norm(axes = layer_norm_35_axes_0, beta = p_encoder_layers_7_norm_feed_forward1_bias_to_fp16, epsilon = const_527_to_fp16, gamma = p_encoder_layers_7_norm_feed_forward1_weight_to_fp16, x = layer_norm_34_cast_fp16)[name = string("layer_norm_35_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_7_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136566400))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139712192))))[name = string("p_encoder_layers_7_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_64_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_7_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_35_cast_fp16)[name = string("linear_64_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_21_cast_fp16 = silu(x = linear_64_cast_fp16)[name = string("silu_21_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_7_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139745024))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142890816))))[name = string("p_encoder_layers_7_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_65_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_7_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_21_cast_fp16)[name = string("linear_65_cast_fp16")];
            fp16 const_529_to_fp16 = const()[name = string("const_529_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_27_cast_fp16 = mul(x = linear_65_cast_fp16, y = const_529_to_fp16)[name = string("mul_27_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_57_cast_fp16 = add(x = layer_norm_34_cast_fp16, y = mul_27_cast_fp16)[name = string("add_57_cast_fp16")];
            tensor<int32, [1]> layer_norm_36_axes_0 = const()[name = string("layer_norm_36_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_7_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_7_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142899072)))];
            tensor<fp16, [1024]> p_encoder_layers_7_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_7_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142901184)))];
            fp16 const_531_to_fp16 = const()[name = string("const_531_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_36_cast_fp16 = layer_norm(axes = layer_norm_36_axes_0, beta = p_encoder_layers_7_norm_self_att_bias_to_fp16, epsilon = const_531_to_fp16, gamma = p_encoder_layers_7_norm_self_att_weight_to_fp16, x = add_57_cast_fp16)[name = string("layer_norm_36_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142903296))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143689792))))[name = string("p_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_66_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_36_cast_fp16)[name = string("linear_66_cast_fp16")];
            tensor<int32, [4]> const_533 = const()[name = string("const_533"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_64_cast_fp16 = reshape(shape = const_533, x = linear_66_cast_fp16)[name = string("view_64_cast_fp16")];
            tensor<int32, [4]> transpose_45_perm_0 = const()[name = string("transpose_45_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143698048))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144484544))))[name = string("p_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_67_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_36_cast_fp16)[name = string("linear_67_cast_fp16")];
            tensor<int32, [4]> const_536 = const()[name = string("const_536"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_65_cast_fp16 = reshape(shape = const_536, x = linear_67_cast_fp16)[name = string("view_65_cast_fp16")];
            tensor<int32, [4]> transpose_46_perm_0 = const()[name = string("transpose_46_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144492800))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145279296))))[name = string("p_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_68_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_36_cast_fp16)[name = string("linear_68_cast_fp16")];
            tensor<int32, [4]> const_539 = const()[name = string("const_539"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_66_cast_fp16 = reshape(shape = const_539, x = linear_68_cast_fp16)[name = string("view_66_cast_fp16")];
            tensor<int32, [4]> transpose_47_perm_0 = const()[name = string("transpose_47_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_67_to_fp16 = const()[name = string("view_67_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145287552)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_45_cast_fp16 = transpose(perm = transpose_45_perm_0, x = view_64_cast_fp16)[name = string("transpose_101")];
            tensor<fp16, [1, 8, 188, 128]> add_58_cast_fp16 = add(x = transpose_45_cast_fp16, y = view_67_to_fp16)[name = string("add_58_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_68_to_fp16 = const()[name = string("view_68_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145289664)))];
            tensor<fp16, [1, 8, 188, 128]> add_59_cast_fp16 = add(x = transpose_45_cast_fp16, y = view_68_to_fp16)[name = string("add_59_cast_fp16")];
            bool matmul_8_transpose_x_0 = const()[name = string("matmul_8_transpose_x_0"), val = bool(false)];
            bool matmul_8_transpose_y_0 = const()[name = string("matmul_8_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_7_to_fp16 = const()[name = string("permute_7_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145291776)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_8_cast_fp16 = matmul(transpose_x = matmul_8_transpose_x_0, transpose_y = matmul_8_transpose_y_0, x = add_59_cast_fp16, y = permute_7_to_fp16)[name = string("matmul_8_cast_fp16")];
            tensor<int32, [8]> pad_7_pad_0 = const()[name = string("pad_7_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_7_mode_0 = const()[name = string("pad_7_mode_0"), val = string("constant")];
            fp16 const_547_to_fp16 = const()[name = string("const_547_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_7_cast_fp16 = pad(constant_val = const_547_to_fp16, mode = pad_7_mode_0, pad = pad_7_pad_0, x = matmul_8_cast_fp16)[name = string("pad_7_cast_fp16")];
            tensor<int32, [4]> const_548 = const()[name = string("const_548"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_70_cast_fp16 = reshape(shape = const_548, x = pad_7_cast_fp16)[name = string("view_70_cast_fp16")];
            tensor<int32, [4]> slice_15_begin_0 = const()[name = string("slice_15_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_15_end_0 = const()[name = string("slice_15_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_15_end_mask_0 = const()[name = string("slice_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_15_cast_fp16 = slice_by_index(begin = slice_15_begin_0, end = slice_15_end_0, end_mask = slice_15_end_mask_0, x = view_70_cast_fp16)[name = string("slice_15_cast_fp16")];
            tensor<int32, [4]> const_552 = const()[name = string("const_552"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_71_cast_fp16 = reshape(shape = const_552, x = slice_15_cast_fp16)[name = string("view_71_cast_fp16")];
            tensor<int32, [4]> slice_16_begin_0 = const()[name = string("slice_16_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_16_end_0 = const()[name = string("slice_16_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_16_end_mask_0 = const()[name = string("slice_16_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_16_cast_fp16 = slice_by_index(begin = slice_16_begin_0, end = slice_16_end_0, end_mask = slice_16_end_mask_0, x = view_71_cast_fp16)[name = string("slice_16_cast_fp16")];
            fp16 const_556_to_fp16 = const()[name = string("const_556_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_28_cast_fp16 = mul(x = slice_16_cast_fp16, y = const_556_to_fp16)[name = string("mul_28_cast_fp16")];
            fp16 const_557_to_fp16 = const()[name = string("const_557_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_14_cast_fp16 = select(a = const_557_to_fp16, b = mul_28_cast_fp16, cond = logical_not)[name = string("masked_fill_14_cast_fp16")];
            fp16 const_558_to_fp16 = const()[name = string("const_558_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_7_1_cast_fp16 = mul(x = add_58_cast_fp16, y = const_558_to_fp16)[name = string("mul_7_1_cast_fp16")];
            bool matmul_7_transpose_y_1 = const()[name = string("matmul_7_transpose_y_1"), val = bool(true)];
            bool matmul_7_transpose_x_1 = const()[name = string("matmul_7_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_46_cast_fp16 = transpose(perm = transpose_46_perm_0, x = view_65_cast_fp16)[name = string("transpose_100")];
            tensor<fp16, [1, 8, 188, 188]> matmul_7_1_cast_fp16 = matmul(transpose_x = matmul_7_transpose_x_1, transpose_y = matmul_7_transpose_y_1, x = mul_7_1_cast_fp16, y = transpose_46_cast_fp16)[name = string("matmul_7_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_7_1_cast_fp16 = add(x = matmul_7_1_cast_fp16, y = masked_fill_14_cast_fp16)[name = string("add_7_1_cast_fp16")];
            int32 softmax_7_axis_0 = const()[name = string("softmax_7_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_7_cast_fp16 = softmax(axis = softmax_7_axis_0, x = add_7_1_cast_fp16)[name = string("softmax_7_cast_fp16")];
            bool scaled_dot_product_attention_7_transpose_x_0 = const()[name = string("scaled_dot_product_attention_7_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_7_transpose_y_0 = const()[name = string("scaled_dot_product_attention_7_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_47_cast_fp16 = transpose(perm = transpose_47_perm_0, x = view_66_cast_fp16)[name = string("transpose_99")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_7_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_7_transpose_x_0, transpose_y = scaled_dot_product_attention_7_transpose_y_0, x = softmax_7_cast_fp16, y = transpose_47_cast_fp16)[name = string("scaled_dot_product_attention_7_cast_fp16")];
            tensor<int32, [4]> transpose_48_perm_0 = const()[name = string("transpose_48_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_561 = const()[name = string("const_561"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_48_cast_fp16 = transpose(perm = transpose_48_perm_0, x = scaled_dot_product_attention_7_cast_fp16)[name = string("transpose_98")];
            tensor<fp16, [1, 188, 1024]> view_72_cast_fp16 = reshape(shape = const_561, x = transpose_48_cast_fp16)[name = string("view_72_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_7_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146059840))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146846336))))[name = string("p_encoder_layers_7_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_70_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_7_self_attn_o_proj_weight_to_fp16_palettized, x = view_72_cast_fp16)[name = string("linear_70_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_60_cast_fp16 = add(x = add_57_cast_fp16, y = linear_70_cast_fp16)[name = string("add_60_cast_fp16")];
            tensor<int32, [1]> layer_norm_37_axes_0 = const()[name = string("layer_norm_37_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_7_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_7_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146854592)))];
            tensor<fp16, [1024]> p_encoder_layers_7_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_7_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146856704)))];
            fp16 const_563_to_fp16 = const()[name = string("const_563_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_37_cast_fp16 = layer_norm(axes = layer_norm_37_axes_0, beta = p_encoder_layers_7_norm_conv_bias_to_fp16, epsilon = const_563_to_fp16, gamma = p_encoder_layers_7_norm_conv_weight_to_fp16, x = add_60_cast_fp16)[name = string("layer_norm_37_cast_fp16")];
            tensor<int32, [3]> transpose_49_perm_0 = const()[name = string("transpose_49_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_21_pad_type_0 = const()[name = string("conv1d_21_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_21_strides_0 = const()[name = string("conv1d_21_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_21_pad_0 = const()[name = string("conv1d_21_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_21_dilations_0 = const()[name = string("conv1d_21_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_21_groups_0 = const()[name = string("conv1d_21_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_7_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146858816))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148431744))))[name = string("p_encoder_layers_7_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_49_cast_fp16 = transpose(perm = transpose_49_perm_0, x = layer_norm_37_cast_fp16)[name = string("transpose_97")];
            tensor<fp16, [1, 2048, 188]> conv1d_21_cast_fp16 = conv(dilations = conv1d_21_dilations_0, groups = conv1d_21_groups_0, pad = conv1d_21_pad_0, pad_type = conv1d_21_pad_type_0, strides = conv1d_21_strides_0, weight = p_encoder_layers_7_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_49_cast_fp16)[name = string("conv1d_21_cast_fp16")];
            int32 glu_7_split_num_splits_0 = const()[name = string("glu_7_split_num_splits_0"), val = int32(2)];
            int32 glu_7_split_axis_0 = const()[name = string("glu_7_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_7_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_7_split_cast_fp16_1 = split(axis = glu_7_split_axis_0, num_splits = glu_7_split_num_splits_0, x = conv1d_21_cast_fp16)[name = string("glu_7_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_7_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_7_split_cast_fp16_1)[name = string("glu_7_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_7_cast_fp16 = mul(x = glu_7_split_cast_fp16_0, y = glu_7_split_1_sigmoid_cast_fp16)[name = string("glu_7_cast_fp16")];
            fp16 const_569_to_fp16 = const()[name = string("const_569_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_15_cast_fp16 = select(a = const_569_to_fp16, b = glu_7_cast_fp16, cond = all_1)[name = string("masked_fill_15_cast_fp16")];
            string conv1d_22_pad_type_0 = const()[name = string("conv1d_22_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_22_pad_0 = const()[name = string("conv1d_22_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_22_groups_0 = const()[name = string("conv1d_22_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_22_strides_0 = const()[name = string("conv1d_22_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_22_dilations_0 = const()[name = string("conv1d_22_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1561_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148448192))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148455168))))[name = string("const_1561_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1562_to_fp16 = const()[name = string("const_1562_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148463424)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_7_cast_fp16 = conv(bias = const_1562_to_fp16, dilations = conv1d_22_dilations_0, groups = conv1d_22_groups_0, pad = conv1d_22_pad_0, pad_type = conv1d_22_pad_type_0, strides = conv1d_22_strides_0, weight = const_1561_to_fp16_palettized, x = masked_fill_15_cast_fp16)[name = string("_native_batch_norm_legit_no_training_7_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_22_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_7_cast_fp16)[name = string("silu_22_cast_fp16")];
            string conv1d_23_pad_type_0 = const()[name = string("conv1d_23_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_23_strides_0 = const()[name = string("conv1d_23_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_23_pad_0 = const()[name = string("conv1d_23_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_23_dilations_0 = const()[name = string("conv1d_23_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_23_groups_0 = const()[name = string("conv1d_23_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_7_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148465536))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149252032))))[name = string("p_encoder_layers_7_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_23_cast_fp16 = conv(dilations = conv1d_23_dilations_0, groups = conv1d_23_groups_0, pad = conv1d_23_pad_0, pad_type = conv1d_23_pad_type_0, strides = conv1d_23_strides_0, weight = p_encoder_layers_7_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_22_cast_fp16)[name = string("conv1d_23_cast_fp16")];
            tensor<int32, [3]> transpose_50_perm_0 = const()[name = string("transpose_50_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_50_cast_fp16 = transpose(perm = transpose_50_perm_0, x = conv1d_23_cast_fp16)[name = string("transpose_96")];
            tensor<fp16, [1, 188, 1024]> add_61_cast_fp16 = add(x = add_60_cast_fp16, y = transpose_50_cast_fp16)[name = string("add_61_cast_fp16")];
            tensor<int32, [1]> layer_norm_38_axes_0 = const()[name = string("layer_norm_38_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_7_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_7_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149260288)))];
            tensor<fp16, [1024]> p_encoder_layers_7_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_7_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149262400)))];
            fp16 const_580_to_fp16 = const()[name = string("const_580_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_38_cast_fp16 = layer_norm(axes = layer_norm_38_axes_0, beta = p_encoder_layers_7_norm_feed_forward2_bias_to_fp16, epsilon = const_580_to_fp16, gamma = p_encoder_layers_7_norm_feed_forward2_weight_to_fp16, x = add_61_cast_fp16)[name = string("layer_norm_38_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_7_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149264512))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152410304))))[name = string("p_encoder_layers_7_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_71_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_7_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_38_cast_fp16)[name = string("linear_71_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_23_cast_fp16 = silu(x = linear_71_cast_fp16)[name = string("silu_23_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_7_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152443136))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155588928))))[name = string("p_encoder_layers_7_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_72_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_7_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_23_cast_fp16)[name = string("linear_72_cast_fp16")];
            fp16 const_582_to_fp16 = const()[name = string("const_582_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_29_cast_fp16 = mul(x = linear_72_cast_fp16, y = const_582_to_fp16)[name = string("mul_29_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_62_cast_fp16 = add(x = add_61_cast_fp16, y = mul_29_cast_fp16)[name = string("add_62_cast_fp16")];
            tensor<int32, [1]> layer_norm_39_axes_0 = const()[name = string("layer_norm_39_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_7_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_7_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155597184)))];
            tensor<fp16, [1024]> p_encoder_layers_7_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_7_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155599296)))];
            fp16 const_584_to_fp16 = const()[name = string("const_584_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_39_cast_fp16 = layer_norm(axes = layer_norm_39_axes_0, beta = p_encoder_layers_7_norm_out_bias_to_fp16, epsilon = const_584_to_fp16, gamma = p_encoder_layers_7_norm_out_weight_to_fp16, x = add_62_cast_fp16)[name = string("layer_norm_39_cast_fp16")];
            tensor<int32, [1]> layer_norm_40_axes_0 = const()[name = string("layer_norm_40_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_8_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_8_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155601408)))];
            tensor<fp16, [1024]> p_encoder_layers_8_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_8_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155603520)))];
            fp16 const_587_to_fp16 = const()[name = string("const_587_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_40_cast_fp16 = layer_norm(axes = layer_norm_40_axes_0, beta = p_encoder_layers_8_norm_feed_forward1_bias_to_fp16, epsilon = const_587_to_fp16, gamma = p_encoder_layers_8_norm_feed_forward1_weight_to_fp16, x = layer_norm_39_cast_fp16)[name = string("layer_norm_40_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_8_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155605632))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158751424))))[name = string("p_encoder_layers_8_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_73_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_8_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_40_cast_fp16)[name = string("linear_73_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_24_cast_fp16 = silu(x = linear_73_cast_fp16)[name = string("silu_24_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_8_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158784256))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161930048))))[name = string("p_encoder_layers_8_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_74_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_8_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_24_cast_fp16)[name = string("linear_74_cast_fp16")];
            fp16 const_589_to_fp16 = const()[name = string("const_589_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_30_cast_fp16 = mul(x = linear_74_cast_fp16, y = const_589_to_fp16)[name = string("mul_30_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_63_cast_fp16 = add(x = layer_norm_39_cast_fp16, y = mul_30_cast_fp16)[name = string("add_63_cast_fp16")];
            tensor<int32, [1]> layer_norm_41_axes_0 = const()[name = string("layer_norm_41_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_8_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_8_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161938304)))];
            tensor<fp16, [1024]> p_encoder_layers_8_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_8_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161940416)))];
            fp16 const_591_to_fp16 = const()[name = string("const_591_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_41_cast_fp16 = layer_norm(axes = layer_norm_41_axes_0, beta = p_encoder_layers_8_norm_self_att_bias_to_fp16, epsilon = const_591_to_fp16, gamma = p_encoder_layers_8_norm_self_att_weight_to_fp16, x = add_63_cast_fp16)[name = string("layer_norm_41_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161942528))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162729024))))[name = string("p_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_75_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_41_cast_fp16)[name = string("linear_75_cast_fp16")];
            tensor<int32, [4]> const_593 = const()[name = string("const_593"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_73_cast_fp16 = reshape(shape = const_593, x = linear_75_cast_fp16)[name = string("view_73_cast_fp16")];
            tensor<int32, [4]> transpose_51_perm_0 = const()[name = string("transpose_51_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162737280))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163523776))))[name = string("p_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_76_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_41_cast_fp16)[name = string("linear_76_cast_fp16")];
            tensor<int32, [4]> const_596 = const()[name = string("const_596"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_74_cast_fp16 = reshape(shape = const_596, x = linear_76_cast_fp16)[name = string("view_74_cast_fp16")];
            tensor<int32, [4]> transpose_52_perm_0 = const()[name = string("transpose_52_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163532032))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164318528))))[name = string("p_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_77_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_41_cast_fp16)[name = string("linear_77_cast_fp16")];
            tensor<int32, [4]> const_599 = const()[name = string("const_599"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_75_cast_fp16 = reshape(shape = const_599, x = linear_77_cast_fp16)[name = string("view_75_cast_fp16")];
            tensor<int32, [4]> transpose_53_perm_0 = const()[name = string("transpose_53_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_76_to_fp16 = const()[name = string("view_76_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164326784)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_51_cast_fp16 = transpose(perm = transpose_51_perm_0, x = view_73_cast_fp16)[name = string("transpose_95")];
            tensor<fp16, [1, 8, 188, 128]> add_64_cast_fp16 = add(x = transpose_51_cast_fp16, y = view_76_to_fp16)[name = string("add_64_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_77_to_fp16 = const()[name = string("view_77_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164328896)))];
            tensor<fp16, [1, 8, 188, 128]> add_65_cast_fp16 = add(x = transpose_51_cast_fp16, y = view_77_to_fp16)[name = string("add_65_cast_fp16")];
            bool matmul_9_transpose_x_0 = const()[name = string("matmul_9_transpose_x_0"), val = bool(false)];
            bool matmul_9_transpose_y_0 = const()[name = string("matmul_9_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_8_to_fp16 = const()[name = string("permute_8_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164331008)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_9_cast_fp16 = matmul(transpose_x = matmul_9_transpose_x_0, transpose_y = matmul_9_transpose_y_0, x = add_65_cast_fp16, y = permute_8_to_fp16)[name = string("matmul_9_cast_fp16")];
            tensor<int32, [8]> pad_8_pad_0 = const()[name = string("pad_8_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_8_mode_0 = const()[name = string("pad_8_mode_0"), val = string("constant")];
            fp16 const_607_to_fp16 = const()[name = string("const_607_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_8_cast_fp16 = pad(constant_val = const_607_to_fp16, mode = pad_8_mode_0, pad = pad_8_pad_0, x = matmul_9_cast_fp16)[name = string("pad_8_cast_fp16")];
            tensor<int32, [4]> const_608 = const()[name = string("const_608"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_79_cast_fp16 = reshape(shape = const_608, x = pad_8_cast_fp16)[name = string("view_79_cast_fp16")];
            tensor<int32, [4]> slice_17_begin_0 = const()[name = string("slice_17_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_17_end_0 = const()[name = string("slice_17_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_17_end_mask_0 = const()[name = string("slice_17_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_17_cast_fp16 = slice_by_index(begin = slice_17_begin_0, end = slice_17_end_0, end_mask = slice_17_end_mask_0, x = view_79_cast_fp16)[name = string("slice_17_cast_fp16")];
            tensor<int32, [4]> const_612 = const()[name = string("const_612"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_80_cast_fp16 = reshape(shape = const_612, x = slice_17_cast_fp16)[name = string("view_80_cast_fp16")];
            tensor<int32, [4]> slice_18_begin_0 = const()[name = string("slice_18_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_18_end_0 = const()[name = string("slice_18_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_18_end_mask_0 = const()[name = string("slice_18_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_18_cast_fp16 = slice_by_index(begin = slice_18_begin_0, end = slice_18_end_0, end_mask = slice_18_end_mask_0, x = view_80_cast_fp16)[name = string("slice_18_cast_fp16")];
            fp16 const_616_to_fp16 = const()[name = string("const_616_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_31_cast_fp16 = mul(x = slice_18_cast_fp16, y = const_616_to_fp16)[name = string("mul_31_cast_fp16")];
            fp16 const_617_to_fp16 = const()[name = string("const_617_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_16_cast_fp16 = select(a = const_617_to_fp16, b = mul_31_cast_fp16, cond = logical_not)[name = string("masked_fill_16_cast_fp16")];
            fp16 const_618_to_fp16 = const()[name = string("const_618_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_8_1_cast_fp16 = mul(x = add_64_cast_fp16, y = const_618_to_fp16)[name = string("mul_8_1_cast_fp16")];
            bool matmul_8_transpose_y_1 = const()[name = string("matmul_8_transpose_y_1"), val = bool(true)];
            bool matmul_8_transpose_x_1 = const()[name = string("matmul_8_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_52_cast_fp16 = transpose(perm = transpose_52_perm_0, x = view_74_cast_fp16)[name = string("transpose_94")];
            tensor<fp16, [1, 8, 188, 188]> matmul_8_1_cast_fp16 = matmul(transpose_x = matmul_8_transpose_x_1, transpose_y = matmul_8_transpose_y_1, x = mul_8_1_cast_fp16, y = transpose_52_cast_fp16)[name = string("matmul_8_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_8_1_cast_fp16 = add(x = matmul_8_1_cast_fp16, y = masked_fill_16_cast_fp16)[name = string("add_8_1_cast_fp16")];
            int32 softmax_8_axis_0 = const()[name = string("softmax_8_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_8_cast_fp16 = softmax(axis = softmax_8_axis_0, x = add_8_1_cast_fp16)[name = string("softmax_8_cast_fp16")];
            bool scaled_dot_product_attention_8_transpose_x_0 = const()[name = string("scaled_dot_product_attention_8_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_8_transpose_y_0 = const()[name = string("scaled_dot_product_attention_8_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_53_cast_fp16 = transpose(perm = transpose_53_perm_0, x = view_75_cast_fp16)[name = string("transpose_93")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_8_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_8_transpose_x_0, transpose_y = scaled_dot_product_attention_8_transpose_y_0, x = softmax_8_cast_fp16, y = transpose_53_cast_fp16)[name = string("scaled_dot_product_attention_8_cast_fp16")];
            tensor<int32, [4]> transpose_54_perm_0 = const()[name = string("transpose_54_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_621 = const()[name = string("const_621"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_54_cast_fp16 = transpose(perm = transpose_54_perm_0, x = scaled_dot_product_attention_8_cast_fp16)[name = string("transpose_92")];
            tensor<fp16, [1, 188, 1024]> view_81_cast_fp16 = reshape(shape = const_621, x = transpose_54_cast_fp16)[name = string("view_81_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_8_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165099072))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165885568))))[name = string("p_encoder_layers_8_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_79_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_8_self_attn_o_proj_weight_to_fp16_palettized, x = view_81_cast_fp16)[name = string("linear_79_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_66_cast_fp16 = add(x = add_63_cast_fp16, y = linear_79_cast_fp16)[name = string("add_66_cast_fp16")];
            tensor<int32, [1]> layer_norm_42_axes_0 = const()[name = string("layer_norm_42_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_8_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_8_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165893824)))];
            tensor<fp16, [1024]> p_encoder_layers_8_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_8_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165895936)))];
            fp16 const_623_to_fp16 = const()[name = string("const_623_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_42_cast_fp16 = layer_norm(axes = layer_norm_42_axes_0, beta = p_encoder_layers_8_norm_conv_bias_to_fp16, epsilon = const_623_to_fp16, gamma = p_encoder_layers_8_norm_conv_weight_to_fp16, x = add_66_cast_fp16)[name = string("layer_norm_42_cast_fp16")];
            tensor<int32, [3]> transpose_55_perm_0 = const()[name = string("transpose_55_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_24_pad_type_0 = const()[name = string("conv1d_24_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_24_strides_0 = const()[name = string("conv1d_24_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_24_pad_0 = const()[name = string("conv1d_24_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_24_dilations_0 = const()[name = string("conv1d_24_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_24_groups_0 = const()[name = string("conv1d_24_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_8_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165898048))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167470976))))[name = string("p_encoder_layers_8_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_55_cast_fp16 = transpose(perm = transpose_55_perm_0, x = layer_norm_42_cast_fp16)[name = string("transpose_91")];
            tensor<fp16, [1, 2048, 188]> conv1d_24_cast_fp16 = conv(dilations = conv1d_24_dilations_0, groups = conv1d_24_groups_0, pad = conv1d_24_pad_0, pad_type = conv1d_24_pad_type_0, strides = conv1d_24_strides_0, weight = p_encoder_layers_8_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_55_cast_fp16)[name = string("conv1d_24_cast_fp16")];
            int32 glu_8_split_num_splits_0 = const()[name = string("glu_8_split_num_splits_0"), val = int32(2)];
            int32 glu_8_split_axis_0 = const()[name = string("glu_8_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_8_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_8_split_cast_fp16_1 = split(axis = glu_8_split_axis_0, num_splits = glu_8_split_num_splits_0, x = conv1d_24_cast_fp16)[name = string("glu_8_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_8_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_8_split_cast_fp16_1)[name = string("glu_8_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_8_cast_fp16 = mul(x = glu_8_split_cast_fp16_0, y = glu_8_split_1_sigmoid_cast_fp16)[name = string("glu_8_cast_fp16")];
            fp16 const_629_to_fp16 = const()[name = string("const_629_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_17_cast_fp16 = select(a = const_629_to_fp16, b = glu_8_cast_fp16, cond = all_1)[name = string("masked_fill_17_cast_fp16")];
            string conv1d_25_pad_type_0 = const()[name = string("conv1d_25_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_25_pad_0 = const()[name = string("conv1d_25_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_25_groups_0 = const()[name = string("conv1d_25_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_25_strides_0 = const()[name = string("conv1d_25_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_25_dilations_0 = const()[name = string("conv1d_25_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1563_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167487424))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167494400))))[name = string("const_1563_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1564_to_fp16 = const()[name = string("const_1564_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167502656)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_8_cast_fp16 = conv(bias = const_1564_to_fp16, dilations = conv1d_25_dilations_0, groups = conv1d_25_groups_0, pad = conv1d_25_pad_0, pad_type = conv1d_25_pad_type_0, strides = conv1d_25_strides_0, weight = const_1563_to_fp16_palettized, x = masked_fill_17_cast_fp16)[name = string("_native_batch_norm_legit_no_training_8_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_25_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_8_cast_fp16)[name = string("silu_25_cast_fp16")];
            string conv1d_26_pad_type_0 = const()[name = string("conv1d_26_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_26_strides_0 = const()[name = string("conv1d_26_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_26_pad_0 = const()[name = string("conv1d_26_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_26_dilations_0 = const()[name = string("conv1d_26_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_26_groups_0 = const()[name = string("conv1d_26_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_8_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167504768))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168291264))))[name = string("p_encoder_layers_8_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_26_cast_fp16 = conv(dilations = conv1d_26_dilations_0, groups = conv1d_26_groups_0, pad = conv1d_26_pad_0, pad_type = conv1d_26_pad_type_0, strides = conv1d_26_strides_0, weight = p_encoder_layers_8_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_25_cast_fp16)[name = string("conv1d_26_cast_fp16")];
            tensor<int32, [3]> transpose_56_perm_0 = const()[name = string("transpose_56_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_56_cast_fp16 = transpose(perm = transpose_56_perm_0, x = conv1d_26_cast_fp16)[name = string("transpose_90")];
            tensor<fp16, [1, 188, 1024]> add_67_cast_fp16 = add(x = add_66_cast_fp16, y = transpose_56_cast_fp16)[name = string("add_67_cast_fp16")];
            tensor<int32, [1]> layer_norm_43_axes_0 = const()[name = string("layer_norm_43_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_8_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_8_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168299520)))];
            tensor<fp16, [1024]> p_encoder_layers_8_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_8_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168301632)))];
            fp16 const_640_to_fp16 = const()[name = string("const_640_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_43_cast_fp16 = layer_norm(axes = layer_norm_43_axes_0, beta = p_encoder_layers_8_norm_feed_forward2_bias_to_fp16, epsilon = const_640_to_fp16, gamma = p_encoder_layers_8_norm_feed_forward2_weight_to_fp16, x = add_67_cast_fp16)[name = string("layer_norm_43_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_8_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168303744))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171449536))))[name = string("p_encoder_layers_8_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_80_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_8_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_43_cast_fp16)[name = string("linear_80_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_26_cast_fp16 = silu(x = linear_80_cast_fp16)[name = string("silu_26_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_8_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171482368))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174628160))))[name = string("p_encoder_layers_8_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_81_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_8_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_26_cast_fp16)[name = string("linear_81_cast_fp16")];
            fp16 const_642_to_fp16 = const()[name = string("const_642_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_32_cast_fp16 = mul(x = linear_81_cast_fp16, y = const_642_to_fp16)[name = string("mul_32_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_68_cast_fp16 = add(x = add_67_cast_fp16, y = mul_32_cast_fp16)[name = string("add_68_cast_fp16")];
            tensor<int32, [1]> layer_norm_44_axes_0 = const()[name = string("layer_norm_44_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_8_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_8_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174636416)))];
            tensor<fp16, [1024]> p_encoder_layers_8_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_8_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174638528)))];
            fp16 const_644_to_fp16 = const()[name = string("const_644_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_44_cast_fp16 = layer_norm(axes = layer_norm_44_axes_0, beta = p_encoder_layers_8_norm_out_bias_to_fp16, epsilon = const_644_to_fp16, gamma = p_encoder_layers_8_norm_out_weight_to_fp16, x = add_68_cast_fp16)[name = string("layer_norm_44_cast_fp16")];
            tensor<int32, [1]> layer_norm_45_axes_0 = const()[name = string("layer_norm_45_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_9_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_9_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174640640)))];
            tensor<fp16, [1024]> p_encoder_layers_9_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_9_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174642752)))];
            fp16 const_647_to_fp16 = const()[name = string("const_647_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_45_cast_fp16 = layer_norm(axes = layer_norm_45_axes_0, beta = p_encoder_layers_9_norm_feed_forward1_bias_to_fp16, epsilon = const_647_to_fp16, gamma = p_encoder_layers_9_norm_feed_forward1_weight_to_fp16, x = layer_norm_44_cast_fp16)[name = string("layer_norm_45_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_9_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174644864))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177790656))))[name = string("p_encoder_layers_9_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_82_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_9_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_45_cast_fp16)[name = string("linear_82_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_27_cast_fp16 = silu(x = linear_82_cast_fp16)[name = string("silu_27_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_9_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177823488))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180969280))))[name = string("p_encoder_layers_9_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_83_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_9_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_27_cast_fp16)[name = string("linear_83_cast_fp16")];
            fp16 const_649_to_fp16 = const()[name = string("const_649_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_33_cast_fp16 = mul(x = linear_83_cast_fp16, y = const_649_to_fp16)[name = string("mul_33_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_69_cast_fp16 = add(x = layer_norm_44_cast_fp16, y = mul_33_cast_fp16)[name = string("add_69_cast_fp16")];
            tensor<int32, [1]> layer_norm_46_axes_0 = const()[name = string("layer_norm_46_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_9_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_9_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180977536)))];
            tensor<fp16, [1024]> p_encoder_layers_9_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_9_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180979648)))];
            fp16 const_651_to_fp16 = const()[name = string("const_651_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_46_cast_fp16 = layer_norm(axes = layer_norm_46_axes_0, beta = p_encoder_layers_9_norm_self_att_bias_to_fp16, epsilon = const_651_to_fp16, gamma = p_encoder_layers_9_norm_self_att_weight_to_fp16, x = add_69_cast_fp16)[name = string("layer_norm_46_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180981760))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181768256))))[name = string("p_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_84_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_46_cast_fp16)[name = string("linear_84_cast_fp16")];
            tensor<int32, [4]> const_653 = const()[name = string("const_653"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_82_cast_fp16 = reshape(shape = const_653, x = linear_84_cast_fp16)[name = string("view_82_cast_fp16")];
            tensor<int32, [4]> transpose_57_perm_0 = const()[name = string("transpose_57_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181776512))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182563008))))[name = string("p_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_85_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_46_cast_fp16)[name = string("linear_85_cast_fp16")];
            tensor<int32, [4]> const_656 = const()[name = string("const_656"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_83_cast_fp16 = reshape(shape = const_656, x = linear_85_cast_fp16)[name = string("view_83_cast_fp16")];
            tensor<int32, [4]> transpose_58_perm_0 = const()[name = string("transpose_58_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182571264))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183357760))))[name = string("p_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_86_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_46_cast_fp16)[name = string("linear_86_cast_fp16")];
            tensor<int32, [4]> const_659 = const()[name = string("const_659"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_84_cast_fp16 = reshape(shape = const_659, x = linear_86_cast_fp16)[name = string("view_84_cast_fp16")];
            tensor<int32, [4]> transpose_59_perm_0 = const()[name = string("transpose_59_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_85_to_fp16 = const()[name = string("view_85_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183366016)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_57_cast_fp16 = transpose(perm = transpose_57_perm_0, x = view_82_cast_fp16)[name = string("transpose_89")];
            tensor<fp16, [1, 8, 188, 128]> add_70_cast_fp16 = add(x = transpose_57_cast_fp16, y = view_85_to_fp16)[name = string("add_70_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_86_to_fp16 = const()[name = string("view_86_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183368128)))];
            tensor<fp16, [1, 8, 188, 128]> add_71_cast_fp16 = add(x = transpose_57_cast_fp16, y = view_86_to_fp16)[name = string("add_71_cast_fp16")];
            bool matmul_10_transpose_x_0 = const()[name = string("matmul_10_transpose_x_0"), val = bool(false)];
            bool matmul_10_transpose_y_0 = const()[name = string("matmul_10_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_9_to_fp16 = const()[name = string("permute_9_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183370240)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_10_cast_fp16 = matmul(transpose_x = matmul_10_transpose_x_0, transpose_y = matmul_10_transpose_y_0, x = add_71_cast_fp16, y = permute_9_to_fp16)[name = string("matmul_10_cast_fp16")];
            tensor<int32, [8]> pad_9_pad_0 = const()[name = string("pad_9_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_9_mode_0 = const()[name = string("pad_9_mode_0"), val = string("constant")];
            fp16 const_667_to_fp16 = const()[name = string("const_667_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_9_cast_fp16 = pad(constant_val = const_667_to_fp16, mode = pad_9_mode_0, pad = pad_9_pad_0, x = matmul_10_cast_fp16)[name = string("pad_9_cast_fp16")];
            tensor<int32, [4]> const_668 = const()[name = string("const_668"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_88_cast_fp16 = reshape(shape = const_668, x = pad_9_cast_fp16)[name = string("view_88_cast_fp16")];
            tensor<int32, [4]> slice_19_begin_0 = const()[name = string("slice_19_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_19_end_0 = const()[name = string("slice_19_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_19_end_mask_0 = const()[name = string("slice_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_19_cast_fp16 = slice_by_index(begin = slice_19_begin_0, end = slice_19_end_0, end_mask = slice_19_end_mask_0, x = view_88_cast_fp16)[name = string("slice_19_cast_fp16")];
            tensor<int32, [4]> const_672 = const()[name = string("const_672"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_89_cast_fp16 = reshape(shape = const_672, x = slice_19_cast_fp16)[name = string("view_89_cast_fp16")];
            tensor<int32, [4]> slice_20_begin_0 = const()[name = string("slice_20_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_20_end_0 = const()[name = string("slice_20_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_20_end_mask_0 = const()[name = string("slice_20_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_20_cast_fp16 = slice_by_index(begin = slice_20_begin_0, end = slice_20_end_0, end_mask = slice_20_end_mask_0, x = view_89_cast_fp16)[name = string("slice_20_cast_fp16")];
            fp16 const_676_to_fp16 = const()[name = string("const_676_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_34_cast_fp16 = mul(x = slice_20_cast_fp16, y = const_676_to_fp16)[name = string("mul_34_cast_fp16")];
            fp16 const_677_to_fp16 = const()[name = string("const_677_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_18_cast_fp16 = select(a = const_677_to_fp16, b = mul_34_cast_fp16, cond = logical_not)[name = string("masked_fill_18_cast_fp16")];
            fp16 const_678_to_fp16 = const()[name = string("const_678_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_9_1_cast_fp16 = mul(x = add_70_cast_fp16, y = const_678_to_fp16)[name = string("mul_9_1_cast_fp16")];
            bool matmul_9_transpose_y_1 = const()[name = string("matmul_9_transpose_y_1"), val = bool(true)];
            bool matmul_9_transpose_x_1 = const()[name = string("matmul_9_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_58_cast_fp16 = transpose(perm = transpose_58_perm_0, x = view_83_cast_fp16)[name = string("transpose_88")];
            tensor<fp16, [1, 8, 188, 188]> matmul_9_1_cast_fp16 = matmul(transpose_x = matmul_9_transpose_x_1, transpose_y = matmul_9_transpose_y_1, x = mul_9_1_cast_fp16, y = transpose_58_cast_fp16)[name = string("matmul_9_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_9_1_cast_fp16 = add(x = matmul_9_1_cast_fp16, y = masked_fill_18_cast_fp16)[name = string("add_9_1_cast_fp16")];
            int32 softmax_9_axis_0 = const()[name = string("softmax_9_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_9_cast_fp16 = softmax(axis = softmax_9_axis_0, x = add_9_1_cast_fp16)[name = string("softmax_9_cast_fp16")];
            bool scaled_dot_product_attention_9_transpose_x_0 = const()[name = string("scaled_dot_product_attention_9_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_9_transpose_y_0 = const()[name = string("scaled_dot_product_attention_9_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_59_cast_fp16 = transpose(perm = transpose_59_perm_0, x = view_84_cast_fp16)[name = string("transpose_87")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_9_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_9_transpose_x_0, transpose_y = scaled_dot_product_attention_9_transpose_y_0, x = softmax_9_cast_fp16, y = transpose_59_cast_fp16)[name = string("scaled_dot_product_attention_9_cast_fp16")];
            tensor<int32, [4]> transpose_60_perm_0 = const()[name = string("transpose_60_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_681 = const()[name = string("const_681"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_60_cast_fp16 = transpose(perm = transpose_60_perm_0, x = scaled_dot_product_attention_9_cast_fp16)[name = string("transpose_86")];
            tensor<fp16, [1, 188, 1024]> view_90_cast_fp16 = reshape(shape = const_681, x = transpose_60_cast_fp16)[name = string("view_90_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_9_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184138304))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184924800))))[name = string("p_encoder_layers_9_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_88_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_9_self_attn_o_proj_weight_to_fp16_palettized, x = view_90_cast_fp16)[name = string("linear_88_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_72_cast_fp16 = add(x = add_69_cast_fp16, y = linear_88_cast_fp16)[name = string("add_72_cast_fp16")];
            tensor<int32, [1]> layer_norm_47_axes_0 = const()[name = string("layer_norm_47_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_9_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_9_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184933056)))];
            tensor<fp16, [1024]> p_encoder_layers_9_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_9_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184935168)))];
            fp16 const_683_to_fp16 = const()[name = string("const_683_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_47_cast_fp16 = layer_norm(axes = layer_norm_47_axes_0, beta = p_encoder_layers_9_norm_conv_bias_to_fp16, epsilon = const_683_to_fp16, gamma = p_encoder_layers_9_norm_conv_weight_to_fp16, x = add_72_cast_fp16)[name = string("layer_norm_47_cast_fp16")];
            tensor<int32, [3]> transpose_61_perm_0 = const()[name = string("transpose_61_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_27_pad_type_0 = const()[name = string("conv1d_27_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_27_strides_0 = const()[name = string("conv1d_27_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_27_pad_0 = const()[name = string("conv1d_27_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_27_dilations_0 = const()[name = string("conv1d_27_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_27_groups_0 = const()[name = string("conv1d_27_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_9_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184937280))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186510208))))[name = string("p_encoder_layers_9_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_61_cast_fp16 = transpose(perm = transpose_61_perm_0, x = layer_norm_47_cast_fp16)[name = string("transpose_85")];
            tensor<fp16, [1, 2048, 188]> conv1d_27_cast_fp16 = conv(dilations = conv1d_27_dilations_0, groups = conv1d_27_groups_0, pad = conv1d_27_pad_0, pad_type = conv1d_27_pad_type_0, strides = conv1d_27_strides_0, weight = p_encoder_layers_9_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_61_cast_fp16)[name = string("conv1d_27_cast_fp16")];
            int32 glu_9_split_num_splits_0 = const()[name = string("glu_9_split_num_splits_0"), val = int32(2)];
            int32 glu_9_split_axis_0 = const()[name = string("glu_9_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_9_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_9_split_cast_fp16_1 = split(axis = glu_9_split_axis_0, num_splits = glu_9_split_num_splits_0, x = conv1d_27_cast_fp16)[name = string("glu_9_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_9_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_9_split_cast_fp16_1)[name = string("glu_9_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_9_cast_fp16 = mul(x = glu_9_split_cast_fp16_0, y = glu_9_split_1_sigmoid_cast_fp16)[name = string("glu_9_cast_fp16")];
            fp16 const_689_to_fp16 = const()[name = string("const_689_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_19_cast_fp16 = select(a = const_689_to_fp16, b = glu_9_cast_fp16, cond = all_1)[name = string("masked_fill_19_cast_fp16")];
            string conv1d_28_pad_type_0 = const()[name = string("conv1d_28_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_28_pad_0 = const()[name = string("conv1d_28_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_28_groups_0 = const()[name = string("conv1d_28_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_28_strides_0 = const()[name = string("conv1d_28_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_28_dilations_0 = const()[name = string("conv1d_28_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1565_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186526656))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186533632))))[name = string("const_1565_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1566_to_fp16 = const()[name = string("const_1566_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186541888)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_9_cast_fp16 = conv(bias = const_1566_to_fp16, dilations = conv1d_28_dilations_0, groups = conv1d_28_groups_0, pad = conv1d_28_pad_0, pad_type = conv1d_28_pad_type_0, strides = conv1d_28_strides_0, weight = const_1565_to_fp16_palettized, x = masked_fill_19_cast_fp16)[name = string("_native_batch_norm_legit_no_training_9_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_28_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_9_cast_fp16)[name = string("silu_28_cast_fp16")];
            string conv1d_29_pad_type_0 = const()[name = string("conv1d_29_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_29_strides_0 = const()[name = string("conv1d_29_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_29_pad_0 = const()[name = string("conv1d_29_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_29_dilations_0 = const()[name = string("conv1d_29_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_29_groups_0 = const()[name = string("conv1d_29_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_9_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186544000))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187330496))))[name = string("p_encoder_layers_9_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_29_cast_fp16 = conv(dilations = conv1d_29_dilations_0, groups = conv1d_29_groups_0, pad = conv1d_29_pad_0, pad_type = conv1d_29_pad_type_0, strides = conv1d_29_strides_0, weight = p_encoder_layers_9_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_28_cast_fp16)[name = string("conv1d_29_cast_fp16")];
            tensor<int32, [3]> transpose_62_perm_0 = const()[name = string("transpose_62_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_62_cast_fp16 = transpose(perm = transpose_62_perm_0, x = conv1d_29_cast_fp16)[name = string("transpose_84")];
            tensor<fp16, [1, 188, 1024]> add_73_cast_fp16 = add(x = add_72_cast_fp16, y = transpose_62_cast_fp16)[name = string("add_73_cast_fp16")];
            tensor<int32, [1]> layer_norm_48_axes_0 = const()[name = string("layer_norm_48_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_9_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_9_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187338752)))];
            tensor<fp16, [1024]> p_encoder_layers_9_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_9_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187340864)))];
            fp16 const_700_to_fp16 = const()[name = string("const_700_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_48_cast_fp16 = layer_norm(axes = layer_norm_48_axes_0, beta = p_encoder_layers_9_norm_feed_forward2_bias_to_fp16, epsilon = const_700_to_fp16, gamma = p_encoder_layers_9_norm_feed_forward2_weight_to_fp16, x = add_73_cast_fp16)[name = string("layer_norm_48_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_9_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187342976))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190488768))))[name = string("p_encoder_layers_9_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_89_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_9_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_48_cast_fp16)[name = string("linear_89_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_29_cast_fp16 = silu(x = linear_89_cast_fp16)[name = string("silu_29_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_9_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190521600))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193667392))))[name = string("p_encoder_layers_9_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_90_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_9_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_29_cast_fp16)[name = string("linear_90_cast_fp16")];
            fp16 const_702_to_fp16 = const()[name = string("const_702_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_35_cast_fp16 = mul(x = linear_90_cast_fp16, y = const_702_to_fp16)[name = string("mul_35_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_74_cast_fp16 = add(x = add_73_cast_fp16, y = mul_35_cast_fp16)[name = string("add_74_cast_fp16")];
            tensor<int32, [1]> layer_norm_49_axes_0 = const()[name = string("layer_norm_49_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_9_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_9_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193675648)))];
            tensor<fp16, [1024]> p_encoder_layers_9_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_9_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193677760)))];
            fp16 const_704_to_fp16 = const()[name = string("const_704_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_49_cast_fp16 = layer_norm(axes = layer_norm_49_axes_0, beta = p_encoder_layers_9_norm_out_bias_to_fp16, epsilon = const_704_to_fp16, gamma = p_encoder_layers_9_norm_out_weight_to_fp16, x = add_74_cast_fp16)[name = string("layer_norm_49_cast_fp16")];
            tensor<int32, [1]> layer_norm_50_axes_0 = const()[name = string("layer_norm_50_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_10_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_10_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193679872)))];
            tensor<fp16, [1024]> p_encoder_layers_10_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_10_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193681984)))];
            fp16 const_707_to_fp16 = const()[name = string("const_707_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_50_cast_fp16 = layer_norm(axes = layer_norm_50_axes_0, beta = p_encoder_layers_10_norm_feed_forward1_bias_to_fp16, epsilon = const_707_to_fp16, gamma = p_encoder_layers_10_norm_feed_forward1_weight_to_fp16, x = layer_norm_49_cast_fp16)[name = string("layer_norm_50_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_10_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193684096))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196829888))))[name = string("p_encoder_layers_10_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_91_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_10_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_50_cast_fp16)[name = string("linear_91_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_30_cast_fp16 = silu(x = linear_91_cast_fp16)[name = string("silu_30_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_10_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196862720))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200008512))))[name = string("p_encoder_layers_10_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_92_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_10_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_30_cast_fp16)[name = string("linear_92_cast_fp16")];
            fp16 const_709_to_fp16 = const()[name = string("const_709_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_36_cast_fp16 = mul(x = linear_92_cast_fp16, y = const_709_to_fp16)[name = string("mul_36_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_75_cast_fp16 = add(x = layer_norm_49_cast_fp16, y = mul_36_cast_fp16)[name = string("add_75_cast_fp16")];
            tensor<int32, [1]> layer_norm_51_axes_0 = const()[name = string("layer_norm_51_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_10_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_10_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200016768)))];
            tensor<fp16, [1024]> p_encoder_layers_10_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_10_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200018880)))];
            fp16 const_711_to_fp16 = const()[name = string("const_711_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_51_cast_fp16 = layer_norm(axes = layer_norm_51_axes_0, beta = p_encoder_layers_10_norm_self_att_bias_to_fp16, epsilon = const_711_to_fp16, gamma = p_encoder_layers_10_norm_self_att_weight_to_fp16, x = add_75_cast_fp16)[name = string("layer_norm_51_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200020992))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200807488))))[name = string("p_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_93_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_51_cast_fp16)[name = string("linear_93_cast_fp16")];
            tensor<int32, [4]> const_713 = const()[name = string("const_713"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_91_cast_fp16 = reshape(shape = const_713, x = linear_93_cast_fp16)[name = string("view_91_cast_fp16")];
            tensor<int32, [4]> transpose_63_perm_0 = const()[name = string("transpose_63_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200815744))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201602240))))[name = string("p_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_94_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_51_cast_fp16)[name = string("linear_94_cast_fp16")];
            tensor<int32, [4]> const_716 = const()[name = string("const_716"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_92_cast_fp16 = reshape(shape = const_716, x = linear_94_cast_fp16)[name = string("view_92_cast_fp16")];
            tensor<int32, [4]> transpose_64_perm_0 = const()[name = string("transpose_64_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201610496))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202396992))))[name = string("p_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_95_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_51_cast_fp16)[name = string("linear_95_cast_fp16")];
            tensor<int32, [4]> const_719 = const()[name = string("const_719"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_93_cast_fp16 = reshape(shape = const_719, x = linear_95_cast_fp16)[name = string("view_93_cast_fp16")];
            tensor<int32, [4]> transpose_65_perm_0 = const()[name = string("transpose_65_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_94_to_fp16 = const()[name = string("view_94_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202405248)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_63_cast_fp16 = transpose(perm = transpose_63_perm_0, x = view_91_cast_fp16)[name = string("transpose_83")];
            tensor<fp16, [1, 8, 188, 128]> add_76_cast_fp16 = add(x = transpose_63_cast_fp16, y = view_94_to_fp16)[name = string("add_76_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_95_to_fp16 = const()[name = string("view_95_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202407360)))];
            tensor<fp16, [1, 8, 188, 128]> add_77_cast_fp16 = add(x = transpose_63_cast_fp16, y = view_95_to_fp16)[name = string("add_77_cast_fp16")];
            bool matmul_11_transpose_x_0 = const()[name = string("matmul_11_transpose_x_0"), val = bool(false)];
            bool matmul_11_transpose_y_0 = const()[name = string("matmul_11_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_10_to_fp16 = const()[name = string("permute_10_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202409472)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_11_cast_fp16 = matmul(transpose_x = matmul_11_transpose_x_0, transpose_y = matmul_11_transpose_y_0, x = add_77_cast_fp16, y = permute_10_to_fp16)[name = string("matmul_11_cast_fp16")];
            tensor<int32, [8]> pad_10_pad_0 = const()[name = string("pad_10_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_10_mode_0 = const()[name = string("pad_10_mode_0"), val = string("constant")];
            fp16 const_727_to_fp16 = const()[name = string("const_727_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_10_cast_fp16 = pad(constant_val = const_727_to_fp16, mode = pad_10_mode_0, pad = pad_10_pad_0, x = matmul_11_cast_fp16)[name = string("pad_10_cast_fp16")];
            tensor<int32, [4]> const_728 = const()[name = string("const_728"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_97_cast_fp16 = reshape(shape = const_728, x = pad_10_cast_fp16)[name = string("view_97_cast_fp16")];
            tensor<int32, [4]> slice_21_begin_0 = const()[name = string("slice_21_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_21_end_0 = const()[name = string("slice_21_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_21_end_mask_0 = const()[name = string("slice_21_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_21_cast_fp16 = slice_by_index(begin = slice_21_begin_0, end = slice_21_end_0, end_mask = slice_21_end_mask_0, x = view_97_cast_fp16)[name = string("slice_21_cast_fp16")];
            tensor<int32, [4]> const_732 = const()[name = string("const_732"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_98_cast_fp16 = reshape(shape = const_732, x = slice_21_cast_fp16)[name = string("view_98_cast_fp16")];
            tensor<int32, [4]> slice_22_begin_0 = const()[name = string("slice_22_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_22_end_0 = const()[name = string("slice_22_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_22_end_mask_0 = const()[name = string("slice_22_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_22_cast_fp16 = slice_by_index(begin = slice_22_begin_0, end = slice_22_end_0, end_mask = slice_22_end_mask_0, x = view_98_cast_fp16)[name = string("slice_22_cast_fp16")];
            fp16 const_736_to_fp16 = const()[name = string("const_736_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_37_cast_fp16 = mul(x = slice_22_cast_fp16, y = const_736_to_fp16)[name = string("mul_37_cast_fp16")];
            fp16 const_737_to_fp16 = const()[name = string("const_737_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_20_cast_fp16 = select(a = const_737_to_fp16, b = mul_37_cast_fp16, cond = logical_not)[name = string("masked_fill_20_cast_fp16")];
            fp16 const_738_to_fp16 = const()[name = string("const_738_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_10_1_cast_fp16 = mul(x = add_76_cast_fp16, y = const_738_to_fp16)[name = string("mul_10_1_cast_fp16")];
            bool matmul_10_transpose_y_1 = const()[name = string("matmul_10_transpose_y_1"), val = bool(true)];
            bool matmul_10_transpose_x_1 = const()[name = string("matmul_10_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_64_cast_fp16 = transpose(perm = transpose_64_perm_0, x = view_92_cast_fp16)[name = string("transpose_82")];
            tensor<fp16, [1, 8, 188, 188]> matmul_10_1_cast_fp16 = matmul(transpose_x = matmul_10_transpose_x_1, transpose_y = matmul_10_transpose_y_1, x = mul_10_1_cast_fp16, y = transpose_64_cast_fp16)[name = string("matmul_10_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_10_1_cast_fp16 = add(x = matmul_10_1_cast_fp16, y = masked_fill_20_cast_fp16)[name = string("add_10_1_cast_fp16")];
            int32 softmax_10_axis_0 = const()[name = string("softmax_10_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_10_cast_fp16 = softmax(axis = softmax_10_axis_0, x = add_10_1_cast_fp16)[name = string("softmax_10_cast_fp16")];
            bool scaled_dot_product_attention_10_transpose_x_0 = const()[name = string("scaled_dot_product_attention_10_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_10_transpose_y_0 = const()[name = string("scaled_dot_product_attention_10_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_65_cast_fp16 = transpose(perm = transpose_65_perm_0, x = view_93_cast_fp16)[name = string("transpose_81")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_10_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_10_transpose_x_0, transpose_y = scaled_dot_product_attention_10_transpose_y_0, x = softmax_10_cast_fp16, y = transpose_65_cast_fp16)[name = string("scaled_dot_product_attention_10_cast_fp16")];
            tensor<int32, [4]> transpose_66_perm_0 = const()[name = string("transpose_66_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_741 = const()[name = string("const_741"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_66_cast_fp16 = transpose(perm = transpose_66_perm_0, x = scaled_dot_product_attention_10_cast_fp16)[name = string("transpose_80")];
            tensor<fp16, [1, 188, 1024]> view_99_cast_fp16 = reshape(shape = const_741, x = transpose_66_cast_fp16)[name = string("view_99_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_10_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203177536))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203964032))))[name = string("p_encoder_layers_10_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_97_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_10_self_attn_o_proj_weight_to_fp16_palettized, x = view_99_cast_fp16)[name = string("linear_97_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_78_cast_fp16 = add(x = add_75_cast_fp16, y = linear_97_cast_fp16)[name = string("add_78_cast_fp16")];
            tensor<int32, [1]> layer_norm_52_axes_0 = const()[name = string("layer_norm_52_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_10_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_10_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203972288)))];
            tensor<fp16, [1024]> p_encoder_layers_10_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_10_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203974400)))];
            fp16 const_743_to_fp16 = const()[name = string("const_743_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_52_cast_fp16 = layer_norm(axes = layer_norm_52_axes_0, beta = p_encoder_layers_10_norm_conv_bias_to_fp16, epsilon = const_743_to_fp16, gamma = p_encoder_layers_10_norm_conv_weight_to_fp16, x = add_78_cast_fp16)[name = string("layer_norm_52_cast_fp16")];
            tensor<int32, [3]> transpose_67_perm_0 = const()[name = string("transpose_67_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_30_pad_type_0 = const()[name = string("conv1d_30_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_30_strides_0 = const()[name = string("conv1d_30_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_30_pad_0 = const()[name = string("conv1d_30_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_30_dilations_0 = const()[name = string("conv1d_30_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_30_groups_0 = const()[name = string("conv1d_30_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_10_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203976512))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205549440))))[name = string("p_encoder_layers_10_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_67_cast_fp16 = transpose(perm = transpose_67_perm_0, x = layer_norm_52_cast_fp16)[name = string("transpose_79")];
            tensor<fp16, [1, 2048, 188]> conv1d_30_cast_fp16 = conv(dilations = conv1d_30_dilations_0, groups = conv1d_30_groups_0, pad = conv1d_30_pad_0, pad_type = conv1d_30_pad_type_0, strides = conv1d_30_strides_0, weight = p_encoder_layers_10_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_67_cast_fp16)[name = string("conv1d_30_cast_fp16")];
            int32 glu_10_split_num_splits_0 = const()[name = string("glu_10_split_num_splits_0"), val = int32(2)];
            int32 glu_10_split_axis_0 = const()[name = string("glu_10_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_10_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_10_split_cast_fp16_1 = split(axis = glu_10_split_axis_0, num_splits = glu_10_split_num_splits_0, x = conv1d_30_cast_fp16)[name = string("glu_10_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_10_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_10_split_cast_fp16_1)[name = string("glu_10_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_10_cast_fp16 = mul(x = glu_10_split_cast_fp16_0, y = glu_10_split_1_sigmoid_cast_fp16)[name = string("glu_10_cast_fp16")];
            fp16 const_749_to_fp16 = const()[name = string("const_749_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_21_cast_fp16 = select(a = const_749_to_fp16, b = glu_10_cast_fp16, cond = all_1)[name = string("masked_fill_21_cast_fp16")];
            string conv1d_31_pad_type_0 = const()[name = string("conv1d_31_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_31_pad_0 = const()[name = string("conv1d_31_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_31_groups_0 = const()[name = string("conv1d_31_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_31_strides_0 = const()[name = string("conv1d_31_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_31_dilations_0 = const()[name = string("conv1d_31_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1567_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205565888))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205572864))))[name = string("const_1567_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1568_to_fp16 = const()[name = string("const_1568_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205581120)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_10_cast_fp16 = conv(bias = const_1568_to_fp16, dilations = conv1d_31_dilations_0, groups = conv1d_31_groups_0, pad = conv1d_31_pad_0, pad_type = conv1d_31_pad_type_0, strides = conv1d_31_strides_0, weight = const_1567_to_fp16_palettized, x = masked_fill_21_cast_fp16)[name = string("_native_batch_norm_legit_no_training_10_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_31_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_10_cast_fp16)[name = string("silu_31_cast_fp16")];
            string conv1d_32_pad_type_0 = const()[name = string("conv1d_32_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_32_strides_0 = const()[name = string("conv1d_32_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_32_pad_0 = const()[name = string("conv1d_32_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_32_dilations_0 = const()[name = string("conv1d_32_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_32_groups_0 = const()[name = string("conv1d_32_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_10_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205583232))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206369728))))[name = string("p_encoder_layers_10_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_32_cast_fp16 = conv(dilations = conv1d_32_dilations_0, groups = conv1d_32_groups_0, pad = conv1d_32_pad_0, pad_type = conv1d_32_pad_type_0, strides = conv1d_32_strides_0, weight = p_encoder_layers_10_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_31_cast_fp16)[name = string("conv1d_32_cast_fp16")];
            tensor<int32, [3]> transpose_68_perm_0 = const()[name = string("transpose_68_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_68_cast_fp16 = transpose(perm = transpose_68_perm_0, x = conv1d_32_cast_fp16)[name = string("transpose_78")];
            tensor<fp16, [1, 188, 1024]> add_79_cast_fp16 = add(x = add_78_cast_fp16, y = transpose_68_cast_fp16)[name = string("add_79_cast_fp16")];
            tensor<int32, [1]> layer_norm_53_axes_0 = const()[name = string("layer_norm_53_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_10_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_10_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206377984)))];
            tensor<fp16, [1024]> p_encoder_layers_10_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_10_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206380096)))];
            fp16 const_760_to_fp16 = const()[name = string("const_760_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_53_cast_fp16 = layer_norm(axes = layer_norm_53_axes_0, beta = p_encoder_layers_10_norm_feed_forward2_bias_to_fp16, epsilon = const_760_to_fp16, gamma = p_encoder_layers_10_norm_feed_forward2_weight_to_fp16, x = add_79_cast_fp16)[name = string("layer_norm_53_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_10_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206382208))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209528000))))[name = string("p_encoder_layers_10_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_98_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_10_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_53_cast_fp16)[name = string("linear_98_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_32_cast_fp16 = silu(x = linear_98_cast_fp16)[name = string("silu_32_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_10_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209560832))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212706624))))[name = string("p_encoder_layers_10_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_99_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_10_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_32_cast_fp16)[name = string("linear_99_cast_fp16")];
            fp16 const_762_to_fp16 = const()[name = string("const_762_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_38_cast_fp16 = mul(x = linear_99_cast_fp16, y = const_762_to_fp16)[name = string("mul_38_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_80_cast_fp16 = add(x = add_79_cast_fp16, y = mul_38_cast_fp16)[name = string("add_80_cast_fp16")];
            tensor<int32, [1]> layer_norm_54_axes_0 = const()[name = string("layer_norm_54_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_10_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_10_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212714880)))];
            tensor<fp16, [1024]> p_encoder_layers_10_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_10_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212716992)))];
            fp16 const_764_to_fp16 = const()[name = string("const_764_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_54_cast_fp16 = layer_norm(axes = layer_norm_54_axes_0, beta = p_encoder_layers_10_norm_out_bias_to_fp16, epsilon = const_764_to_fp16, gamma = p_encoder_layers_10_norm_out_weight_to_fp16, x = add_80_cast_fp16)[name = string("layer_norm_54_cast_fp16")];
            tensor<int32, [1]> layer_norm_55_axes_0 = const()[name = string("layer_norm_55_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_11_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_11_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212719104)))];
            tensor<fp16, [1024]> p_encoder_layers_11_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_11_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212721216)))];
            fp16 const_767_to_fp16 = const()[name = string("const_767_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_55_cast_fp16 = layer_norm(axes = layer_norm_55_axes_0, beta = p_encoder_layers_11_norm_feed_forward1_bias_to_fp16, epsilon = const_767_to_fp16, gamma = p_encoder_layers_11_norm_feed_forward1_weight_to_fp16, x = layer_norm_54_cast_fp16)[name = string("layer_norm_55_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_11_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212723328))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215869120))))[name = string("p_encoder_layers_11_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_100_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_11_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_55_cast_fp16)[name = string("linear_100_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_33_cast_fp16 = silu(x = linear_100_cast_fp16)[name = string("silu_33_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_11_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215901952))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219047744))))[name = string("p_encoder_layers_11_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_101_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_11_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_33_cast_fp16)[name = string("linear_101_cast_fp16")];
            fp16 const_769_to_fp16 = const()[name = string("const_769_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_39_cast_fp16 = mul(x = linear_101_cast_fp16, y = const_769_to_fp16)[name = string("mul_39_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_81_cast_fp16 = add(x = layer_norm_54_cast_fp16, y = mul_39_cast_fp16)[name = string("add_81_cast_fp16")];
            tensor<int32, [1]> layer_norm_56_axes_0 = const()[name = string("layer_norm_56_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_11_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_11_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219056000)))];
            tensor<fp16, [1024]> p_encoder_layers_11_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_11_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219058112)))];
            fp16 const_771_to_fp16 = const()[name = string("const_771_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_56_cast_fp16 = layer_norm(axes = layer_norm_56_axes_0, beta = p_encoder_layers_11_norm_self_att_bias_to_fp16, epsilon = const_771_to_fp16, gamma = p_encoder_layers_11_norm_self_att_weight_to_fp16, x = add_81_cast_fp16)[name = string("layer_norm_56_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219060224))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219846720))))[name = string("p_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_102_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_56_cast_fp16)[name = string("linear_102_cast_fp16")];
            tensor<int32, [4]> const_773 = const()[name = string("const_773"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_100_cast_fp16 = reshape(shape = const_773, x = linear_102_cast_fp16)[name = string("view_100_cast_fp16")];
            tensor<int32, [4]> transpose_69_perm_0 = const()[name = string("transpose_69_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219854976))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220641472))))[name = string("p_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_103_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_56_cast_fp16)[name = string("linear_103_cast_fp16")];
            tensor<int32, [4]> const_776 = const()[name = string("const_776"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_101_cast_fp16 = reshape(shape = const_776, x = linear_103_cast_fp16)[name = string("view_101_cast_fp16")];
            tensor<int32, [4]> transpose_70_perm_0 = const()[name = string("transpose_70_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220649728))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221436224))))[name = string("p_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_104_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_56_cast_fp16)[name = string("linear_104_cast_fp16")];
            tensor<int32, [4]> const_779 = const()[name = string("const_779"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_102_cast_fp16 = reshape(shape = const_779, x = linear_104_cast_fp16)[name = string("view_102_cast_fp16")];
            tensor<int32, [4]> transpose_71_perm_0 = const()[name = string("transpose_71_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_103_to_fp16 = const()[name = string("view_103_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221444480)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_69_cast_fp16 = transpose(perm = transpose_69_perm_0, x = view_100_cast_fp16)[name = string("transpose_77")];
            tensor<fp16, [1, 8, 188, 128]> add_82_cast_fp16 = add(x = transpose_69_cast_fp16, y = view_103_to_fp16)[name = string("add_82_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_104_to_fp16 = const()[name = string("view_104_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221446592)))];
            tensor<fp16, [1, 8, 188, 128]> add_83_cast_fp16 = add(x = transpose_69_cast_fp16, y = view_104_to_fp16)[name = string("add_83_cast_fp16")];
            bool matmul_12_transpose_x_0 = const()[name = string("matmul_12_transpose_x_0"), val = bool(false)];
            bool matmul_12_transpose_y_0 = const()[name = string("matmul_12_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_11_to_fp16 = const()[name = string("permute_11_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221448704)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_12_cast_fp16 = matmul(transpose_x = matmul_12_transpose_x_0, transpose_y = matmul_12_transpose_y_0, x = add_83_cast_fp16, y = permute_11_to_fp16)[name = string("matmul_12_cast_fp16")];
            tensor<int32, [8]> pad_11_pad_0 = const()[name = string("pad_11_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_11_mode_0 = const()[name = string("pad_11_mode_0"), val = string("constant")];
            fp16 const_787_to_fp16 = const()[name = string("const_787_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_11_cast_fp16 = pad(constant_val = const_787_to_fp16, mode = pad_11_mode_0, pad = pad_11_pad_0, x = matmul_12_cast_fp16)[name = string("pad_11_cast_fp16")];
            tensor<int32, [4]> const_788 = const()[name = string("const_788"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_106_cast_fp16 = reshape(shape = const_788, x = pad_11_cast_fp16)[name = string("view_106_cast_fp16")];
            tensor<int32, [4]> slice_23_begin_0 = const()[name = string("slice_23_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_23_end_0 = const()[name = string("slice_23_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_23_end_mask_0 = const()[name = string("slice_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_23_cast_fp16 = slice_by_index(begin = slice_23_begin_0, end = slice_23_end_0, end_mask = slice_23_end_mask_0, x = view_106_cast_fp16)[name = string("slice_23_cast_fp16")];
            tensor<int32, [4]> const_792 = const()[name = string("const_792"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_107_cast_fp16 = reshape(shape = const_792, x = slice_23_cast_fp16)[name = string("view_107_cast_fp16")];
            tensor<int32, [4]> slice_24_begin_0 = const()[name = string("slice_24_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_24_end_0 = const()[name = string("slice_24_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_24_end_mask_0 = const()[name = string("slice_24_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_24_cast_fp16 = slice_by_index(begin = slice_24_begin_0, end = slice_24_end_0, end_mask = slice_24_end_mask_0, x = view_107_cast_fp16)[name = string("slice_24_cast_fp16")];
            fp16 const_796_to_fp16 = const()[name = string("const_796_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_40_cast_fp16 = mul(x = slice_24_cast_fp16, y = const_796_to_fp16)[name = string("mul_40_cast_fp16")];
            fp16 const_797_to_fp16 = const()[name = string("const_797_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_22_cast_fp16 = select(a = const_797_to_fp16, b = mul_40_cast_fp16, cond = logical_not)[name = string("masked_fill_22_cast_fp16")];
            fp16 const_798_to_fp16 = const()[name = string("const_798_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_11_1_cast_fp16 = mul(x = add_82_cast_fp16, y = const_798_to_fp16)[name = string("mul_11_1_cast_fp16")];
            bool matmul_11_transpose_y_1 = const()[name = string("matmul_11_transpose_y_1"), val = bool(true)];
            bool matmul_11_transpose_x_1 = const()[name = string("matmul_11_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_70_cast_fp16 = transpose(perm = transpose_70_perm_0, x = view_101_cast_fp16)[name = string("transpose_76")];
            tensor<fp16, [1, 8, 188, 188]> matmul_11_1_cast_fp16 = matmul(transpose_x = matmul_11_transpose_x_1, transpose_y = matmul_11_transpose_y_1, x = mul_11_1_cast_fp16, y = transpose_70_cast_fp16)[name = string("matmul_11_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_11_1_cast_fp16 = add(x = matmul_11_1_cast_fp16, y = masked_fill_22_cast_fp16)[name = string("add_11_1_cast_fp16")];
            int32 softmax_11_axis_0 = const()[name = string("softmax_11_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_11_cast_fp16 = softmax(axis = softmax_11_axis_0, x = add_11_1_cast_fp16)[name = string("softmax_11_cast_fp16")];
            bool scaled_dot_product_attention_11_transpose_x_0 = const()[name = string("scaled_dot_product_attention_11_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_11_transpose_y_0 = const()[name = string("scaled_dot_product_attention_11_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_71_cast_fp16 = transpose(perm = transpose_71_perm_0, x = view_102_cast_fp16)[name = string("transpose_75")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_11_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_11_transpose_x_0, transpose_y = scaled_dot_product_attention_11_transpose_y_0, x = softmax_11_cast_fp16, y = transpose_71_cast_fp16)[name = string("scaled_dot_product_attention_11_cast_fp16")];
            tensor<int32, [4]> transpose_72_perm_0 = const()[name = string("transpose_72_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_801 = const()[name = string("const_801"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_72_cast_fp16 = transpose(perm = transpose_72_perm_0, x = scaled_dot_product_attention_11_cast_fp16)[name = string("transpose_74")];
            tensor<fp16, [1, 188, 1024]> view_108_cast_fp16 = reshape(shape = const_801, x = transpose_72_cast_fp16)[name = string("view_108_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_11_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222216768))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223003264))))[name = string("p_encoder_layers_11_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_106_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_11_self_attn_o_proj_weight_to_fp16_palettized, x = view_108_cast_fp16)[name = string("linear_106_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_84_cast_fp16 = add(x = add_81_cast_fp16, y = linear_106_cast_fp16)[name = string("add_84_cast_fp16")];
            tensor<int32, [1]> layer_norm_57_axes_0 = const()[name = string("layer_norm_57_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_11_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_11_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223011520)))];
            tensor<fp16, [1024]> p_encoder_layers_11_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_11_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223013632)))];
            fp16 const_803_to_fp16 = const()[name = string("const_803_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_57_cast_fp16 = layer_norm(axes = layer_norm_57_axes_0, beta = p_encoder_layers_11_norm_conv_bias_to_fp16, epsilon = const_803_to_fp16, gamma = p_encoder_layers_11_norm_conv_weight_to_fp16, x = add_84_cast_fp16)[name = string("layer_norm_57_cast_fp16")];
            tensor<int32, [3]> transpose_73_perm_0 = const()[name = string("transpose_73_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_33_pad_type_0 = const()[name = string("conv1d_33_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_33_strides_0 = const()[name = string("conv1d_33_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_33_pad_0 = const()[name = string("conv1d_33_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_33_dilations_0 = const()[name = string("conv1d_33_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_33_groups_0 = const()[name = string("conv1d_33_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_11_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223015744))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224588672))))[name = string("p_encoder_layers_11_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_73_cast_fp16 = transpose(perm = transpose_73_perm_0, x = layer_norm_57_cast_fp16)[name = string("transpose_73")];
            tensor<fp16, [1, 2048, 188]> conv1d_33_cast_fp16 = conv(dilations = conv1d_33_dilations_0, groups = conv1d_33_groups_0, pad = conv1d_33_pad_0, pad_type = conv1d_33_pad_type_0, strides = conv1d_33_strides_0, weight = p_encoder_layers_11_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_73_cast_fp16)[name = string("conv1d_33_cast_fp16")];
            int32 glu_11_split_num_splits_0 = const()[name = string("glu_11_split_num_splits_0"), val = int32(2)];
            int32 glu_11_split_axis_0 = const()[name = string("glu_11_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_11_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_11_split_cast_fp16_1 = split(axis = glu_11_split_axis_0, num_splits = glu_11_split_num_splits_0, x = conv1d_33_cast_fp16)[name = string("glu_11_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_11_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_11_split_cast_fp16_1)[name = string("glu_11_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_11_cast_fp16 = mul(x = glu_11_split_cast_fp16_0, y = glu_11_split_1_sigmoid_cast_fp16)[name = string("glu_11_cast_fp16")];
            fp16 const_809_to_fp16 = const()[name = string("const_809_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_23_cast_fp16 = select(a = const_809_to_fp16, b = glu_11_cast_fp16, cond = all_1)[name = string("masked_fill_23_cast_fp16")];
            string conv1d_34_pad_type_0 = const()[name = string("conv1d_34_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_34_pad_0 = const()[name = string("conv1d_34_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_34_groups_0 = const()[name = string("conv1d_34_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_34_strides_0 = const()[name = string("conv1d_34_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_34_dilations_0 = const()[name = string("conv1d_34_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1569_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224605120))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224612096))))[name = string("const_1569_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1570_to_fp16 = const()[name = string("const_1570_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224620352)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_11_cast_fp16 = conv(bias = const_1570_to_fp16, dilations = conv1d_34_dilations_0, groups = conv1d_34_groups_0, pad = conv1d_34_pad_0, pad_type = conv1d_34_pad_type_0, strides = conv1d_34_strides_0, weight = const_1569_to_fp16_palettized, x = masked_fill_23_cast_fp16)[name = string("_native_batch_norm_legit_no_training_11_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_34_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_11_cast_fp16)[name = string("silu_34_cast_fp16")];
            string conv1d_35_pad_type_0 = const()[name = string("conv1d_35_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_35_strides_0 = const()[name = string("conv1d_35_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_35_pad_0 = const()[name = string("conv1d_35_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_35_dilations_0 = const()[name = string("conv1d_35_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_35_groups_0 = const()[name = string("conv1d_35_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_11_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224622464))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225408960))))[name = string("p_encoder_layers_11_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_35_cast_fp16 = conv(dilations = conv1d_35_dilations_0, groups = conv1d_35_groups_0, pad = conv1d_35_pad_0, pad_type = conv1d_35_pad_type_0, strides = conv1d_35_strides_0, weight = p_encoder_layers_11_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_34_cast_fp16)[name = string("conv1d_35_cast_fp16")];
            tensor<int32, [3]> transpose_74_perm_0 = const()[name = string("transpose_74_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_74_cast_fp16 = transpose(perm = transpose_74_perm_0, x = conv1d_35_cast_fp16)[name = string("transpose_72")];
            tensor<fp16, [1, 188, 1024]> add_85_cast_fp16 = add(x = add_84_cast_fp16, y = transpose_74_cast_fp16)[name = string("add_85_cast_fp16")];
            tensor<int32, [1]> layer_norm_58_axes_0 = const()[name = string("layer_norm_58_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_11_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_11_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225417216)))];
            tensor<fp16, [1024]> p_encoder_layers_11_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_11_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225419328)))];
            fp16 const_820_to_fp16 = const()[name = string("const_820_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_58_cast_fp16 = layer_norm(axes = layer_norm_58_axes_0, beta = p_encoder_layers_11_norm_feed_forward2_bias_to_fp16, epsilon = const_820_to_fp16, gamma = p_encoder_layers_11_norm_feed_forward2_weight_to_fp16, x = add_85_cast_fp16)[name = string("layer_norm_58_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_11_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225421440))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228567232))))[name = string("p_encoder_layers_11_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_107_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_11_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_58_cast_fp16)[name = string("linear_107_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_35_cast_fp16 = silu(x = linear_107_cast_fp16)[name = string("silu_35_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_11_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228600064))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231745856))))[name = string("p_encoder_layers_11_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_108_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_11_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_35_cast_fp16)[name = string("linear_108_cast_fp16")];
            fp16 const_822_to_fp16 = const()[name = string("const_822_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_41_cast_fp16 = mul(x = linear_108_cast_fp16, y = const_822_to_fp16)[name = string("mul_41_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_86_cast_fp16 = add(x = add_85_cast_fp16, y = mul_41_cast_fp16)[name = string("add_86_cast_fp16")];
            tensor<int32, [1]> layer_norm_59_axes_0 = const()[name = string("layer_norm_59_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_11_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_11_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231754112)))];
            tensor<fp16, [1024]> p_encoder_layers_11_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_11_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231756224)))];
            fp16 const_824_to_fp16 = const()[name = string("const_824_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_59_cast_fp16 = layer_norm(axes = layer_norm_59_axes_0, beta = p_encoder_layers_11_norm_out_bias_to_fp16, epsilon = const_824_to_fp16, gamma = p_encoder_layers_11_norm_out_weight_to_fp16, x = add_86_cast_fp16)[name = string("layer_norm_59_cast_fp16")];
            tensor<int32, [1]> layer_norm_60_axes_0 = const()[name = string("layer_norm_60_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_12_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_12_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231758336)))];
            tensor<fp16, [1024]> p_encoder_layers_12_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_12_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231760448)))];
            fp16 const_827_to_fp16 = const()[name = string("const_827_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_60_cast_fp16 = layer_norm(axes = layer_norm_60_axes_0, beta = p_encoder_layers_12_norm_feed_forward1_bias_to_fp16, epsilon = const_827_to_fp16, gamma = p_encoder_layers_12_norm_feed_forward1_weight_to_fp16, x = layer_norm_59_cast_fp16)[name = string("layer_norm_60_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_12_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231762560))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234908352))))[name = string("p_encoder_layers_12_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_109_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_12_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_60_cast_fp16)[name = string("linear_109_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_36_cast_fp16 = silu(x = linear_109_cast_fp16)[name = string("silu_36_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_12_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234941184))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238086976))))[name = string("p_encoder_layers_12_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_110_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_12_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_36_cast_fp16)[name = string("linear_110_cast_fp16")];
            fp16 const_829_to_fp16 = const()[name = string("const_829_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_42_cast_fp16 = mul(x = linear_110_cast_fp16, y = const_829_to_fp16)[name = string("mul_42_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_87_cast_fp16 = add(x = layer_norm_59_cast_fp16, y = mul_42_cast_fp16)[name = string("add_87_cast_fp16")];
            tensor<int32, [1]> layer_norm_61_axes_0 = const()[name = string("layer_norm_61_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_12_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_12_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238095232)))];
            tensor<fp16, [1024]> p_encoder_layers_12_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_12_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238097344)))];
            fp16 const_831_to_fp16 = const()[name = string("const_831_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_61_cast_fp16 = layer_norm(axes = layer_norm_61_axes_0, beta = p_encoder_layers_12_norm_self_att_bias_to_fp16, epsilon = const_831_to_fp16, gamma = p_encoder_layers_12_norm_self_att_weight_to_fp16, x = add_87_cast_fp16)[name = string("layer_norm_61_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_12_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238099456))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238885952))))[name = string("p_encoder_layers_12_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_111_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_12_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_61_cast_fp16)[name = string("linear_111_cast_fp16")];
            tensor<int32, [4]> const_833 = const()[name = string("const_833"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_109_cast_fp16 = reshape(shape = const_833, x = linear_111_cast_fp16)[name = string("view_109_cast_fp16")];
            tensor<int32, [4]> transpose_75_perm_0 = const()[name = string("transpose_75_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_12_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238894208))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239680704))))[name = string("p_encoder_layers_12_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_112_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_12_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_61_cast_fp16)[name = string("linear_112_cast_fp16")];
            tensor<int32, [4]> const_836 = const()[name = string("const_836"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_110_cast_fp16 = reshape(shape = const_836, x = linear_112_cast_fp16)[name = string("view_110_cast_fp16")];
            tensor<int32, [4]> transpose_76_perm_0 = const()[name = string("transpose_76_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_12_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239688960))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240475456))))[name = string("p_encoder_layers_12_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_113_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_12_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_61_cast_fp16)[name = string("linear_113_cast_fp16")];
            tensor<int32, [4]> const_839 = const()[name = string("const_839"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_111_cast_fp16 = reshape(shape = const_839, x = linear_113_cast_fp16)[name = string("view_111_cast_fp16")];
            tensor<int32, [4]> transpose_77_perm_0 = const()[name = string("transpose_77_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_112_to_fp16 = const()[name = string("view_112_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240483712)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_75_cast_fp16 = transpose(perm = transpose_75_perm_0, x = view_109_cast_fp16)[name = string("transpose_71")];
            tensor<fp16, [1, 8, 188, 128]> add_88_cast_fp16 = add(x = transpose_75_cast_fp16, y = view_112_to_fp16)[name = string("add_88_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_113_to_fp16 = const()[name = string("view_113_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240485824)))];
            tensor<fp16, [1, 8, 188, 128]> add_89_cast_fp16 = add(x = transpose_75_cast_fp16, y = view_113_to_fp16)[name = string("add_89_cast_fp16")];
            bool matmul_13_transpose_x_0 = const()[name = string("matmul_13_transpose_x_0"), val = bool(false)];
            bool matmul_13_transpose_y_0 = const()[name = string("matmul_13_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_12_to_fp16 = const()[name = string("permute_12_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240487936)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_13_cast_fp16 = matmul(transpose_x = matmul_13_transpose_x_0, transpose_y = matmul_13_transpose_y_0, x = add_89_cast_fp16, y = permute_12_to_fp16)[name = string("matmul_13_cast_fp16")];
            tensor<int32, [8]> pad_12_pad_0 = const()[name = string("pad_12_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_12_mode_0 = const()[name = string("pad_12_mode_0"), val = string("constant")];
            fp16 const_847_to_fp16 = const()[name = string("const_847_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_12_cast_fp16 = pad(constant_val = const_847_to_fp16, mode = pad_12_mode_0, pad = pad_12_pad_0, x = matmul_13_cast_fp16)[name = string("pad_12_cast_fp16")];
            tensor<int32, [4]> const_848 = const()[name = string("const_848"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_115_cast_fp16 = reshape(shape = const_848, x = pad_12_cast_fp16)[name = string("view_115_cast_fp16")];
            tensor<int32, [4]> slice_25_begin_0 = const()[name = string("slice_25_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_25_end_0 = const()[name = string("slice_25_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_25_end_mask_0 = const()[name = string("slice_25_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_25_cast_fp16 = slice_by_index(begin = slice_25_begin_0, end = slice_25_end_0, end_mask = slice_25_end_mask_0, x = view_115_cast_fp16)[name = string("slice_25_cast_fp16")];
            tensor<int32, [4]> const_852 = const()[name = string("const_852"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_116_cast_fp16 = reshape(shape = const_852, x = slice_25_cast_fp16)[name = string("view_116_cast_fp16")];
            tensor<int32, [4]> slice_26_begin_0 = const()[name = string("slice_26_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_26_end_0 = const()[name = string("slice_26_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_26_end_mask_0 = const()[name = string("slice_26_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_26_cast_fp16 = slice_by_index(begin = slice_26_begin_0, end = slice_26_end_0, end_mask = slice_26_end_mask_0, x = view_116_cast_fp16)[name = string("slice_26_cast_fp16")];
            fp16 const_856_to_fp16 = const()[name = string("const_856_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_43_cast_fp16 = mul(x = slice_26_cast_fp16, y = const_856_to_fp16)[name = string("mul_43_cast_fp16")];
            fp16 const_857_to_fp16 = const()[name = string("const_857_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_24_cast_fp16 = select(a = const_857_to_fp16, b = mul_43_cast_fp16, cond = logical_not)[name = string("masked_fill_24_cast_fp16")];
            fp16 const_858_to_fp16 = const()[name = string("const_858_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_12_1_cast_fp16 = mul(x = add_88_cast_fp16, y = const_858_to_fp16)[name = string("mul_12_1_cast_fp16")];
            bool matmul_12_transpose_y_1 = const()[name = string("matmul_12_transpose_y_1"), val = bool(true)];
            bool matmul_12_transpose_x_1 = const()[name = string("matmul_12_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_76_cast_fp16 = transpose(perm = transpose_76_perm_0, x = view_110_cast_fp16)[name = string("transpose_70")];
            tensor<fp16, [1, 8, 188, 188]> matmul_12_1_cast_fp16 = matmul(transpose_x = matmul_12_transpose_x_1, transpose_y = matmul_12_transpose_y_1, x = mul_12_1_cast_fp16, y = transpose_76_cast_fp16)[name = string("matmul_12_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_12_1_cast_fp16 = add(x = matmul_12_1_cast_fp16, y = masked_fill_24_cast_fp16)[name = string("add_12_1_cast_fp16")];
            int32 softmax_12_axis_0 = const()[name = string("softmax_12_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_12_cast_fp16 = softmax(axis = softmax_12_axis_0, x = add_12_1_cast_fp16)[name = string("softmax_12_cast_fp16")];
            bool scaled_dot_product_attention_12_transpose_x_0 = const()[name = string("scaled_dot_product_attention_12_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_12_transpose_y_0 = const()[name = string("scaled_dot_product_attention_12_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_77_cast_fp16 = transpose(perm = transpose_77_perm_0, x = view_111_cast_fp16)[name = string("transpose_69")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_12_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_12_transpose_x_0, transpose_y = scaled_dot_product_attention_12_transpose_y_0, x = softmax_12_cast_fp16, y = transpose_77_cast_fp16)[name = string("scaled_dot_product_attention_12_cast_fp16")];
            tensor<int32, [4]> transpose_78_perm_0 = const()[name = string("transpose_78_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_861 = const()[name = string("const_861"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_78_cast_fp16 = transpose(perm = transpose_78_perm_0, x = scaled_dot_product_attention_12_cast_fp16)[name = string("transpose_68")];
            tensor<fp16, [1, 188, 1024]> view_117_cast_fp16 = reshape(shape = const_861, x = transpose_78_cast_fp16)[name = string("view_117_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_12_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241256000))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242042496))))[name = string("p_encoder_layers_12_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_115_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_12_self_attn_o_proj_weight_to_fp16_palettized, x = view_117_cast_fp16)[name = string("linear_115_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_90_cast_fp16 = add(x = add_87_cast_fp16, y = linear_115_cast_fp16)[name = string("add_90_cast_fp16")];
            tensor<int32, [1]> layer_norm_62_axes_0 = const()[name = string("layer_norm_62_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_12_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_12_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242050752)))];
            tensor<fp16, [1024]> p_encoder_layers_12_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_12_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242052864)))];
            fp16 const_863_to_fp16 = const()[name = string("const_863_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_62_cast_fp16 = layer_norm(axes = layer_norm_62_axes_0, beta = p_encoder_layers_12_norm_conv_bias_to_fp16, epsilon = const_863_to_fp16, gamma = p_encoder_layers_12_norm_conv_weight_to_fp16, x = add_90_cast_fp16)[name = string("layer_norm_62_cast_fp16")];
            tensor<int32, [3]> transpose_79_perm_0 = const()[name = string("transpose_79_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_36_pad_type_0 = const()[name = string("conv1d_36_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_36_strides_0 = const()[name = string("conv1d_36_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_36_pad_0 = const()[name = string("conv1d_36_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_36_dilations_0 = const()[name = string("conv1d_36_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_36_groups_0 = const()[name = string("conv1d_36_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_12_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242054976))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243627904))))[name = string("p_encoder_layers_12_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_79_cast_fp16 = transpose(perm = transpose_79_perm_0, x = layer_norm_62_cast_fp16)[name = string("transpose_67")];
            tensor<fp16, [1, 2048, 188]> conv1d_36_cast_fp16 = conv(dilations = conv1d_36_dilations_0, groups = conv1d_36_groups_0, pad = conv1d_36_pad_0, pad_type = conv1d_36_pad_type_0, strides = conv1d_36_strides_0, weight = p_encoder_layers_12_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_79_cast_fp16)[name = string("conv1d_36_cast_fp16")];
            int32 glu_12_split_num_splits_0 = const()[name = string("glu_12_split_num_splits_0"), val = int32(2)];
            int32 glu_12_split_axis_0 = const()[name = string("glu_12_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_12_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_12_split_cast_fp16_1 = split(axis = glu_12_split_axis_0, num_splits = glu_12_split_num_splits_0, x = conv1d_36_cast_fp16)[name = string("glu_12_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_12_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_12_split_cast_fp16_1)[name = string("glu_12_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_12_cast_fp16 = mul(x = glu_12_split_cast_fp16_0, y = glu_12_split_1_sigmoid_cast_fp16)[name = string("glu_12_cast_fp16")];
            fp16 const_869_to_fp16 = const()[name = string("const_869_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_25_cast_fp16 = select(a = const_869_to_fp16, b = glu_12_cast_fp16, cond = all_1)[name = string("masked_fill_25_cast_fp16")];
            string conv1d_37_pad_type_0 = const()[name = string("conv1d_37_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_37_pad_0 = const()[name = string("conv1d_37_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_37_groups_0 = const()[name = string("conv1d_37_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_37_strides_0 = const()[name = string("conv1d_37_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_37_dilations_0 = const()[name = string("conv1d_37_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1571_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243644352))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243651328))))[name = string("const_1571_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1572_to_fp16 = const()[name = string("const_1572_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243659584)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_12_cast_fp16 = conv(bias = const_1572_to_fp16, dilations = conv1d_37_dilations_0, groups = conv1d_37_groups_0, pad = conv1d_37_pad_0, pad_type = conv1d_37_pad_type_0, strides = conv1d_37_strides_0, weight = const_1571_to_fp16_palettized, x = masked_fill_25_cast_fp16)[name = string("_native_batch_norm_legit_no_training_12_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_37_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_12_cast_fp16)[name = string("silu_37_cast_fp16")];
            string conv1d_38_pad_type_0 = const()[name = string("conv1d_38_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_38_strides_0 = const()[name = string("conv1d_38_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_38_pad_0 = const()[name = string("conv1d_38_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_38_dilations_0 = const()[name = string("conv1d_38_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_38_groups_0 = const()[name = string("conv1d_38_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_12_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243661696))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244448192))))[name = string("p_encoder_layers_12_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_38_cast_fp16 = conv(dilations = conv1d_38_dilations_0, groups = conv1d_38_groups_0, pad = conv1d_38_pad_0, pad_type = conv1d_38_pad_type_0, strides = conv1d_38_strides_0, weight = p_encoder_layers_12_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_37_cast_fp16)[name = string("conv1d_38_cast_fp16")];
            tensor<int32, [3]> transpose_80_perm_0 = const()[name = string("transpose_80_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_80_cast_fp16 = transpose(perm = transpose_80_perm_0, x = conv1d_38_cast_fp16)[name = string("transpose_66")];
            tensor<fp16, [1, 188, 1024]> add_91_cast_fp16 = add(x = add_90_cast_fp16, y = transpose_80_cast_fp16)[name = string("add_91_cast_fp16")];
            tensor<int32, [1]> layer_norm_63_axes_0 = const()[name = string("layer_norm_63_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_12_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_12_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244456448)))];
            tensor<fp16, [1024]> p_encoder_layers_12_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_12_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244458560)))];
            fp16 const_880_to_fp16 = const()[name = string("const_880_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_63_cast_fp16 = layer_norm(axes = layer_norm_63_axes_0, beta = p_encoder_layers_12_norm_feed_forward2_bias_to_fp16, epsilon = const_880_to_fp16, gamma = p_encoder_layers_12_norm_feed_forward2_weight_to_fp16, x = add_91_cast_fp16)[name = string("layer_norm_63_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_12_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244460672))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247606464))))[name = string("p_encoder_layers_12_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_116_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_12_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_63_cast_fp16)[name = string("linear_116_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_38_cast_fp16 = silu(x = linear_116_cast_fp16)[name = string("silu_38_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_12_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247639296))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250785088))))[name = string("p_encoder_layers_12_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_117_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_12_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_38_cast_fp16)[name = string("linear_117_cast_fp16")];
            fp16 const_882_to_fp16 = const()[name = string("const_882_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_44_cast_fp16 = mul(x = linear_117_cast_fp16, y = const_882_to_fp16)[name = string("mul_44_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_92_cast_fp16 = add(x = add_91_cast_fp16, y = mul_44_cast_fp16)[name = string("add_92_cast_fp16")];
            tensor<int32, [1]> layer_norm_64_axes_0 = const()[name = string("layer_norm_64_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_12_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_12_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250793344)))];
            tensor<fp16, [1024]> p_encoder_layers_12_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_12_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250795456)))];
            fp16 const_884_to_fp16 = const()[name = string("const_884_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_64_cast_fp16 = layer_norm(axes = layer_norm_64_axes_0, beta = p_encoder_layers_12_norm_out_bias_to_fp16, epsilon = const_884_to_fp16, gamma = p_encoder_layers_12_norm_out_weight_to_fp16, x = add_92_cast_fp16)[name = string("layer_norm_64_cast_fp16")];
            tensor<int32, [1]> layer_norm_65_axes_0 = const()[name = string("layer_norm_65_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_13_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_13_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250797568)))];
            tensor<fp16, [1024]> p_encoder_layers_13_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_13_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250799680)))];
            fp16 const_887_to_fp16 = const()[name = string("const_887_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_65_cast_fp16 = layer_norm(axes = layer_norm_65_axes_0, beta = p_encoder_layers_13_norm_feed_forward1_bias_to_fp16, epsilon = const_887_to_fp16, gamma = p_encoder_layers_13_norm_feed_forward1_weight_to_fp16, x = layer_norm_64_cast_fp16)[name = string("layer_norm_65_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_13_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250801792))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253947584))))[name = string("p_encoder_layers_13_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_118_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_13_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_65_cast_fp16)[name = string("linear_118_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_39_cast_fp16 = silu(x = linear_118_cast_fp16)[name = string("silu_39_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_13_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253980416))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257126208))))[name = string("p_encoder_layers_13_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_119_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_13_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_39_cast_fp16)[name = string("linear_119_cast_fp16")];
            fp16 const_889_to_fp16 = const()[name = string("const_889_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_45_cast_fp16 = mul(x = linear_119_cast_fp16, y = const_889_to_fp16)[name = string("mul_45_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_93_cast_fp16 = add(x = layer_norm_64_cast_fp16, y = mul_45_cast_fp16)[name = string("add_93_cast_fp16")];
            tensor<int32, [1]> layer_norm_66_axes_0 = const()[name = string("layer_norm_66_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_13_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_13_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257134464)))];
            tensor<fp16, [1024]> p_encoder_layers_13_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_13_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257136576)))];
            fp16 const_891_to_fp16 = const()[name = string("const_891_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_66_cast_fp16 = layer_norm(axes = layer_norm_66_axes_0, beta = p_encoder_layers_13_norm_self_att_bias_to_fp16, epsilon = const_891_to_fp16, gamma = p_encoder_layers_13_norm_self_att_weight_to_fp16, x = add_93_cast_fp16)[name = string("layer_norm_66_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_13_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257138688))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257925184))))[name = string("p_encoder_layers_13_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_120_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_13_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_66_cast_fp16)[name = string("linear_120_cast_fp16")];
            tensor<int32, [4]> const_893 = const()[name = string("const_893"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_118_cast_fp16 = reshape(shape = const_893, x = linear_120_cast_fp16)[name = string("view_118_cast_fp16")];
            tensor<int32, [4]> transpose_81_perm_0 = const()[name = string("transpose_81_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_13_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257933440))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258719936))))[name = string("p_encoder_layers_13_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_121_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_13_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_66_cast_fp16)[name = string("linear_121_cast_fp16")];
            tensor<int32, [4]> const_896 = const()[name = string("const_896"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_119_cast_fp16 = reshape(shape = const_896, x = linear_121_cast_fp16)[name = string("view_119_cast_fp16")];
            tensor<int32, [4]> transpose_82_perm_0 = const()[name = string("transpose_82_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_13_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258728192))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259514688))))[name = string("p_encoder_layers_13_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_122_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_13_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_66_cast_fp16)[name = string("linear_122_cast_fp16")];
            tensor<int32, [4]> const_899 = const()[name = string("const_899"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_120_cast_fp16 = reshape(shape = const_899, x = linear_122_cast_fp16)[name = string("view_120_cast_fp16")];
            tensor<int32, [4]> transpose_83_perm_0 = const()[name = string("transpose_83_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_121_to_fp16 = const()[name = string("view_121_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259522944)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_81_cast_fp16 = transpose(perm = transpose_81_perm_0, x = view_118_cast_fp16)[name = string("transpose_65")];
            tensor<fp16, [1, 8, 188, 128]> add_94_cast_fp16 = add(x = transpose_81_cast_fp16, y = view_121_to_fp16)[name = string("add_94_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_122_to_fp16 = const()[name = string("view_122_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259525056)))];
            tensor<fp16, [1, 8, 188, 128]> add_95_cast_fp16 = add(x = transpose_81_cast_fp16, y = view_122_to_fp16)[name = string("add_95_cast_fp16")];
            bool matmul_14_transpose_x_0 = const()[name = string("matmul_14_transpose_x_0"), val = bool(false)];
            bool matmul_14_transpose_y_0 = const()[name = string("matmul_14_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_13_to_fp16 = const()[name = string("permute_13_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259527168)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_14_cast_fp16 = matmul(transpose_x = matmul_14_transpose_x_0, transpose_y = matmul_14_transpose_y_0, x = add_95_cast_fp16, y = permute_13_to_fp16)[name = string("matmul_14_cast_fp16")];
            tensor<int32, [8]> pad_13_pad_0 = const()[name = string("pad_13_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_13_mode_0 = const()[name = string("pad_13_mode_0"), val = string("constant")];
            fp16 const_907_to_fp16 = const()[name = string("const_907_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_13_cast_fp16 = pad(constant_val = const_907_to_fp16, mode = pad_13_mode_0, pad = pad_13_pad_0, x = matmul_14_cast_fp16)[name = string("pad_13_cast_fp16")];
            tensor<int32, [4]> const_908 = const()[name = string("const_908"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_124_cast_fp16 = reshape(shape = const_908, x = pad_13_cast_fp16)[name = string("view_124_cast_fp16")];
            tensor<int32, [4]> slice_27_begin_0 = const()[name = string("slice_27_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_27_end_0 = const()[name = string("slice_27_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_27_end_mask_0 = const()[name = string("slice_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_27_cast_fp16 = slice_by_index(begin = slice_27_begin_0, end = slice_27_end_0, end_mask = slice_27_end_mask_0, x = view_124_cast_fp16)[name = string("slice_27_cast_fp16")];
            tensor<int32, [4]> const_912 = const()[name = string("const_912"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_125_cast_fp16 = reshape(shape = const_912, x = slice_27_cast_fp16)[name = string("view_125_cast_fp16")];
            tensor<int32, [4]> slice_28_begin_0 = const()[name = string("slice_28_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_28_end_0 = const()[name = string("slice_28_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_28_end_mask_0 = const()[name = string("slice_28_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_28_cast_fp16 = slice_by_index(begin = slice_28_begin_0, end = slice_28_end_0, end_mask = slice_28_end_mask_0, x = view_125_cast_fp16)[name = string("slice_28_cast_fp16")];
            fp16 const_916_to_fp16 = const()[name = string("const_916_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_46_cast_fp16 = mul(x = slice_28_cast_fp16, y = const_916_to_fp16)[name = string("mul_46_cast_fp16")];
            fp16 const_917_to_fp16 = const()[name = string("const_917_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_26_cast_fp16 = select(a = const_917_to_fp16, b = mul_46_cast_fp16, cond = logical_not)[name = string("masked_fill_26_cast_fp16")];
            fp16 const_918_to_fp16 = const()[name = string("const_918_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_13_1_cast_fp16 = mul(x = add_94_cast_fp16, y = const_918_to_fp16)[name = string("mul_13_1_cast_fp16")];
            bool matmul_13_transpose_y_1 = const()[name = string("matmul_13_transpose_y_1"), val = bool(true)];
            bool matmul_13_transpose_x_1 = const()[name = string("matmul_13_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_82_cast_fp16 = transpose(perm = transpose_82_perm_0, x = view_119_cast_fp16)[name = string("transpose_64")];
            tensor<fp16, [1, 8, 188, 188]> matmul_13_1_cast_fp16 = matmul(transpose_x = matmul_13_transpose_x_1, transpose_y = matmul_13_transpose_y_1, x = mul_13_1_cast_fp16, y = transpose_82_cast_fp16)[name = string("matmul_13_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_13_1_cast_fp16 = add(x = matmul_13_1_cast_fp16, y = masked_fill_26_cast_fp16)[name = string("add_13_1_cast_fp16")];
            int32 softmax_13_axis_0 = const()[name = string("softmax_13_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_13_cast_fp16 = softmax(axis = softmax_13_axis_0, x = add_13_1_cast_fp16)[name = string("softmax_13_cast_fp16")];
            bool scaled_dot_product_attention_13_transpose_x_0 = const()[name = string("scaled_dot_product_attention_13_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_13_transpose_y_0 = const()[name = string("scaled_dot_product_attention_13_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_83_cast_fp16 = transpose(perm = transpose_83_perm_0, x = view_120_cast_fp16)[name = string("transpose_63")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_13_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_13_transpose_x_0, transpose_y = scaled_dot_product_attention_13_transpose_y_0, x = softmax_13_cast_fp16, y = transpose_83_cast_fp16)[name = string("scaled_dot_product_attention_13_cast_fp16")];
            tensor<int32, [4]> transpose_84_perm_0 = const()[name = string("transpose_84_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_921 = const()[name = string("const_921"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_84_cast_fp16 = transpose(perm = transpose_84_perm_0, x = scaled_dot_product_attention_13_cast_fp16)[name = string("transpose_62")];
            tensor<fp16, [1, 188, 1024]> view_126_cast_fp16 = reshape(shape = const_921, x = transpose_84_cast_fp16)[name = string("view_126_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_13_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260295232))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261081728))))[name = string("p_encoder_layers_13_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_124_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_13_self_attn_o_proj_weight_to_fp16_palettized, x = view_126_cast_fp16)[name = string("linear_124_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_96_cast_fp16 = add(x = add_93_cast_fp16, y = linear_124_cast_fp16)[name = string("add_96_cast_fp16")];
            tensor<int32, [1]> layer_norm_67_axes_0 = const()[name = string("layer_norm_67_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_13_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_13_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261089984)))];
            tensor<fp16, [1024]> p_encoder_layers_13_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_13_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261092096)))];
            fp16 const_923_to_fp16 = const()[name = string("const_923_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_67_cast_fp16 = layer_norm(axes = layer_norm_67_axes_0, beta = p_encoder_layers_13_norm_conv_bias_to_fp16, epsilon = const_923_to_fp16, gamma = p_encoder_layers_13_norm_conv_weight_to_fp16, x = add_96_cast_fp16)[name = string("layer_norm_67_cast_fp16")];
            tensor<int32, [3]> transpose_85_perm_0 = const()[name = string("transpose_85_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_39_pad_type_0 = const()[name = string("conv1d_39_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_39_strides_0 = const()[name = string("conv1d_39_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_39_pad_0 = const()[name = string("conv1d_39_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_39_dilations_0 = const()[name = string("conv1d_39_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_39_groups_0 = const()[name = string("conv1d_39_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_13_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261094208))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262667136))))[name = string("p_encoder_layers_13_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_85_cast_fp16 = transpose(perm = transpose_85_perm_0, x = layer_norm_67_cast_fp16)[name = string("transpose_61")];
            tensor<fp16, [1, 2048, 188]> conv1d_39_cast_fp16 = conv(dilations = conv1d_39_dilations_0, groups = conv1d_39_groups_0, pad = conv1d_39_pad_0, pad_type = conv1d_39_pad_type_0, strides = conv1d_39_strides_0, weight = p_encoder_layers_13_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_85_cast_fp16)[name = string("conv1d_39_cast_fp16")];
            int32 glu_13_split_num_splits_0 = const()[name = string("glu_13_split_num_splits_0"), val = int32(2)];
            int32 glu_13_split_axis_0 = const()[name = string("glu_13_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_13_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_13_split_cast_fp16_1 = split(axis = glu_13_split_axis_0, num_splits = glu_13_split_num_splits_0, x = conv1d_39_cast_fp16)[name = string("glu_13_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_13_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_13_split_cast_fp16_1)[name = string("glu_13_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_13_cast_fp16 = mul(x = glu_13_split_cast_fp16_0, y = glu_13_split_1_sigmoid_cast_fp16)[name = string("glu_13_cast_fp16")];
            fp16 const_929_to_fp16 = const()[name = string("const_929_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_27_cast_fp16 = select(a = const_929_to_fp16, b = glu_13_cast_fp16, cond = all_1)[name = string("masked_fill_27_cast_fp16")];
            string conv1d_40_pad_type_0 = const()[name = string("conv1d_40_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_40_pad_0 = const()[name = string("conv1d_40_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_40_groups_0 = const()[name = string("conv1d_40_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_40_strides_0 = const()[name = string("conv1d_40_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_40_dilations_0 = const()[name = string("conv1d_40_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1573_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262683584))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262690560))))[name = string("const_1573_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1574_to_fp16 = const()[name = string("const_1574_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262698816)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_13_cast_fp16 = conv(bias = const_1574_to_fp16, dilations = conv1d_40_dilations_0, groups = conv1d_40_groups_0, pad = conv1d_40_pad_0, pad_type = conv1d_40_pad_type_0, strides = conv1d_40_strides_0, weight = const_1573_to_fp16_palettized, x = masked_fill_27_cast_fp16)[name = string("_native_batch_norm_legit_no_training_13_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_40_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_13_cast_fp16)[name = string("silu_40_cast_fp16")];
            string conv1d_41_pad_type_0 = const()[name = string("conv1d_41_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_41_strides_0 = const()[name = string("conv1d_41_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_41_pad_0 = const()[name = string("conv1d_41_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_41_dilations_0 = const()[name = string("conv1d_41_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_41_groups_0 = const()[name = string("conv1d_41_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_13_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262700928))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263487424))))[name = string("p_encoder_layers_13_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_41_cast_fp16 = conv(dilations = conv1d_41_dilations_0, groups = conv1d_41_groups_0, pad = conv1d_41_pad_0, pad_type = conv1d_41_pad_type_0, strides = conv1d_41_strides_0, weight = p_encoder_layers_13_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_40_cast_fp16)[name = string("conv1d_41_cast_fp16")];
            tensor<int32, [3]> transpose_86_perm_0 = const()[name = string("transpose_86_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_86_cast_fp16 = transpose(perm = transpose_86_perm_0, x = conv1d_41_cast_fp16)[name = string("transpose_60")];
            tensor<fp16, [1, 188, 1024]> add_97_cast_fp16 = add(x = add_96_cast_fp16, y = transpose_86_cast_fp16)[name = string("add_97_cast_fp16")];
            tensor<int32, [1]> layer_norm_68_axes_0 = const()[name = string("layer_norm_68_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_13_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_13_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263495680)))];
            tensor<fp16, [1024]> p_encoder_layers_13_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_13_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263497792)))];
            fp16 const_940_to_fp16 = const()[name = string("const_940_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_68_cast_fp16 = layer_norm(axes = layer_norm_68_axes_0, beta = p_encoder_layers_13_norm_feed_forward2_bias_to_fp16, epsilon = const_940_to_fp16, gamma = p_encoder_layers_13_norm_feed_forward2_weight_to_fp16, x = add_97_cast_fp16)[name = string("layer_norm_68_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_13_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263499904))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(266645696))))[name = string("p_encoder_layers_13_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_125_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_13_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_68_cast_fp16)[name = string("linear_125_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_41_cast_fp16 = silu(x = linear_125_cast_fp16)[name = string("silu_41_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_13_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(266678528))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269824320))))[name = string("p_encoder_layers_13_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_126_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_13_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_41_cast_fp16)[name = string("linear_126_cast_fp16")];
            fp16 const_942_to_fp16 = const()[name = string("const_942_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_47_cast_fp16 = mul(x = linear_126_cast_fp16, y = const_942_to_fp16)[name = string("mul_47_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_98_cast_fp16 = add(x = add_97_cast_fp16, y = mul_47_cast_fp16)[name = string("add_98_cast_fp16")];
            tensor<int32, [1]> layer_norm_69_axes_0 = const()[name = string("layer_norm_69_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_13_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_13_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269832576)))];
            tensor<fp16, [1024]> p_encoder_layers_13_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_13_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269834688)))];
            fp16 const_944_to_fp16 = const()[name = string("const_944_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_69_cast_fp16 = layer_norm(axes = layer_norm_69_axes_0, beta = p_encoder_layers_13_norm_out_bias_to_fp16, epsilon = const_944_to_fp16, gamma = p_encoder_layers_13_norm_out_weight_to_fp16, x = add_98_cast_fp16)[name = string("layer_norm_69_cast_fp16")];
            tensor<int32, [1]> layer_norm_70_axes_0 = const()[name = string("layer_norm_70_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_14_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_14_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269836800)))];
            tensor<fp16, [1024]> p_encoder_layers_14_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_14_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269838912)))];
            fp16 const_947_to_fp16 = const()[name = string("const_947_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_70_cast_fp16 = layer_norm(axes = layer_norm_70_axes_0, beta = p_encoder_layers_14_norm_feed_forward1_bias_to_fp16, epsilon = const_947_to_fp16, gamma = p_encoder_layers_14_norm_feed_forward1_weight_to_fp16, x = layer_norm_69_cast_fp16)[name = string("layer_norm_70_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_14_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269841024))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272986816))))[name = string("p_encoder_layers_14_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_127_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_14_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_70_cast_fp16)[name = string("linear_127_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_42_cast_fp16 = silu(x = linear_127_cast_fp16)[name = string("silu_42_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_14_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273019648))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276165440))))[name = string("p_encoder_layers_14_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_128_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_14_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_42_cast_fp16)[name = string("linear_128_cast_fp16")];
            fp16 const_949_to_fp16 = const()[name = string("const_949_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_48_cast_fp16 = mul(x = linear_128_cast_fp16, y = const_949_to_fp16)[name = string("mul_48_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_99_cast_fp16 = add(x = layer_norm_69_cast_fp16, y = mul_48_cast_fp16)[name = string("add_99_cast_fp16")];
            tensor<int32, [1]> layer_norm_71_axes_0 = const()[name = string("layer_norm_71_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_14_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_14_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276173696)))];
            tensor<fp16, [1024]> p_encoder_layers_14_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_14_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276175808)))];
            fp16 const_951_to_fp16 = const()[name = string("const_951_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_71_cast_fp16 = layer_norm(axes = layer_norm_71_axes_0, beta = p_encoder_layers_14_norm_self_att_bias_to_fp16, epsilon = const_951_to_fp16, gamma = p_encoder_layers_14_norm_self_att_weight_to_fp16, x = add_99_cast_fp16)[name = string("layer_norm_71_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_14_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276177920))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276964416))))[name = string("p_encoder_layers_14_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_129_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_14_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_71_cast_fp16)[name = string("linear_129_cast_fp16")];
            tensor<int32, [4]> const_953 = const()[name = string("const_953"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_127_cast_fp16 = reshape(shape = const_953, x = linear_129_cast_fp16)[name = string("view_127_cast_fp16")];
            tensor<int32, [4]> transpose_87_perm_0 = const()[name = string("transpose_87_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_14_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276972672))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277759168))))[name = string("p_encoder_layers_14_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_130_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_14_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_71_cast_fp16)[name = string("linear_130_cast_fp16")];
            tensor<int32, [4]> const_956 = const()[name = string("const_956"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_128_cast_fp16 = reshape(shape = const_956, x = linear_130_cast_fp16)[name = string("view_128_cast_fp16")];
            tensor<int32, [4]> transpose_88_perm_0 = const()[name = string("transpose_88_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_14_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277767424))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278553920))))[name = string("p_encoder_layers_14_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_131_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_14_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_71_cast_fp16)[name = string("linear_131_cast_fp16")];
            tensor<int32, [4]> const_959 = const()[name = string("const_959"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_129_cast_fp16 = reshape(shape = const_959, x = linear_131_cast_fp16)[name = string("view_129_cast_fp16")];
            tensor<int32, [4]> transpose_89_perm_0 = const()[name = string("transpose_89_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_130_to_fp16 = const()[name = string("view_130_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278562176)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_87_cast_fp16 = transpose(perm = transpose_87_perm_0, x = view_127_cast_fp16)[name = string("transpose_59")];
            tensor<fp16, [1, 8, 188, 128]> add_100_cast_fp16 = add(x = transpose_87_cast_fp16, y = view_130_to_fp16)[name = string("add_100_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_131_to_fp16 = const()[name = string("view_131_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278564288)))];
            tensor<fp16, [1, 8, 188, 128]> add_101_cast_fp16 = add(x = transpose_87_cast_fp16, y = view_131_to_fp16)[name = string("add_101_cast_fp16")];
            bool matmul_15_transpose_x_0 = const()[name = string("matmul_15_transpose_x_0"), val = bool(false)];
            bool matmul_15_transpose_y_0 = const()[name = string("matmul_15_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_14_to_fp16 = const()[name = string("permute_14_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278566400)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_15_cast_fp16 = matmul(transpose_x = matmul_15_transpose_x_0, transpose_y = matmul_15_transpose_y_0, x = add_101_cast_fp16, y = permute_14_to_fp16)[name = string("matmul_15_cast_fp16")];
            tensor<int32, [8]> pad_14_pad_0 = const()[name = string("pad_14_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_14_mode_0 = const()[name = string("pad_14_mode_0"), val = string("constant")];
            fp16 const_967_to_fp16 = const()[name = string("const_967_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_14_cast_fp16 = pad(constant_val = const_967_to_fp16, mode = pad_14_mode_0, pad = pad_14_pad_0, x = matmul_15_cast_fp16)[name = string("pad_14_cast_fp16")];
            tensor<int32, [4]> const_968 = const()[name = string("const_968"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_133_cast_fp16 = reshape(shape = const_968, x = pad_14_cast_fp16)[name = string("view_133_cast_fp16")];
            tensor<int32, [4]> slice_29_begin_0 = const()[name = string("slice_29_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_29_end_0 = const()[name = string("slice_29_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_29_end_mask_0 = const()[name = string("slice_29_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_29_cast_fp16 = slice_by_index(begin = slice_29_begin_0, end = slice_29_end_0, end_mask = slice_29_end_mask_0, x = view_133_cast_fp16)[name = string("slice_29_cast_fp16")];
            tensor<int32, [4]> const_972 = const()[name = string("const_972"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_134_cast_fp16 = reshape(shape = const_972, x = slice_29_cast_fp16)[name = string("view_134_cast_fp16")];
            tensor<int32, [4]> slice_30_begin_0 = const()[name = string("slice_30_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_30_end_0 = const()[name = string("slice_30_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_30_end_mask_0 = const()[name = string("slice_30_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_30_cast_fp16 = slice_by_index(begin = slice_30_begin_0, end = slice_30_end_0, end_mask = slice_30_end_mask_0, x = view_134_cast_fp16)[name = string("slice_30_cast_fp16")];
            fp16 const_976_to_fp16 = const()[name = string("const_976_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_49_cast_fp16 = mul(x = slice_30_cast_fp16, y = const_976_to_fp16)[name = string("mul_49_cast_fp16")];
            fp16 const_977_to_fp16 = const()[name = string("const_977_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_28_cast_fp16 = select(a = const_977_to_fp16, b = mul_49_cast_fp16, cond = logical_not)[name = string("masked_fill_28_cast_fp16")];
            fp16 const_978_to_fp16 = const()[name = string("const_978_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_14_1_cast_fp16 = mul(x = add_100_cast_fp16, y = const_978_to_fp16)[name = string("mul_14_1_cast_fp16")];
            bool matmul_14_transpose_y_1 = const()[name = string("matmul_14_transpose_y_1"), val = bool(true)];
            bool matmul_14_transpose_x_1 = const()[name = string("matmul_14_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_88_cast_fp16 = transpose(perm = transpose_88_perm_0, x = view_128_cast_fp16)[name = string("transpose_58")];
            tensor<fp16, [1, 8, 188, 188]> matmul_14_1_cast_fp16 = matmul(transpose_x = matmul_14_transpose_x_1, transpose_y = matmul_14_transpose_y_1, x = mul_14_1_cast_fp16, y = transpose_88_cast_fp16)[name = string("matmul_14_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_14_1_cast_fp16 = add(x = matmul_14_1_cast_fp16, y = masked_fill_28_cast_fp16)[name = string("add_14_1_cast_fp16")];
            int32 softmax_14_axis_0 = const()[name = string("softmax_14_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_14_cast_fp16 = softmax(axis = softmax_14_axis_0, x = add_14_1_cast_fp16)[name = string("softmax_14_cast_fp16")];
            bool scaled_dot_product_attention_14_transpose_x_0 = const()[name = string("scaled_dot_product_attention_14_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_14_transpose_y_0 = const()[name = string("scaled_dot_product_attention_14_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_89_cast_fp16 = transpose(perm = transpose_89_perm_0, x = view_129_cast_fp16)[name = string("transpose_57")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_14_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_14_transpose_x_0, transpose_y = scaled_dot_product_attention_14_transpose_y_0, x = softmax_14_cast_fp16, y = transpose_89_cast_fp16)[name = string("scaled_dot_product_attention_14_cast_fp16")];
            tensor<int32, [4]> transpose_90_perm_0 = const()[name = string("transpose_90_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_981 = const()[name = string("const_981"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_90_cast_fp16 = transpose(perm = transpose_90_perm_0, x = scaled_dot_product_attention_14_cast_fp16)[name = string("transpose_56")];
            tensor<fp16, [1, 188, 1024]> view_135_cast_fp16 = reshape(shape = const_981, x = transpose_90_cast_fp16)[name = string("view_135_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_14_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279334464))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280120960))))[name = string("p_encoder_layers_14_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_133_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_14_self_attn_o_proj_weight_to_fp16_palettized, x = view_135_cast_fp16)[name = string("linear_133_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_102_cast_fp16 = add(x = add_99_cast_fp16, y = linear_133_cast_fp16)[name = string("add_102_cast_fp16")];
            tensor<int32, [1]> layer_norm_72_axes_0 = const()[name = string("layer_norm_72_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_14_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_14_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280129216)))];
            tensor<fp16, [1024]> p_encoder_layers_14_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_14_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280131328)))];
            fp16 const_983_to_fp16 = const()[name = string("const_983_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_72_cast_fp16 = layer_norm(axes = layer_norm_72_axes_0, beta = p_encoder_layers_14_norm_conv_bias_to_fp16, epsilon = const_983_to_fp16, gamma = p_encoder_layers_14_norm_conv_weight_to_fp16, x = add_102_cast_fp16)[name = string("layer_norm_72_cast_fp16")];
            tensor<int32, [3]> transpose_91_perm_0 = const()[name = string("transpose_91_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_42_pad_type_0 = const()[name = string("conv1d_42_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_42_strides_0 = const()[name = string("conv1d_42_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_42_pad_0 = const()[name = string("conv1d_42_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_42_dilations_0 = const()[name = string("conv1d_42_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_42_groups_0 = const()[name = string("conv1d_42_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_14_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280133440))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281706368))))[name = string("p_encoder_layers_14_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_91_cast_fp16 = transpose(perm = transpose_91_perm_0, x = layer_norm_72_cast_fp16)[name = string("transpose_55")];
            tensor<fp16, [1, 2048, 188]> conv1d_42_cast_fp16 = conv(dilations = conv1d_42_dilations_0, groups = conv1d_42_groups_0, pad = conv1d_42_pad_0, pad_type = conv1d_42_pad_type_0, strides = conv1d_42_strides_0, weight = p_encoder_layers_14_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_91_cast_fp16)[name = string("conv1d_42_cast_fp16")];
            int32 glu_14_split_num_splits_0 = const()[name = string("glu_14_split_num_splits_0"), val = int32(2)];
            int32 glu_14_split_axis_0 = const()[name = string("glu_14_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_14_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_14_split_cast_fp16_1 = split(axis = glu_14_split_axis_0, num_splits = glu_14_split_num_splits_0, x = conv1d_42_cast_fp16)[name = string("glu_14_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_14_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_14_split_cast_fp16_1)[name = string("glu_14_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_14_cast_fp16 = mul(x = glu_14_split_cast_fp16_0, y = glu_14_split_1_sigmoid_cast_fp16)[name = string("glu_14_cast_fp16")];
            fp16 const_989_to_fp16 = const()[name = string("const_989_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_29_cast_fp16 = select(a = const_989_to_fp16, b = glu_14_cast_fp16, cond = all_1)[name = string("masked_fill_29_cast_fp16")];
            string conv1d_43_pad_type_0 = const()[name = string("conv1d_43_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_43_pad_0 = const()[name = string("conv1d_43_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_43_groups_0 = const()[name = string("conv1d_43_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_43_strides_0 = const()[name = string("conv1d_43_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_43_dilations_0 = const()[name = string("conv1d_43_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1575_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281722816))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281729792))))[name = string("const_1575_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1576_to_fp16 = const()[name = string("const_1576_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281738048)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_14_cast_fp16 = conv(bias = const_1576_to_fp16, dilations = conv1d_43_dilations_0, groups = conv1d_43_groups_0, pad = conv1d_43_pad_0, pad_type = conv1d_43_pad_type_0, strides = conv1d_43_strides_0, weight = const_1575_to_fp16_palettized, x = masked_fill_29_cast_fp16)[name = string("_native_batch_norm_legit_no_training_14_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_43_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_14_cast_fp16)[name = string("silu_43_cast_fp16")];
            string conv1d_44_pad_type_0 = const()[name = string("conv1d_44_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_44_strides_0 = const()[name = string("conv1d_44_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_44_pad_0 = const()[name = string("conv1d_44_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_44_dilations_0 = const()[name = string("conv1d_44_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_44_groups_0 = const()[name = string("conv1d_44_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_14_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281740160))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282526656))))[name = string("p_encoder_layers_14_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_44_cast_fp16 = conv(dilations = conv1d_44_dilations_0, groups = conv1d_44_groups_0, pad = conv1d_44_pad_0, pad_type = conv1d_44_pad_type_0, strides = conv1d_44_strides_0, weight = p_encoder_layers_14_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_43_cast_fp16)[name = string("conv1d_44_cast_fp16")];
            tensor<int32, [3]> transpose_92_perm_0 = const()[name = string("transpose_92_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_92_cast_fp16 = transpose(perm = transpose_92_perm_0, x = conv1d_44_cast_fp16)[name = string("transpose_54")];
            tensor<fp16, [1, 188, 1024]> add_103_cast_fp16 = add(x = add_102_cast_fp16, y = transpose_92_cast_fp16)[name = string("add_103_cast_fp16")];
            tensor<int32, [1]> layer_norm_73_axes_0 = const()[name = string("layer_norm_73_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_14_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_14_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282534912)))];
            tensor<fp16, [1024]> p_encoder_layers_14_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_14_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282537024)))];
            fp16 const_1000_to_fp16 = const()[name = string("const_1000_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_73_cast_fp16 = layer_norm(axes = layer_norm_73_axes_0, beta = p_encoder_layers_14_norm_feed_forward2_bias_to_fp16, epsilon = const_1000_to_fp16, gamma = p_encoder_layers_14_norm_feed_forward2_weight_to_fp16, x = add_103_cast_fp16)[name = string("layer_norm_73_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_14_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282539136))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285684928))))[name = string("p_encoder_layers_14_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_134_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_14_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_73_cast_fp16)[name = string("linear_134_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_44_cast_fp16 = silu(x = linear_134_cast_fp16)[name = string("silu_44_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_14_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285717760))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288863552))))[name = string("p_encoder_layers_14_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_135_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_14_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_44_cast_fp16)[name = string("linear_135_cast_fp16")];
            fp16 const_1002_to_fp16 = const()[name = string("const_1002_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_50_cast_fp16 = mul(x = linear_135_cast_fp16, y = const_1002_to_fp16)[name = string("mul_50_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_104_cast_fp16 = add(x = add_103_cast_fp16, y = mul_50_cast_fp16)[name = string("add_104_cast_fp16")];
            tensor<int32, [1]> layer_norm_74_axes_0 = const()[name = string("layer_norm_74_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_14_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_14_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288871808)))];
            tensor<fp16, [1024]> p_encoder_layers_14_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_14_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288873920)))];
            fp16 const_1004_to_fp16 = const()[name = string("const_1004_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_74_cast_fp16 = layer_norm(axes = layer_norm_74_axes_0, beta = p_encoder_layers_14_norm_out_bias_to_fp16, epsilon = const_1004_to_fp16, gamma = p_encoder_layers_14_norm_out_weight_to_fp16, x = add_104_cast_fp16)[name = string("layer_norm_74_cast_fp16")];
            tensor<int32, [1]> layer_norm_75_axes_0 = const()[name = string("layer_norm_75_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_15_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_15_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288876032)))];
            tensor<fp16, [1024]> p_encoder_layers_15_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_15_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288878144)))];
            fp16 const_1007_to_fp16 = const()[name = string("const_1007_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_75_cast_fp16 = layer_norm(axes = layer_norm_75_axes_0, beta = p_encoder_layers_15_norm_feed_forward1_bias_to_fp16, epsilon = const_1007_to_fp16, gamma = p_encoder_layers_15_norm_feed_forward1_weight_to_fp16, x = layer_norm_74_cast_fp16)[name = string("layer_norm_75_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_15_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288880256))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292026048))))[name = string("p_encoder_layers_15_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_136_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_15_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_75_cast_fp16)[name = string("linear_136_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_45_cast_fp16 = silu(x = linear_136_cast_fp16)[name = string("silu_45_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_15_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292058880))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295204672))))[name = string("p_encoder_layers_15_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_137_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_15_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_45_cast_fp16)[name = string("linear_137_cast_fp16")];
            fp16 const_1009_to_fp16 = const()[name = string("const_1009_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_51_cast_fp16 = mul(x = linear_137_cast_fp16, y = const_1009_to_fp16)[name = string("mul_51_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_105_cast_fp16 = add(x = layer_norm_74_cast_fp16, y = mul_51_cast_fp16)[name = string("add_105_cast_fp16")];
            tensor<int32, [1]> layer_norm_76_axes_0 = const()[name = string("layer_norm_76_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_15_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_15_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295212928)))];
            tensor<fp16, [1024]> p_encoder_layers_15_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_15_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295215040)))];
            fp16 const_1011_to_fp16 = const()[name = string("const_1011_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_76_cast_fp16 = layer_norm(axes = layer_norm_76_axes_0, beta = p_encoder_layers_15_norm_self_att_bias_to_fp16, epsilon = const_1011_to_fp16, gamma = p_encoder_layers_15_norm_self_att_weight_to_fp16, x = add_105_cast_fp16)[name = string("layer_norm_76_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_15_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295217152))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296003648))))[name = string("p_encoder_layers_15_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_138_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_15_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_76_cast_fp16)[name = string("linear_138_cast_fp16")];
            tensor<int32, [4]> const_1013 = const()[name = string("const_1013"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_136_cast_fp16 = reshape(shape = const_1013, x = linear_138_cast_fp16)[name = string("view_136_cast_fp16")];
            tensor<int32, [4]> transpose_93_perm_0 = const()[name = string("transpose_93_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_15_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296011904))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296798400))))[name = string("p_encoder_layers_15_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_139_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_15_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_76_cast_fp16)[name = string("linear_139_cast_fp16")];
            tensor<int32, [4]> const_1016 = const()[name = string("const_1016"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_137_cast_fp16 = reshape(shape = const_1016, x = linear_139_cast_fp16)[name = string("view_137_cast_fp16")];
            tensor<int32, [4]> transpose_94_perm_0 = const()[name = string("transpose_94_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_15_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296806656))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297593152))))[name = string("p_encoder_layers_15_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_140_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_15_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_76_cast_fp16)[name = string("linear_140_cast_fp16")];
            tensor<int32, [4]> const_1019 = const()[name = string("const_1019"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_138_cast_fp16 = reshape(shape = const_1019, x = linear_140_cast_fp16)[name = string("view_138_cast_fp16")];
            tensor<int32, [4]> transpose_95_perm_0 = const()[name = string("transpose_95_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_139_to_fp16 = const()[name = string("view_139_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297601408)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_93_cast_fp16 = transpose(perm = transpose_93_perm_0, x = view_136_cast_fp16)[name = string("transpose_53")];
            tensor<fp16, [1, 8, 188, 128]> add_106_cast_fp16 = add(x = transpose_93_cast_fp16, y = view_139_to_fp16)[name = string("add_106_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_140_to_fp16 = const()[name = string("view_140_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297603520)))];
            tensor<fp16, [1, 8, 188, 128]> add_107_cast_fp16 = add(x = transpose_93_cast_fp16, y = view_140_to_fp16)[name = string("add_107_cast_fp16")];
            bool matmul_16_transpose_x_0 = const()[name = string("matmul_16_transpose_x_0"), val = bool(false)];
            bool matmul_16_transpose_y_0 = const()[name = string("matmul_16_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_15_to_fp16 = const()[name = string("permute_15_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297605632)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_16_cast_fp16 = matmul(transpose_x = matmul_16_transpose_x_0, transpose_y = matmul_16_transpose_y_0, x = add_107_cast_fp16, y = permute_15_to_fp16)[name = string("matmul_16_cast_fp16")];
            tensor<int32, [8]> pad_15_pad_0 = const()[name = string("pad_15_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_15_mode_0 = const()[name = string("pad_15_mode_0"), val = string("constant")];
            fp16 const_1027_to_fp16 = const()[name = string("const_1027_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_15_cast_fp16 = pad(constant_val = const_1027_to_fp16, mode = pad_15_mode_0, pad = pad_15_pad_0, x = matmul_16_cast_fp16)[name = string("pad_15_cast_fp16")];
            tensor<int32, [4]> const_1028 = const()[name = string("const_1028"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_142_cast_fp16 = reshape(shape = const_1028, x = pad_15_cast_fp16)[name = string("view_142_cast_fp16")];
            tensor<int32, [4]> slice_31_begin_0 = const()[name = string("slice_31_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_31_end_0 = const()[name = string("slice_31_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_31_end_mask_0 = const()[name = string("slice_31_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_31_cast_fp16 = slice_by_index(begin = slice_31_begin_0, end = slice_31_end_0, end_mask = slice_31_end_mask_0, x = view_142_cast_fp16)[name = string("slice_31_cast_fp16")];
            tensor<int32, [4]> const_1032 = const()[name = string("const_1032"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_143_cast_fp16 = reshape(shape = const_1032, x = slice_31_cast_fp16)[name = string("view_143_cast_fp16")];
            tensor<int32, [4]> slice_32_begin_0 = const()[name = string("slice_32_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_32_end_0 = const()[name = string("slice_32_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_32_end_mask_0 = const()[name = string("slice_32_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_32_cast_fp16 = slice_by_index(begin = slice_32_begin_0, end = slice_32_end_0, end_mask = slice_32_end_mask_0, x = view_143_cast_fp16)[name = string("slice_32_cast_fp16")];
            fp16 const_1036_to_fp16 = const()[name = string("const_1036_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_52_cast_fp16 = mul(x = slice_32_cast_fp16, y = const_1036_to_fp16)[name = string("mul_52_cast_fp16")];
            fp16 const_1037_to_fp16 = const()[name = string("const_1037_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_30_cast_fp16 = select(a = const_1037_to_fp16, b = mul_52_cast_fp16, cond = logical_not)[name = string("masked_fill_30_cast_fp16")];
            fp16 const_1038_to_fp16 = const()[name = string("const_1038_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_15_1_cast_fp16 = mul(x = add_106_cast_fp16, y = const_1038_to_fp16)[name = string("mul_15_1_cast_fp16")];
            bool matmul_15_transpose_y_1 = const()[name = string("matmul_15_transpose_y_1"), val = bool(true)];
            bool matmul_15_transpose_x_1 = const()[name = string("matmul_15_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_94_cast_fp16 = transpose(perm = transpose_94_perm_0, x = view_137_cast_fp16)[name = string("transpose_52")];
            tensor<fp16, [1, 8, 188, 188]> matmul_15_1_cast_fp16 = matmul(transpose_x = matmul_15_transpose_x_1, transpose_y = matmul_15_transpose_y_1, x = mul_15_1_cast_fp16, y = transpose_94_cast_fp16)[name = string("matmul_15_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_15_1_cast_fp16 = add(x = matmul_15_1_cast_fp16, y = masked_fill_30_cast_fp16)[name = string("add_15_1_cast_fp16")];
            int32 softmax_15_axis_0 = const()[name = string("softmax_15_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_15_cast_fp16 = softmax(axis = softmax_15_axis_0, x = add_15_1_cast_fp16)[name = string("softmax_15_cast_fp16")];
            bool scaled_dot_product_attention_15_transpose_x_0 = const()[name = string("scaled_dot_product_attention_15_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_15_transpose_y_0 = const()[name = string("scaled_dot_product_attention_15_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_95_cast_fp16 = transpose(perm = transpose_95_perm_0, x = view_138_cast_fp16)[name = string("transpose_51")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_15_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_15_transpose_x_0, transpose_y = scaled_dot_product_attention_15_transpose_y_0, x = softmax_15_cast_fp16, y = transpose_95_cast_fp16)[name = string("scaled_dot_product_attention_15_cast_fp16")];
            tensor<int32, [4]> transpose_96_perm_0 = const()[name = string("transpose_96_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_1041 = const()[name = string("const_1041"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_96_cast_fp16 = transpose(perm = transpose_96_perm_0, x = scaled_dot_product_attention_15_cast_fp16)[name = string("transpose_50")];
            tensor<fp16, [1, 188, 1024]> view_144_cast_fp16 = reshape(shape = const_1041, x = transpose_96_cast_fp16)[name = string("view_144_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_15_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298373696))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299160192))))[name = string("p_encoder_layers_15_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_142_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_15_self_attn_o_proj_weight_to_fp16_palettized, x = view_144_cast_fp16)[name = string("linear_142_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_108_cast_fp16 = add(x = add_105_cast_fp16, y = linear_142_cast_fp16)[name = string("add_108_cast_fp16")];
            tensor<int32, [1]> layer_norm_77_axes_0 = const()[name = string("layer_norm_77_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_15_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_15_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299168448)))];
            tensor<fp16, [1024]> p_encoder_layers_15_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_15_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299170560)))];
            fp16 const_1043_to_fp16 = const()[name = string("const_1043_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_77_cast_fp16 = layer_norm(axes = layer_norm_77_axes_0, beta = p_encoder_layers_15_norm_conv_bias_to_fp16, epsilon = const_1043_to_fp16, gamma = p_encoder_layers_15_norm_conv_weight_to_fp16, x = add_108_cast_fp16)[name = string("layer_norm_77_cast_fp16")];
            tensor<int32, [3]> transpose_97_perm_0 = const()[name = string("transpose_97_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_45_pad_type_0 = const()[name = string("conv1d_45_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_45_strides_0 = const()[name = string("conv1d_45_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_45_pad_0 = const()[name = string("conv1d_45_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_45_dilations_0 = const()[name = string("conv1d_45_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_45_groups_0 = const()[name = string("conv1d_45_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_15_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299172672))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300745600))))[name = string("p_encoder_layers_15_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_97_cast_fp16 = transpose(perm = transpose_97_perm_0, x = layer_norm_77_cast_fp16)[name = string("transpose_49")];
            tensor<fp16, [1, 2048, 188]> conv1d_45_cast_fp16 = conv(dilations = conv1d_45_dilations_0, groups = conv1d_45_groups_0, pad = conv1d_45_pad_0, pad_type = conv1d_45_pad_type_0, strides = conv1d_45_strides_0, weight = p_encoder_layers_15_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_97_cast_fp16)[name = string("conv1d_45_cast_fp16")];
            int32 glu_15_split_num_splits_0 = const()[name = string("glu_15_split_num_splits_0"), val = int32(2)];
            int32 glu_15_split_axis_0 = const()[name = string("glu_15_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_15_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_15_split_cast_fp16_1 = split(axis = glu_15_split_axis_0, num_splits = glu_15_split_num_splits_0, x = conv1d_45_cast_fp16)[name = string("glu_15_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_15_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_15_split_cast_fp16_1)[name = string("glu_15_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_15_cast_fp16 = mul(x = glu_15_split_cast_fp16_0, y = glu_15_split_1_sigmoid_cast_fp16)[name = string("glu_15_cast_fp16")];
            fp16 const_1049_to_fp16 = const()[name = string("const_1049_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_31_cast_fp16 = select(a = const_1049_to_fp16, b = glu_15_cast_fp16, cond = all_1)[name = string("masked_fill_31_cast_fp16")];
            string conv1d_46_pad_type_0 = const()[name = string("conv1d_46_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_46_pad_0 = const()[name = string("conv1d_46_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_46_groups_0 = const()[name = string("conv1d_46_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_46_strides_0 = const()[name = string("conv1d_46_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_46_dilations_0 = const()[name = string("conv1d_46_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1577_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300762048))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300769024))))[name = string("const_1577_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1578_to_fp16 = const()[name = string("const_1578_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300777280)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_15_cast_fp16 = conv(bias = const_1578_to_fp16, dilations = conv1d_46_dilations_0, groups = conv1d_46_groups_0, pad = conv1d_46_pad_0, pad_type = conv1d_46_pad_type_0, strides = conv1d_46_strides_0, weight = const_1577_to_fp16_palettized, x = masked_fill_31_cast_fp16)[name = string("_native_batch_norm_legit_no_training_15_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_46_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_15_cast_fp16)[name = string("silu_46_cast_fp16")];
            string conv1d_47_pad_type_0 = const()[name = string("conv1d_47_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_47_strides_0 = const()[name = string("conv1d_47_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_47_pad_0 = const()[name = string("conv1d_47_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_47_dilations_0 = const()[name = string("conv1d_47_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_47_groups_0 = const()[name = string("conv1d_47_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_15_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300779392))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301565888))))[name = string("p_encoder_layers_15_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_47_cast_fp16 = conv(dilations = conv1d_47_dilations_0, groups = conv1d_47_groups_0, pad = conv1d_47_pad_0, pad_type = conv1d_47_pad_type_0, strides = conv1d_47_strides_0, weight = p_encoder_layers_15_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_46_cast_fp16)[name = string("conv1d_47_cast_fp16")];
            tensor<int32, [3]> transpose_98_perm_0 = const()[name = string("transpose_98_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_98_cast_fp16 = transpose(perm = transpose_98_perm_0, x = conv1d_47_cast_fp16)[name = string("transpose_48")];
            tensor<fp16, [1, 188, 1024]> add_109_cast_fp16 = add(x = add_108_cast_fp16, y = transpose_98_cast_fp16)[name = string("add_109_cast_fp16")];
            tensor<int32, [1]> layer_norm_78_axes_0 = const()[name = string("layer_norm_78_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_15_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_15_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301574144)))];
            tensor<fp16, [1024]> p_encoder_layers_15_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_15_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301576256)))];
            fp16 const_1060_to_fp16 = const()[name = string("const_1060_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_78_cast_fp16 = layer_norm(axes = layer_norm_78_axes_0, beta = p_encoder_layers_15_norm_feed_forward2_bias_to_fp16, epsilon = const_1060_to_fp16, gamma = p_encoder_layers_15_norm_feed_forward2_weight_to_fp16, x = add_109_cast_fp16)[name = string("layer_norm_78_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_15_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301578368))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(304724160))))[name = string("p_encoder_layers_15_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_143_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_15_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_78_cast_fp16)[name = string("linear_143_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_47_cast_fp16 = silu(x = linear_143_cast_fp16)[name = string("silu_47_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_15_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(304756992))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307902784))))[name = string("p_encoder_layers_15_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_144_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_15_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_47_cast_fp16)[name = string("linear_144_cast_fp16")];
            fp16 const_1062_to_fp16 = const()[name = string("const_1062_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_53_cast_fp16 = mul(x = linear_144_cast_fp16, y = const_1062_to_fp16)[name = string("mul_53_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_110_cast_fp16 = add(x = add_109_cast_fp16, y = mul_53_cast_fp16)[name = string("add_110_cast_fp16")];
            tensor<int32, [1]> layer_norm_79_axes_0 = const()[name = string("layer_norm_79_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_15_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_15_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307911040)))];
            tensor<fp16, [1024]> p_encoder_layers_15_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_15_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307913152)))];
            fp16 const_1064_to_fp16 = const()[name = string("const_1064_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_79_cast_fp16 = layer_norm(axes = layer_norm_79_axes_0, beta = p_encoder_layers_15_norm_out_bias_to_fp16, epsilon = const_1064_to_fp16, gamma = p_encoder_layers_15_norm_out_weight_to_fp16, x = add_110_cast_fp16)[name = string("layer_norm_79_cast_fp16")];
            tensor<int32, [1]> layer_norm_80_axes_0 = const()[name = string("layer_norm_80_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_16_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_16_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307915264)))];
            tensor<fp16, [1024]> p_encoder_layers_16_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_16_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307917376)))];
            fp16 const_1067_to_fp16 = const()[name = string("const_1067_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_80_cast_fp16 = layer_norm(axes = layer_norm_80_axes_0, beta = p_encoder_layers_16_norm_feed_forward1_bias_to_fp16, epsilon = const_1067_to_fp16, gamma = p_encoder_layers_16_norm_feed_forward1_weight_to_fp16, x = layer_norm_79_cast_fp16)[name = string("layer_norm_80_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_16_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307919488))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311065280))))[name = string("p_encoder_layers_16_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_145_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_16_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_80_cast_fp16)[name = string("linear_145_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_48_cast_fp16 = silu(x = linear_145_cast_fp16)[name = string("silu_48_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_16_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311098112))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314243904))))[name = string("p_encoder_layers_16_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_146_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_16_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_48_cast_fp16)[name = string("linear_146_cast_fp16")];
            fp16 const_1069_to_fp16 = const()[name = string("const_1069_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_54_cast_fp16 = mul(x = linear_146_cast_fp16, y = const_1069_to_fp16)[name = string("mul_54_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_111_cast_fp16 = add(x = layer_norm_79_cast_fp16, y = mul_54_cast_fp16)[name = string("add_111_cast_fp16")];
            tensor<int32, [1]> layer_norm_81_axes_0 = const()[name = string("layer_norm_81_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_16_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_16_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314252160)))];
            tensor<fp16, [1024]> p_encoder_layers_16_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_16_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314254272)))];
            fp16 const_1071_to_fp16 = const()[name = string("const_1071_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_81_cast_fp16 = layer_norm(axes = layer_norm_81_axes_0, beta = p_encoder_layers_16_norm_self_att_bias_to_fp16, epsilon = const_1071_to_fp16, gamma = p_encoder_layers_16_norm_self_att_weight_to_fp16, x = add_111_cast_fp16)[name = string("layer_norm_81_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_16_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314256384))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315042880))))[name = string("p_encoder_layers_16_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_147_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_16_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_81_cast_fp16)[name = string("linear_147_cast_fp16")];
            tensor<int32, [4]> const_1073 = const()[name = string("const_1073"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_145_cast_fp16 = reshape(shape = const_1073, x = linear_147_cast_fp16)[name = string("view_145_cast_fp16")];
            tensor<int32, [4]> transpose_99_perm_0 = const()[name = string("transpose_99_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_16_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315051136))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315837632))))[name = string("p_encoder_layers_16_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_148_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_16_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_81_cast_fp16)[name = string("linear_148_cast_fp16")];
            tensor<int32, [4]> const_1076 = const()[name = string("const_1076"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_146_cast_fp16 = reshape(shape = const_1076, x = linear_148_cast_fp16)[name = string("view_146_cast_fp16")];
            tensor<int32, [4]> transpose_100_perm_0 = const()[name = string("transpose_100_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_16_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315845888))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316632384))))[name = string("p_encoder_layers_16_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_149_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_16_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_81_cast_fp16)[name = string("linear_149_cast_fp16")];
            tensor<int32, [4]> const_1079 = const()[name = string("const_1079"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_147_cast_fp16 = reshape(shape = const_1079, x = linear_149_cast_fp16)[name = string("view_147_cast_fp16")];
            tensor<int32, [4]> transpose_101_perm_0 = const()[name = string("transpose_101_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_148_to_fp16 = const()[name = string("view_148_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316640640)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_99_cast_fp16 = transpose(perm = transpose_99_perm_0, x = view_145_cast_fp16)[name = string("transpose_47")];
            tensor<fp16, [1, 8, 188, 128]> add_112_cast_fp16 = add(x = transpose_99_cast_fp16, y = view_148_to_fp16)[name = string("add_112_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_149_to_fp16 = const()[name = string("view_149_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316642752)))];
            tensor<fp16, [1, 8, 188, 128]> add_113_cast_fp16 = add(x = transpose_99_cast_fp16, y = view_149_to_fp16)[name = string("add_113_cast_fp16")];
            bool matmul_17_transpose_x_0 = const()[name = string("matmul_17_transpose_x_0"), val = bool(false)];
            bool matmul_17_transpose_y_0 = const()[name = string("matmul_17_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_16_to_fp16 = const()[name = string("permute_16_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316644864)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_17_cast_fp16 = matmul(transpose_x = matmul_17_transpose_x_0, transpose_y = matmul_17_transpose_y_0, x = add_113_cast_fp16, y = permute_16_to_fp16)[name = string("matmul_17_cast_fp16")];
            tensor<int32, [8]> pad_16_pad_0 = const()[name = string("pad_16_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_16_mode_0 = const()[name = string("pad_16_mode_0"), val = string("constant")];
            fp16 const_1087_to_fp16 = const()[name = string("const_1087_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_16_cast_fp16 = pad(constant_val = const_1087_to_fp16, mode = pad_16_mode_0, pad = pad_16_pad_0, x = matmul_17_cast_fp16)[name = string("pad_16_cast_fp16")];
            tensor<int32, [4]> const_1088 = const()[name = string("const_1088"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_151_cast_fp16 = reshape(shape = const_1088, x = pad_16_cast_fp16)[name = string("view_151_cast_fp16")];
            tensor<int32, [4]> slice_33_begin_0 = const()[name = string("slice_33_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_33_end_0 = const()[name = string("slice_33_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_33_end_mask_0 = const()[name = string("slice_33_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_33_cast_fp16 = slice_by_index(begin = slice_33_begin_0, end = slice_33_end_0, end_mask = slice_33_end_mask_0, x = view_151_cast_fp16)[name = string("slice_33_cast_fp16")];
            tensor<int32, [4]> const_1092 = const()[name = string("const_1092"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_152_cast_fp16 = reshape(shape = const_1092, x = slice_33_cast_fp16)[name = string("view_152_cast_fp16")];
            tensor<int32, [4]> slice_34_begin_0 = const()[name = string("slice_34_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_34_end_0 = const()[name = string("slice_34_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_34_end_mask_0 = const()[name = string("slice_34_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_34_cast_fp16 = slice_by_index(begin = slice_34_begin_0, end = slice_34_end_0, end_mask = slice_34_end_mask_0, x = view_152_cast_fp16)[name = string("slice_34_cast_fp16")];
            fp16 const_1096_to_fp16 = const()[name = string("const_1096_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_55_cast_fp16 = mul(x = slice_34_cast_fp16, y = const_1096_to_fp16)[name = string("mul_55_cast_fp16")];
            fp16 const_1097_to_fp16 = const()[name = string("const_1097_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_32_cast_fp16 = select(a = const_1097_to_fp16, b = mul_55_cast_fp16, cond = logical_not)[name = string("masked_fill_32_cast_fp16")];
            fp16 const_1098_to_fp16 = const()[name = string("const_1098_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_16_1_cast_fp16 = mul(x = add_112_cast_fp16, y = const_1098_to_fp16)[name = string("mul_16_1_cast_fp16")];
            bool matmul_16_transpose_y_1 = const()[name = string("matmul_16_transpose_y_1"), val = bool(true)];
            bool matmul_16_transpose_x_1 = const()[name = string("matmul_16_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_100_cast_fp16 = transpose(perm = transpose_100_perm_0, x = view_146_cast_fp16)[name = string("transpose_46")];
            tensor<fp16, [1, 8, 188, 188]> matmul_16_1_cast_fp16 = matmul(transpose_x = matmul_16_transpose_x_1, transpose_y = matmul_16_transpose_y_1, x = mul_16_1_cast_fp16, y = transpose_100_cast_fp16)[name = string("matmul_16_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_16_1_cast_fp16 = add(x = matmul_16_1_cast_fp16, y = masked_fill_32_cast_fp16)[name = string("add_16_1_cast_fp16")];
            int32 softmax_16_axis_0 = const()[name = string("softmax_16_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_16_cast_fp16 = softmax(axis = softmax_16_axis_0, x = add_16_1_cast_fp16)[name = string("softmax_16_cast_fp16")];
            bool scaled_dot_product_attention_16_transpose_x_0 = const()[name = string("scaled_dot_product_attention_16_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_16_transpose_y_0 = const()[name = string("scaled_dot_product_attention_16_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_101_cast_fp16 = transpose(perm = transpose_101_perm_0, x = view_147_cast_fp16)[name = string("transpose_45")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_16_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_16_transpose_x_0, transpose_y = scaled_dot_product_attention_16_transpose_y_0, x = softmax_16_cast_fp16, y = transpose_101_cast_fp16)[name = string("scaled_dot_product_attention_16_cast_fp16")];
            tensor<int32, [4]> transpose_102_perm_0 = const()[name = string("transpose_102_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_1101 = const()[name = string("const_1101"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_102_cast_fp16 = transpose(perm = transpose_102_perm_0, x = scaled_dot_product_attention_16_cast_fp16)[name = string("transpose_44")];
            tensor<fp16, [1, 188, 1024]> view_153_cast_fp16 = reshape(shape = const_1101, x = transpose_102_cast_fp16)[name = string("view_153_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_16_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(317412928))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(318199424))))[name = string("p_encoder_layers_16_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_151_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_16_self_attn_o_proj_weight_to_fp16_palettized, x = view_153_cast_fp16)[name = string("linear_151_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_114_cast_fp16 = add(x = add_111_cast_fp16, y = linear_151_cast_fp16)[name = string("add_114_cast_fp16")];
            tensor<int32, [1]> layer_norm_82_axes_0 = const()[name = string("layer_norm_82_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_16_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_16_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(318207680)))];
            tensor<fp16, [1024]> p_encoder_layers_16_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_16_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(318209792)))];
            fp16 const_1103_to_fp16 = const()[name = string("const_1103_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_82_cast_fp16 = layer_norm(axes = layer_norm_82_axes_0, beta = p_encoder_layers_16_norm_conv_bias_to_fp16, epsilon = const_1103_to_fp16, gamma = p_encoder_layers_16_norm_conv_weight_to_fp16, x = add_114_cast_fp16)[name = string("layer_norm_82_cast_fp16")];
            tensor<int32, [3]> transpose_103_perm_0 = const()[name = string("transpose_103_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_48_pad_type_0 = const()[name = string("conv1d_48_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_48_strides_0 = const()[name = string("conv1d_48_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_48_pad_0 = const()[name = string("conv1d_48_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_48_dilations_0 = const()[name = string("conv1d_48_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_48_groups_0 = const()[name = string("conv1d_48_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_16_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(318211904))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(319784832))))[name = string("p_encoder_layers_16_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_103_cast_fp16 = transpose(perm = transpose_103_perm_0, x = layer_norm_82_cast_fp16)[name = string("transpose_43")];
            tensor<fp16, [1, 2048, 188]> conv1d_48_cast_fp16 = conv(dilations = conv1d_48_dilations_0, groups = conv1d_48_groups_0, pad = conv1d_48_pad_0, pad_type = conv1d_48_pad_type_0, strides = conv1d_48_strides_0, weight = p_encoder_layers_16_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_103_cast_fp16)[name = string("conv1d_48_cast_fp16")];
            int32 glu_16_split_num_splits_0 = const()[name = string("glu_16_split_num_splits_0"), val = int32(2)];
            int32 glu_16_split_axis_0 = const()[name = string("glu_16_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_16_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_16_split_cast_fp16_1 = split(axis = glu_16_split_axis_0, num_splits = glu_16_split_num_splits_0, x = conv1d_48_cast_fp16)[name = string("glu_16_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_16_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_16_split_cast_fp16_1)[name = string("glu_16_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_16_cast_fp16 = mul(x = glu_16_split_cast_fp16_0, y = glu_16_split_1_sigmoid_cast_fp16)[name = string("glu_16_cast_fp16")];
            fp16 const_1109_to_fp16 = const()[name = string("const_1109_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_33_cast_fp16 = select(a = const_1109_to_fp16, b = glu_16_cast_fp16, cond = all_1)[name = string("masked_fill_33_cast_fp16")];
            string conv1d_49_pad_type_0 = const()[name = string("conv1d_49_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_49_pad_0 = const()[name = string("conv1d_49_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_49_groups_0 = const()[name = string("conv1d_49_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_49_strides_0 = const()[name = string("conv1d_49_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_49_dilations_0 = const()[name = string("conv1d_49_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1579_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(319801280))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(319808256))))[name = string("const_1579_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1580_to_fp16 = const()[name = string("const_1580_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(319816512)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_16_cast_fp16 = conv(bias = const_1580_to_fp16, dilations = conv1d_49_dilations_0, groups = conv1d_49_groups_0, pad = conv1d_49_pad_0, pad_type = conv1d_49_pad_type_0, strides = conv1d_49_strides_0, weight = const_1579_to_fp16_palettized, x = masked_fill_33_cast_fp16)[name = string("_native_batch_norm_legit_no_training_16_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_49_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_16_cast_fp16)[name = string("silu_49_cast_fp16")];
            string conv1d_50_pad_type_0 = const()[name = string("conv1d_50_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_50_strides_0 = const()[name = string("conv1d_50_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_50_pad_0 = const()[name = string("conv1d_50_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_50_dilations_0 = const()[name = string("conv1d_50_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_50_groups_0 = const()[name = string("conv1d_50_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_16_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(319818624))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320605120))))[name = string("p_encoder_layers_16_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_50_cast_fp16 = conv(dilations = conv1d_50_dilations_0, groups = conv1d_50_groups_0, pad = conv1d_50_pad_0, pad_type = conv1d_50_pad_type_0, strides = conv1d_50_strides_0, weight = p_encoder_layers_16_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_49_cast_fp16)[name = string("conv1d_50_cast_fp16")];
            tensor<int32, [3]> transpose_104_perm_0 = const()[name = string("transpose_104_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_104_cast_fp16 = transpose(perm = transpose_104_perm_0, x = conv1d_50_cast_fp16)[name = string("transpose_42")];
            tensor<fp16, [1, 188, 1024]> add_115_cast_fp16 = add(x = add_114_cast_fp16, y = transpose_104_cast_fp16)[name = string("add_115_cast_fp16")];
            tensor<int32, [1]> layer_norm_83_axes_0 = const()[name = string("layer_norm_83_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_16_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_16_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320613376)))];
            tensor<fp16, [1024]> p_encoder_layers_16_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_16_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320615488)))];
            fp16 const_1120_to_fp16 = const()[name = string("const_1120_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_83_cast_fp16 = layer_norm(axes = layer_norm_83_axes_0, beta = p_encoder_layers_16_norm_feed_forward2_bias_to_fp16, epsilon = const_1120_to_fp16, gamma = p_encoder_layers_16_norm_feed_forward2_weight_to_fp16, x = add_115_cast_fp16)[name = string("layer_norm_83_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_16_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320617600))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(323763392))))[name = string("p_encoder_layers_16_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_152_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_16_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_83_cast_fp16)[name = string("linear_152_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_50_cast_fp16 = silu(x = linear_152_cast_fp16)[name = string("silu_50_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_16_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(323796224))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326942016))))[name = string("p_encoder_layers_16_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_153_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_16_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_50_cast_fp16)[name = string("linear_153_cast_fp16")];
            fp16 const_1122_to_fp16 = const()[name = string("const_1122_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_56_cast_fp16 = mul(x = linear_153_cast_fp16, y = const_1122_to_fp16)[name = string("mul_56_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_116_cast_fp16 = add(x = add_115_cast_fp16, y = mul_56_cast_fp16)[name = string("add_116_cast_fp16")];
            tensor<int32, [1]> layer_norm_84_axes_0 = const()[name = string("layer_norm_84_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_16_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_16_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326950272)))];
            tensor<fp16, [1024]> p_encoder_layers_16_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_16_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326952384)))];
            fp16 const_1124_to_fp16 = const()[name = string("const_1124_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_84_cast_fp16 = layer_norm(axes = layer_norm_84_axes_0, beta = p_encoder_layers_16_norm_out_bias_to_fp16, epsilon = const_1124_to_fp16, gamma = p_encoder_layers_16_norm_out_weight_to_fp16, x = add_116_cast_fp16)[name = string("layer_norm_84_cast_fp16")];
            tensor<int32, [1]> layer_norm_85_axes_0 = const()[name = string("layer_norm_85_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_17_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_17_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326954496)))];
            tensor<fp16, [1024]> p_encoder_layers_17_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_17_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326956608)))];
            fp16 const_1127_to_fp16 = const()[name = string("const_1127_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_85_cast_fp16 = layer_norm(axes = layer_norm_85_axes_0, beta = p_encoder_layers_17_norm_feed_forward1_bias_to_fp16, epsilon = const_1127_to_fp16, gamma = p_encoder_layers_17_norm_feed_forward1_weight_to_fp16, x = layer_norm_84_cast_fp16)[name = string("layer_norm_85_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_17_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326958720))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(330104512))))[name = string("p_encoder_layers_17_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_154_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_17_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_85_cast_fp16)[name = string("linear_154_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_51_cast_fp16 = silu(x = linear_154_cast_fp16)[name = string("silu_51_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_17_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(330137344))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(333283136))))[name = string("p_encoder_layers_17_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_155_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_17_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_51_cast_fp16)[name = string("linear_155_cast_fp16")];
            fp16 const_1129_to_fp16 = const()[name = string("const_1129_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_57_cast_fp16 = mul(x = linear_155_cast_fp16, y = const_1129_to_fp16)[name = string("mul_57_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_117_cast_fp16 = add(x = layer_norm_84_cast_fp16, y = mul_57_cast_fp16)[name = string("add_117_cast_fp16")];
            tensor<int32, [1]> layer_norm_86_axes_0 = const()[name = string("layer_norm_86_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_17_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_17_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(333291392)))];
            tensor<fp16, [1024]> p_encoder_layers_17_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_17_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(333293504)))];
            fp16 const_1131_to_fp16 = const()[name = string("const_1131_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_86_cast_fp16 = layer_norm(axes = layer_norm_86_axes_0, beta = p_encoder_layers_17_norm_self_att_bias_to_fp16, epsilon = const_1131_to_fp16, gamma = p_encoder_layers_17_norm_self_att_weight_to_fp16, x = add_117_cast_fp16)[name = string("layer_norm_86_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_17_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(333295616))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(334082112))))[name = string("p_encoder_layers_17_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_156_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_17_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_86_cast_fp16)[name = string("linear_156_cast_fp16")];
            tensor<int32, [4]> const_1133 = const()[name = string("const_1133"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_154_cast_fp16 = reshape(shape = const_1133, x = linear_156_cast_fp16)[name = string("view_154_cast_fp16")];
            tensor<int32, [4]> transpose_105_perm_0 = const()[name = string("transpose_105_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_17_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(334090368))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(334876864))))[name = string("p_encoder_layers_17_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_157_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_17_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_86_cast_fp16)[name = string("linear_157_cast_fp16")];
            tensor<int32, [4]> const_1136 = const()[name = string("const_1136"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_155_cast_fp16 = reshape(shape = const_1136, x = linear_157_cast_fp16)[name = string("view_155_cast_fp16")];
            tensor<int32, [4]> transpose_106_perm_0 = const()[name = string("transpose_106_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_17_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(334885120))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(335671616))))[name = string("p_encoder_layers_17_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_158_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_17_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_86_cast_fp16)[name = string("linear_158_cast_fp16")];
            tensor<int32, [4]> const_1139 = const()[name = string("const_1139"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_156_cast_fp16 = reshape(shape = const_1139, x = linear_158_cast_fp16)[name = string("view_156_cast_fp16")];
            tensor<int32, [4]> transpose_107_perm_0 = const()[name = string("transpose_107_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_157_to_fp16 = const()[name = string("view_157_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(335679872)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_105_cast_fp16 = transpose(perm = transpose_105_perm_0, x = view_154_cast_fp16)[name = string("transpose_41")];
            tensor<fp16, [1, 8, 188, 128]> add_118_cast_fp16 = add(x = transpose_105_cast_fp16, y = view_157_to_fp16)[name = string("add_118_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_158_to_fp16 = const()[name = string("view_158_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(335681984)))];
            tensor<fp16, [1, 8, 188, 128]> add_119_cast_fp16 = add(x = transpose_105_cast_fp16, y = view_158_to_fp16)[name = string("add_119_cast_fp16")];
            bool matmul_18_transpose_x_0 = const()[name = string("matmul_18_transpose_x_0"), val = bool(false)];
            bool matmul_18_transpose_y_0 = const()[name = string("matmul_18_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_17_to_fp16 = const()[name = string("permute_17_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(335684096)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_18_cast_fp16 = matmul(transpose_x = matmul_18_transpose_x_0, transpose_y = matmul_18_transpose_y_0, x = add_119_cast_fp16, y = permute_17_to_fp16)[name = string("matmul_18_cast_fp16")];
            tensor<int32, [8]> pad_17_pad_0 = const()[name = string("pad_17_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_17_mode_0 = const()[name = string("pad_17_mode_0"), val = string("constant")];
            fp16 const_1147_to_fp16 = const()[name = string("const_1147_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_17_cast_fp16 = pad(constant_val = const_1147_to_fp16, mode = pad_17_mode_0, pad = pad_17_pad_0, x = matmul_18_cast_fp16)[name = string("pad_17_cast_fp16")];
            tensor<int32, [4]> const_1148 = const()[name = string("const_1148"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_160_cast_fp16 = reshape(shape = const_1148, x = pad_17_cast_fp16)[name = string("view_160_cast_fp16")];
            tensor<int32, [4]> slice_35_begin_0 = const()[name = string("slice_35_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_35_end_0 = const()[name = string("slice_35_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_35_end_mask_0 = const()[name = string("slice_35_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_35_cast_fp16 = slice_by_index(begin = slice_35_begin_0, end = slice_35_end_0, end_mask = slice_35_end_mask_0, x = view_160_cast_fp16)[name = string("slice_35_cast_fp16")];
            tensor<int32, [4]> const_1152 = const()[name = string("const_1152"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_161_cast_fp16 = reshape(shape = const_1152, x = slice_35_cast_fp16)[name = string("view_161_cast_fp16")];
            tensor<int32, [4]> slice_36_begin_0 = const()[name = string("slice_36_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_36_end_0 = const()[name = string("slice_36_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_36_end_mask_0 = const()[name = string("slice_36_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_36_cast_fp16 = slice_by_index(begin = slice_36_begin_0, end = slice_36_end_0, end_mask = slice_36_end_mask_0, x = view_161_cast_fp16)[name = string("slice_36_cast_fp16")];
            fp16 const_1156_to_fp16 = const()[name = string("const_1156_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_58_cast_fp16 = mul(x = slice_36_cast_fp16, y = const_1156_to_fp16)[name = string("mul_58_cast_fp16")];
            fp16 const_1157_to_fp16 = const()[name = string("const_1157_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_34_cast_fp16 = select(a = const_1157_to_fp16, b = mul_58_cast_fp16, cond = logical_not)[name = string("masked_fill_34_cast_fp16")];
            fp16 const_1158_to_fp16 = const()[name = string("const_1158_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_17_1_cast_fp16 = mul(x = add_118_cast_fp16, y = const_1158_to_fp16)[name = string("mul_17_1_cast_fp16")];
            bool matmul_17_transpose_y_1 = const()[name = string("matmul_17_transpose_y_1"), val = bool(true)];
            bool matmul_17_transpose_x_1 = const()[name = string("matmul_17_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_106_cast_fp16 = transpose(perm = transpose_106_perm_0, x = view_155_cast_fp16)[name = string("transpose_40")];
            tensor<fp16, [1, 8, 188, 188]> matmul_17_1_cast_fp16 = matmul(transpose_x = matmul_17_transpose_x_1, transpose_y = matmul_17_transpose_y_1, x = mul_17_1_cast_fp16, y = transpose_106_cast_fp16)[name = string("matmul_17_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_17_1_cast_fp16 = add(x = matmul_17_1_cast_fp16, y = masked_fill_34_cast_fp16)[name = string("add_17_1_cast_fp16")];
            int32 softmax_17_axis_0 = const()[name = string("softmax_17_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_17_cast_fp16 = softmax(axis = softmax_17_axis_0, x = add_17_1_cast_fp16)[name = string("softmax_17_cast_fp16")];
            bool scaled_dot_product_attention_17_transpose_x_0 = const()[name = string("scaled_dot_product_attention_17_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_17_transpose_y_0 = const()[name = string("scaled_dot_product_attention_17_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_107_cast_fp16 = transpose(perm = transpose_107_perm_0, x = view_156_cast_fp16)[name = string("transpose_39")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_17_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_17_transpose_x_0, transpose_y = scaled_dot_product_attention_17_transpose_y_0, x = softmax_17_cast_fp16, y = transpose_107_cast_fp16)[name = string("scaled_dot_product_attention_17_cast_fp16")];
            tensor<int32, [4]> transpose_108_perm_0 = const()[name = string("transpose_108_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_1161 = const()[name = string("const_1161"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_108_cast_fp16 = transpose(perm = transpose_108_perm_0, x = scaled_dot_product_attention_17_cast_fp16)[name = string("transpose_38")];
            tensor<fp16, [1, 188, 1024]> view_162_cast_fp16 = reshape(shape = const_1161, x = transpose_108_cast_fp16)[name = string("view_162_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_17_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(336452160))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337238656))))[name = string("p_encoder_layers_17_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_160_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_17_self_attn_o_proj_weight_to_fp16_palettized, x = view_162_cast_fp16)[name = string("linear_160_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_120_cast_fp16 = add(x = add_117_cast_fp16, y = linear_160_cast_fp16)[name = string("add_120_cast_fp16")];
            tensor<int32, [1]> layer_norm_87_axes_0 = const()[name = string("layer_norm_87_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_17_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_17_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337246912)))];
            tensor<fp16, [1024]> p_encoder_layers_17_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_17_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337249024)))];
            fp16 const_1163_to_fp16 = const()[name = string("const_1163_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_87_cast_fp16 = layer_norm(axes = layer_norm_87_axes_0, beta = p_encoder_layers_17_norm_conv_bias_to_fp16, epsilon = const_1163_to_fp16, gamma = p_encoder_layers_17_norm_conv_weight_to_fp16, x = add_120_cast_fp16)[name = string("layer_norm_87_cast_fp16")];
            tensor<int32, [3]> transpose_109_perm_0 = const()[name = string("transpose_109_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_51_pad_type_0 = const()[name = string("conv1d_51_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_51_strides_0 = const()[name = string("conv1d_51_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_51_pad_0 = const()[name = string("conv1d_51_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_51_dilations_0 = const()[name = string("conv1d_51_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_51_groups_0 = const()[name = string("conv1d_51_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_17_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337251136))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(338824064))))[name = string("p_encoder_layers_17_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_109_cast_fp16 = transpose(perm = transpose_109_perm_0, x = layer_norm_87_cast_fp16)[name = string("transpose_37")];
            tensor<fp16, [1, 2048, 188]> conv1d_51_cast_fp16 = conv(dilations = conv1d_51_dilations_0, groups = conv1d_51_groups_0, pad = conv1d_51_pad_0, pad_type = conv1d_51_pad_type_0, strides = conv1d_51_strides_0, weight = p_encoder_layers_17_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_109_cast_fp16)[name = string("conv1d_51_cast_fp16")];
            int32 glu_17_split_num_splits_0 = const()[name = string("glu_17_split_num_splits_0"), val = int32(2)];
            int32 glu_17_split_axis_0 = const()[name = string("glu_17_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_17_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_17_split_cast_fp16_1 = split(axis = glu_17_split_axis_0, num_splits = glu_17_split_num_splits_0, x = conv1d_51_cast_fp16)[name = string("glu_17_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_17_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_17_split_cast_fp16_1)[name = string("glu_17_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_17_cast_fp16 = mul(x = glu_17_split_cast_fp16_0, y = glu_17_split_1_sigmoid_cast_fp16)[name = string("glu_17_cast_fp16")];
            fp16 const_1169_to_fp16 = const()[name = string("const_1169_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_35_cast_fp16 = select(a = const_1169_to_fp16, b = glu_17_cast_fp16, cond = all_1)[name = string("masked_fill_35_cast_fp16")];
            string conv1d_52_pad_type_0 = const()[name = string("conv1d_52_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_52_pad_0 = const()[name = string("conv1d_52_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_52_groups_0 = const()[name = string("conv1d_52_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_52_strides_0 = const()[name = string("conv1d_52_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_52_dilations_0 = const()[name = string("conv1d_52_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1581_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(338840512))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(338847488))))[name = string("const_1581_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1582_to_fp16 = const()[name = string("const_1582_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(338855744)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_17_cast_fp16 = conv(bias = const_1582_to_fp16, dilations = conv1d_52_dilations_0, groups = conv1d_52_groups_0, pad = conv1d_52_pad_0, pad_type = conv1d_52_pad_type_0, strides = conv1d_52_strides_0, weight = const_1581_to_fp16_palettized, x = masked_fill_35_cast_fp16)[name = string("_native_batch_norm_legit_no_training_17_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_52_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_17_cast_fp16)[name = string("silu_52_cast_fp16")];
            string conv1d_53_pad_type_0 = const()[name = string("conv1d_53_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_53_strides_0 = const()[name = string("conv1d_53_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_53_pad_0 = const()[name = string("conv1d_53_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_53_dilations_0 = const()[name = string("conv1d_53_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_53_groups_0 = const()[name = string("conv1d_53_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_17_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(338857856))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339644352))))[name = string("p_encoder_layers_17_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_53_cast_fp16 = conv(dilations = conv1d_53_dilations_0, groups = conv1d_53_groups_0, pad = conv1d_53_pad_0, pad_type = conv1d_53_pad_type_0, strides = conv1d_53_strides_0, weight = p_encoder_layers_17_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_52_cast_fp16)[name = string("conv1d_53_cast_fp16")];
            tensor<int32, [3]> transpose_110_perm_0 = const()[name = string("transpose_110_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_110_cast_fp16 = transpose(perm = transpose_110_perm_0, x = conv1d_53_cast_fp16)[name = string("transpose_36")];
            tensor<fp16, [1, 188, 1024]> add_121_cast_fp16 = add(x = add_120_cast_fp16, y = transpose_110_cast_fp16)[name = string("add_121_cast_fp16")];
            tensor<int32, [1]> layer_norm_88_axes_0 = const()[name = string("layer_norm_88_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_17_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_17_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339652608)))];
            tensor<fp16, [1024]> p_encoder_layers_17_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_17_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339654720)))];
            fp16 const_1180_to_fp16 = const()[name = string("const_1180_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_88_cast_fp16 = layer_norm(axes = layer_norm_88_axes_0, beta = p_encoder_layers_17_norm_feed_forward2_bias_to_fp16, epsilon = const_1180_to_fp16, gamma = p_encoder_layers_17_norm_feed_forward2_weight_to_fp16, x = add_121_cast_fp16)[name = string("layer_norm_88_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_17_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339656832))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342802624))))[name = string("p_encoder_layers_17_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_161_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_17_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_88_cast_fp16)[name = string("linear_161_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_53_cast_fp16 = silu(x = linear_161_cast_fp16)[name = string("silu_53_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_17_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342835456))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345981248))))[name = string("p_encoder_layers_17_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_162_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_17_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_53_cast_fp16)[name = string("linear_162_cast_fp16")];
            fp16 const_1182_to_fp16 = const()[name = string("const_1182_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_59_cast_fp16 = mul(x = linear_162_cast_fp16, y = const_1182_to_fp16)[name = string("mul_59_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_122_cast_fp16 = add(x = add_121_cast_fp16, y = mul_59_cast_fp16)[name = string("add_122_cast_fp16")];
            tensor<int32, [1]> layer_norm_89_axes_0 = const()[name = string("layer_norm_89_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_17_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_17_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345989504)))];
            tensor<fp16, [1024]> p_encoder_layers_17_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_17_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345991616)))];
            fp16 const_1184_to_fp16 = const()[name = string("const_1184_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_89_cast_fp16 = layer_norm(axes = layer_norm_89_axes_0, beta = p_encoder_layers_17_norm_out_bias_to_fp16, epsilon = const_1184_to_fp16, gamma = p_encoder_layers_17_norm_out_weight_to_fp16, x = add_122_cast_fp16)[name = string("layer_norm_89_cast_fp16")];
            tensor<int32, [1]> layer_norm_90_axes_0 = const()[name = string("layer_norm_90_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_18_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_18_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345993728)))];
            tensor<fp16, [1024]> p_encoder_layers_18_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_18_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345995840)))];
            fp16 const_1187_to_fp16 = const()[name = string("const_1187_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_90_cast_fp16 = layer_norm(axes = layer_norm_90_axes_0, beta = p_encoder_layers_18_norm_feed_forward1_bias_to_fp16, epsilon = const_1187_to_fp16, gamma = p_encoder_layers_18_norm_feed_forward1_weight_to_fp16, x = layer_norm_89_cast_fp16)[name = string("layer_norm_90_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_18_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345997952))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(349143744))))[name = string("p_encoder_layers_18_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_163_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_18_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_90_cast_fp16)[name = string("linear_163_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_54_cast_fp16 = silu(x = linear_163_cast_fp16)[name = string("silu_54_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_18_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(349176576))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(352322368))))[name = string("p_encoder_layers_18_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_164_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_18_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_54_cast_fp16)[name = string("linear_164_cast_fp16")];
            fp16 const_1189_to_fp16 = const()[name = string("const_1189_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_60_cast_fp16 = mul(x = linear_164_cast_fp16, y = const_1189_to_fp16)[name = string("mul_60_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_123_cast_fp16 = add(x = layer_norm_89_cast_fp16, y = mul_60_cast_fp16)[name = string("add_123_cast_fp16")];
            tensor<int32, [1]> layer_norm_91_axes_0 = const()[name = string("layer_norm_91_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_18_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_18_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(352330624)))];
            tensor<fp16, [1024]> p_encoder_layers_18_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_18_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(352332736)))];
            fp16 const_1191_to_fp16 = const()[name = string("const_1191_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_91_cast_fp16 = layer_norm(axes = layer_norm_91_axes_0, beta = p_encoder_layers_18_norm_self_att_bias_to_fp16, epsilon = const_1191_to_fp16, gamma = p_encoder_layers_18_norm_self_att_weight_to_fp16, x = add_123_cast_fp16)[name = string("layer_norm_91_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_18_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(352334848))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(353121344))))[name = string("p_encoder_layers_18_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_165_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_18_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_91_cast_fp16)[name = string("linear_165_cast_fp16")];
            tensor<int32, [4]> const_1193 = const()[name = string("const_1193"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_163_cast_fp16 = reshape(shape = const_1193, x = linear_165_cast_fp16)[name = string("view_163_cast_fp16")];
            tensor<int32, [4]> transpose_111_perm_0 = const()[name = string("transpose_111_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_18_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(353129600))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(353916096))))[name = string("p_encoder_layers_18_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_166_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_18_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_91_cast_fp16)[name = string("linear_166_cast_fp16")];
            tensor<int32, [4]> const_1196 = const()[name = string("const_1196"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_164_cast_fp16 = reshape(shape = const_1196, x = linear_166_cast_fp16)[name = string("view_164_cast_fp16")];
            tensor<int32, [4]> transpose_112_perm_0 = const()[name = string("transpose_112_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_18_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(353924352))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354710848))))[name = string("p_encoder_layers_18_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_167_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_18_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_91_cast_fp16)[name = string("linear_167_cast_fp16")];
            tensor<int32, [4]> const_1199 = const()[name = string("const_1199"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_165_cast_fp16 = reshape(shape = const_1199, x = linear_167_cast_fp16)[name = string("view_165_cast_fp16")];
            tensor<int32, [4]> transpose_113_perm_0 = const()[name = string("transpose_113_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_166_to_fp16 = const()[name = string("view_166_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354719104)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_111_cast_fp16 = transpose(perm = transpose_111_perm_0, x = view_163_cast_fp16)[name = string("transpose_35")];
            tensor<fp16, [1, 8, 188, 128]> add_124_cast_fp16 = add(x = transpose_111_cast_fp16, y = view_166_to_fp16)[name = string("add_124_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_167_to_fp16 = const()[name = string("view_167_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354721216)))];
            tensor<fp16, [1, 8, 188, 128]> add_125_cast_fp16 = add(x = transpose_111_cast_fp16, y = view_167_to_fp16)[name = string("add_125_cast_fp16")];
            bool matmul_19_transpose_x_0 = const()[name = string("matmul_19_transpose_x_0"), val = bool(false)];
            bool matmul_19_transpose_y_0 = const()[name = string("matmul_19_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_18_to_fp16 = const()[name = string("permute_18_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354723328)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_19_cast_fp16 = matmul(transpose_x = matmul_19_transpose_x_0, transpose_y = matmul_19_transpose_y_0, x = add_125_cast_fp16, y = permute_18_to_fp16)[name = string("matmul_19_cast_fp16")];
            tensor<int32, [8]> pad_18_pad_0 = const()[name = string("pad_18_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_18_mode_0 = const()[name = string("pad_18_mode_0"), val = string("constant")];
            fp16 const_1207_to_fp16 = const()[name = string("const_1207_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_18_cast_fp16 = pad(constant_val = const_1207_to_fp16, mode = pad_18_mode_0, pad = pad_18_pad_0, x = matmul_19_cast_fp16)[name = string("pad_18_cast_fp16")];
            tensor<int32, [4]> const_1208 = const()[name = string("const_1208"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_169_cast_fp16 = reshape(shape = const_1208, x = pad_18_cast_fp16)[name = string("view_169_cast_fp16")];
            tensor<int32, [4]> slice_37_begin_0 = const()[name = string("slice_37_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_37_end_0 = const()[name = string("slice_37_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_37_end_mask_0 = const()[name = string("slice_37_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_37_cast_fp16 = slice_by_index(begin = slice_37_begin_0, end = slice_37_end_0, end_mask = slice_37_end_mask_0, x = view_169_cast_fp16)[name = string("slice_37_cast_fp16")];
            tensor<int32, [4]> const_1212 = const()[name = string("const_1212"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_170_cast_fp16 = reshape(shape = const_1212, x = slice_37_cast_fp16)[name = string("view_170_cast_fp16")];
            tensor<int32, [4]> slice_38_begin_0 = const()[name = string("slice_38_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_38_end_0 = const()[name = string("slice_38_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_38_end_mask_0 = const()[name = string("slice_38_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_38_cast_fp16 = slice_by_index(begin = slice_38_begin_0, end = slice_38_end_0, end_mask = slice_38_end_mask_0, x = view_170_cast_fp16)[name = string("slice_38_cast_fp16")];
            fp16 const_1216_to_fp16 = const()[name = string("const_1216_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_61_cast_fp16 = mul(x = slice_38_cast_fp16, y = const_1216_to_fp16)[name = string("mul_61_cast_fp16")];
            fp16 const_1217_to_fp16 = const()[name = string("const_1217_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_36_cast_fp16 = select(a = const_1217_to_fp16, b = mul_61_cast_fp16, cond = logical_not)[name = string("masked_fill_36_cast_fp16")];
            fp16 const_1218_to_fp16 = const()[name = string("const_1218_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_18_1_cast_fp16 = mul(x = add_124_cast_fp16, y = const_1218_to_fp16)[name = string("mul_18_1_cast_fp16")];
            bool matmul_18_transpose_y_1 = const()[name = string("matmul_18_transpose_y_1"), val = bool(true)];
            bool matmul_18_transpose_x_1 = const()[name = string("matmul_18_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_112_cast_fp16 = transpose(perm = transpose_112_perm_0, x = view_164_cast_fp16)[name = string("transpose_34")];
            tensor<fp16, [1, 8, 188, 188]> matmul_18_1_cast_fp16 = matmul(transpose_x = matmul_18_transpose_x_1, transpose_y = matmul_18_transpose_y_1, x = mul_18_1_cast_fp16, y = transpose_112_cast_fp16)[name = string("matmul_18_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_18_1_cast_fp16 = add(x = matmul_18_1_cast_fp16, y = masked_fill_36_cast_fp16)[name = string("add_18_1_cast_fp16")];
            int32 softmax_18_axis_0 = const()[name = string("softmax_18_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_18_cast_fp16 = softmax(axis = softmax_18_axis_0, x = add_18_1_cast_fp16)[name = string("softmax_18_cast_fp16")];
            bool scaled_dot_product_attention_18_transpose_x_0 = const()[name = string("scaled_dot_product_attention_18_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_18_transpose_y_0 = const()[name = string("scaled_dot_product_attention_18_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_113_cast_fp16 = transpose(perm = transpose_113_perm_0, x = view_165_cast_fp16)[name = string("transpose_33")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_18_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_18_transpose_x_0, transpose_y = scaled_dot_product_attention_18_transpose_y_0, x = softmax_18_cast_fp16, y = transpose_113_cast_fp16)[name = string("scaled_dot_product_attention_18_cast_fp16")];
            tensor<int32, [4]> transpose_114_perm_0 = const()[name = string("transpose_114_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_1221 = const()[name = string("const_1221"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_114_cast_fp16 = transpose(perm = transpose_114_perm_0, x = scaled_dot_product_attention_18_cast_fp16)[name = string("transpose_32")];
            tensor<fp16, [1, 188, 1024]> view_171_cast_fp16 = reshape(shape = const_1221, x = transpose_114_cast_fp16)[name = string("view_171_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_18_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355491392))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(356277888))))[name = string("p_encoder_layers_18_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_169_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_18_self_attn_o_proj_weight_to_fp16_palettized, x = view_171_cast_fp16)[name = string("linear_169_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_126_cast_fp16 = add(x = add_123_cast_fp16, y = linear_169_cast_fp16)[name = string("add_126_cast_fp16")];
            tensor<int32, [1]> layer_norm_92_axes_0 = const()[name = string("layer_norm_92_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_18_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_18_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(356286144)))];
            tensor<fp16, [1024]> p_encoder_layers_18_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_18_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(356288256)))];
            fp16 const_1223_to_fp16 = const()[name = string("const_1223_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_92_cast_fp16 = layer_norm(axes = layer_norm_92_axes_0, beta = p_encoder_layers_18_norm_conv_bias_to_fp16, epsilon = const_1223_to_fp16, gamma = p_encoder_layers_18_norm_conv_weight_to_fp16, x = add_126_cast_fp16)[name = string("layer_norm_92_cast_fp16")];
            tensor<int32, [3]> transpose_115_perm_0 = const()[name = string("transpose_115_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_54_pad_type_0 = const()[name = string("conv1d_54_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_54_strides_0 = const()[name = string("conv1d_54_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_54_pad_0 = const()[name = string("conv1d_54_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_54_dilations_0 = const()[name = string("conv1d_54_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_54_groups_0 = const()[name = string("conv1d_54_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_18_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(356290368))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(357863296))))[name = string("p_encoder_layers_18_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_115_cast_fp16 = transpose(perm = transpose_115_perm_0, x = layer_norm_92_cast_fp16)[name = string("transpose_31")];
            tensor<fp16, [1, 2048, 188]> conv1d_54_cast_fp16 = conv(dilations = conv1d_54_dilations_0, groups = conv1d_54_groups_0, pad = conv1d_54_pad_0, pad_type = conv1d_54_pad_type_0, strides = conv1d_54_strides_0, weight = p_encoder_layers_18_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_115_cast_fp16)[name = string("conv1d_54_cast_fp16")];
            int32 glu_18_split_num_splits_0 = const()[name = string("glu_18_split_num_splits_0"), val = int32(2)];
            int32 glu_18_split_axis_0 = const()[name = string("glu_18_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_18_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_18_split_cast_fp16_1 = split(axis = glu_18_split_axis_0, num_splits = glu_18_split_num_splits_0, x = conv1d_54_cast_fp16)[name = string("glu_18_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_18_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_18_split_cast_fp16_1)[name = string("glu_18_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_18_cast_fp16 = mul(x = glu_18_split_cast_fp16_0, y = glu_18_split_1_sigmoid_cast_fp16)[name = string("glu_18_cast_fp16")];
            fp16 const_1229_to_fp16 = const()[name = string("const_1229_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_37_cast_fp16 = select(a = const_1229_to_fp16, b = glu_18_cast_fp16, cond = all_1)[name = string("masked_fill_37_cast_fp16")];
            string conv1d_55_pad_type_0 = const()[name = string("conv1d_55_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_55_pad_0 = const()[name = string("conv1d_55_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_55_groups_0 = const()[name = string("conv1d_55_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_55_strides_0 = const()[name = string("conv1d_55_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_55_dilations_0 = const()[name = string("conv1d_55_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1583_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(357879744))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(357886720))))[name = string("const_1583_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1584_to_fp16 = const()[name = string("const_1584_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(357894976)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_18_cast_fp16 = conv(bias = const_1584_to_fp16, dilations = conv1d_55_dilations_0, groups = conv1d_55_groups_0, pad = conv1d_55_pad_0, pad_type = conv1d_55_pad_type_0, strides = conv1d_55_strides_0, weight = const_1583_to_fp16_palettized, x = masked_fill_37_cast_fp16)[name = string("_native_batch_norm_legit_no_training_18_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_55_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_18_cast_fp16)[name = string("silu_55_cast_fp16")];
            string conv1d_56_pad_type_0 = const()[name = string("conv1d_56_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_56_strides_0 = const()[name = string("conv1d_56_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_56_pad_0 = const()[name = string("conv1d_56_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_56_dilations_0 = const()[name = string("conv1d_56_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_56_groups_0 = const()[name = string("conv1d_56_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_18_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(357897088))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(358683584))))[name = string("p_encoder_layers_18_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_56_cast_fp16 = conv(dilations = conv1d_56_dilations_0, groups = conv1d_56_groups_0, pad = conv1d_56_pad_0, pad_type = conv1d_56_pad_type_0, strides = conv1d_56_strides_0, weight = p_encoder_layers_18_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_55_cast_fp16)[name = string("conv1d_56_cast_fp16")];
            tensor<int32, [3]> transpose_116_perm_0 = const()[name = string("transpose_116_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_116_cast_fp16 = transpose(perm = transpose_116_perm_0, x = conv1d_56_cast_fp16)[name = string("transpose_30")];
            tensor<fp16, [1, 188, 1024]> add_127_cast_fp16 = add(x = add_126_cast_fp16, y = transpose_116_cast_fp16)[name = string("add_127_cast_fp16")];
            tensor<int32, [1]> layer_norm_93_axes_0 = const()[name = string("layer_norm_93_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_18_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_18_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(358691840)))];
            tensor<fp16, [1024]> p_encoder_layers_18_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_18_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(358693952)))];
            fp16 const_1240_to_fp16 = const()[name = string("const_1240_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_93_cast_fp16 = layer_norm(axes = layer_norm_93_axes_0, beta = p_encoder_layers_18_norm_feed_forward2_bias_to_fp16, epsilon = const_1240_to_fp16, gamma = p_encoder_layers_18_norm_feed_forward2_weight_to_fp16, x = add_127_cast_fp16)[name = string("layer_norm_93_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_18_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(358696064))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(361841856))))[name = string("p_encoder_layers_18_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_170_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_18_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_93_cast_fp16)[name = string("linear_170_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_56_cast_fp16 = silu(x = linear_170_cast_fp16)[name = string("silu_56_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_18_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(361874688))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(365020480))))[name = string("p_encoder_layers_18_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_171_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_18_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_56_cast_fp16)[name = string("linear_171_cast_fp16")];
            fp16 const_1242_to_fp16 = const()[name = string("const_1242_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_62_cast_fp16 = mul(x = linear_171_cast_fp16, y = const_1242_to_fp16)[name = string("mul_62_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_128_cast_fp16 = add(x = add_127_cast_fp16, y = mul_62_cast_fp16)[name = string("add_128_cast_fp16")];
            tensor<int32, [1]> layer_norm_94_axes_0 = const()[name = string("layer_norm_94_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_18_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_18_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(365028736)))];
            tensor<fp16, [1024]> p_encoder_layers_18_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_18_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(365030848)))];
            fp16 const_1244_to_fp16 = const()[name = string("const_1244_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_94_cast_fp16 = layer_norm(axes = layer_norm_94_axes_0, beta = p_encoder_layers_18_norm_out_bias_to_fp16, epsilon = const_1244_to_fp16, gamma = p_encoder_layers_18_norm_out_weight_to_fp16, x = add_128_cast_fp16)[name = string("layer_norm_94_cast_fp16")];
            tensor<int32, [1]> layer_norm_95_axes_0 = const()[name = string("layer_norm_95_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_19_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_19_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(365032960)))];
            tensor<fp16, [1024]> p_encoder_layers_19_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_19_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(365035072)))];
            fp16 const_1247_to_fp16 = const()[name = string("const_1247_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_95_cast_fp16 = layer_norm(axes = layer_norm_95_axes_0, beta = p_encoder_layers_19_norm_feed_forward1_bias_to_fp16, epsilon = const_1247_to_fp16, gamma = p_encoder_layers_19_norm_feed_forward1_weight_to_fp16, x = layer_norm_94_cast_fp16)[name = string("layer_norm_95_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_19_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(365037184))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368182976))))[name = string("p_encoder_layers_19_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_172_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_19_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_95_cast_fp16)[name = string("linear_172_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_57_cast_fp16 = silu(x = linear_172_cast_fp16)[name = string("silu_57_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_19_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368215808))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(371361600))))[name = string("p_encoder_layers_19_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_173_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_19_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_57_cast_fp16)[name = string("linear_173_cast_fp16")];
            fp16 const_1249_to_fp16 = const()[name = string("const_1249_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_63_cast_fp16 = mul(x = linear_173_cast_fp16, y = const_1249_to_fp16)[name = string("mul_63_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_129_cast_fp16 = add(x = layer_norm_94_cast_fp16, y = mul_63_cast_fp16)[name = string("add_129_cast_fp16")];
            tensor<int32, [1]> layer_norm_96_axes_0 = const()[name = string("layer_norm_96_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_19_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_19_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(371369856)))];
            tensor<fp16, [1024]> p_encoder_layers_19_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_19_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(371371968)))];
            fp16 const_1251_to_fp16 = const()[name = string("const_1251_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_96_cast_fp16 = layer_norm(axes = layer_norm_96_axes_0, beta = p_encoder_layers_19_norm_self_att_bias_to_fp16, epsilon = const_1251_to_fp16, gamma = p_encoder_layers_19_norm_self_att_weight_to_fp16, x = add_129_cast_fp16)[name = string("layer_norm_96_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_19_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(371374080))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(372160576))))[name = string("p_encoder_layers_19_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_174_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_19_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_96_cast_fp16)[name = string("linear_174_cast_fp16")];
            tensor<int32, [4]> const_1253 = const()[name = string("const_1253"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_172_cast_fp16 = reshape(shape = const_1253, x = linear_174_cast_fp16)[name = string("view_172_cast_fp16")];
            tensor<int32, [4]> transpose_117_perm_0 = const()[name = string("transpose_117_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_19_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(372168832))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(372955328))))[name = string("p_encoder_layers_19_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_175_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_19_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_96_cast_fp16)[name = string("linear_175_cast_fp16")];
            tensor<int32, [4]> const_1256 = const()[name = string("const_1256"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_173_cast_fp16 = reshape(shape = const_1256, x = linear_175_cast_fp16)[name = string("view_173_cast_fp16")];
            tensor<int32, [4]> transpose_118_perm_0 = const()[name = string("transpose_118_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_19_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(372963584))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(373750080))))[name = string("p_encoder_layers_19_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_176_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_19_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_96_cast_fp16)[name = string("linear_176_cast_fp16")];
            tensor<int32, [4]> const_1259 = const()[name = string("const_1259"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_174_cast_fp16 = reshape(shape = const_1259, x = linear_176_cast_fp16)[name = string("view_174_cast_fp16")];
            tensor<int32, [4]> transpose_119_perm_0 = const()[name = string("transpose_119_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_175_to_fp16 = const()[name = string("view_175_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(373758336)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_117_cast_fp16 = transpose(perm = transpose_117_perm_0, x = view_172_cast_fp16)[name = string("transpose_29")];
            tensor<fp16, [1, 8, 188, 128]> add_130_cast_fp16 = add(x = transpose_117_cast_fp16, y = view_175_to_fp16)[name = string("add_130_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_176_to_fp16 = const()[name = string("view_176_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(373760448)))];
            tensor<fp16, [1, 8, 188, 128]> add_131_cast_fp16 = add(x = transpose_117_cast_fp16, y = view_176_to_fp16)[name = string("add_131_cast_fp16")];
            bool matmul_20_transpose_x_0 = const()[name = string("matmul_20_transpose_x_0"), val = bool(false)];
            bool matmul_20_transpose_y_0 = const()[name = string("matmul_20_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_19_to_fp16 = const()[name = string("permute_19_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(373762560)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_20_cast_fp16 = matmul(transpose_x = matmul_20_transpose_x_0, transpose_y = matmul_20_transpose_y_0, x = add_131_cast_fp16, y = permute_19_to_fp16)[name = string("matmul_20_cast_fp16")];
            tensor<int32, [8]> pad_19_pad_0 = const()[name = string("pad_19_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_19_mode_0 = const()[name = string("pad_19_mode_0"), val = string("constant")];
            fp16 const_1267_to_fp16 = const()[name = string("const_1267_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_19_cast_fp16 = pad(constant_val = const_1267_to_fp16, mode = pad_19_mode_0, pad = pad_19_pad_0, x = matmul_20_cast_fp16)[name = string("pad_19_cast_fp16")];
            tensor<int32, [4]> const_1268 = const()[name = string("const_1268"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_178_cast_fp16 = reshape(shape = const_1268, x = pad_19_cast_fp16)[name = string("view_178_cast_fp16")];
            tensor<int32, [4]> slice_39_begin_0 = const()[name = string("slice_39_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_39_end_0 = const()[name = string("slice_39_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_39_end_mask_0 = const()[name = string("slice_39_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_39_cast_fp16 = slice_by_index(begin = slice_39_begin_0, end = slice_39_end_0, end_mask = slice_39_end_mask_0, x = view_178_cast_fp16)[name = string("slice_39_cast_fp16")];
            tensor<int32, [4]> const_1272 = const()[name = string("const_1272"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_179_cast_fp16 = reshape(shape = const_1272, x = slice_39_cast_fp16)[name = string("view_179_cast_fp16")];
            tensor<int32, [4]> slice_40_begin_0 = const()[name = string("slice_40_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_40_end_0 = const()[name = string("slice_40_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_40_end_mask_0 = const()[name = string("slice_40_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_40_cast_fp16 = slice_by_index(begin = slice_40_begin_0, end = slice_40_end_0, end_mask = slice_40_end_mask_0, x = view_179_cast_fp16)[name = string("slice_40_cast_fp16")];
            fp16 const_1276_to_fp16 = const()[name = string("const_1276_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_64_cast_fp16 = mul(x = slice_40_cast_fp16, y = const_1276_to_fp16)[name = string("mul_64_cast_fp16")];
            fp16 const_1277_to_fp16 = const()[name = string("const_1277_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_38_cast_fp16 = select(a = const_1277_to_fp16, b = mul_64_cast_fp16, cond = logical_not)[name = string("masked_fill_38_cast_fp16")];
            fp16 const_1278_to_fp16 = const()[name = string("const_1278_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_19_1_cast_fp16 = mul(x = add_130_cast_fp16, y = const_1278_to_fp16)[name = string("mul_19_1_cast_fp16")];
            bool matmul_19_transpose_y_1 = const()[name = string("matmul_19_transpose_y_1"), val = bool(true)];
            bool matmul_19_transpose_x_1 = const()[name = string("matmul_19_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_118_cast_fp16 = transpose(perm = transpose_118_perm_0, x = view_173_cast_fp16)[name = string("transpose_28")];
            tensor<fp16, [1, 8, 188, 188]> matmul_19_1_cast_fp16 = matmul(transpose_x = matmul_19_transpose_x_1, transpose_y = matmul_19_transpose_y_1, x = mul_19_1_cast_fp16, y = transpose_118_cast_fp16)[name = string("matmul_19_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_19_1_cast_fp16 = add(x = matmul_19_1_cast_fp16, y = masked_fill_38_cast_fp16)[name = string("add_19_1_cast_fp16")];
            int32 softmax_19_axis_0 = const()[name = string("softmax_19_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_19_cast_fp16 = softmax(axis = softmax_19_axis_0, x = add_19_1_cast_fp16)[name = string("softmax_19_cast_fp16")];
            bool scaled_dot_product_attention_19_transpose_x_0 = const()[name = string("scaled_dot_product_attention_19_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_19_transpose_y_0 = const()[name = string("scaled_dot_product_attention_19_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_119_cast_fp16 = transpose(perm = transpose_119_perm_0, x = view_174_cast_fp16)[name = string("transpose_27")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_19_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_19_transpose_x_0, transpose_y = scaled_dot_product_attention_19_transpose_y_0, x = softmax_19_cast_fp16, y = transpose_119_cast_fp16)[name = string("scaled_dot_product_attention_19_cast_fp16")];
            tensor<int32, [4]> transpose_120_perm_0 = const()[name = string("transpose_120_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_1281 = const()[name = string("const_1281"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_120_cast_fp16 = transpose(perm = transpose_120_perm_0, x = scaled_dot_product_attention_19_cast_fp16)[name = string("transpose_26")];
            tensor<fp16, [1, 188, 1024]> view_180_cast_fp16 = reshape(shape = const_1281, x = transpose_120_cast_fp16)[name = string("view_180_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_19_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(374530624))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(375317120))))[name = string("p_encoder_layers_19_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_178_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_19_self_attn_o_proj_weight_to_fp16_palettized, x = view_180_cast_fp16)[name = string("linear_178_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_132_cast_fp16 = add(x = add_129_cast_fp16, y = linear_178_cast_fp16)[name = string("add_132_cast_fp16")];
            tensor<int32, [1]> layer_norm_97_axes_0 = const()[name = string("layer_norm_97_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_19_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_19_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(375325376)))];
            tensor<fp16, [1024]> p_encoder_layers_19_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_19_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(375327488)))];
            fp16 const_1283_to_fp16 = const()[name = string("const_1283_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_97_cast_fp16 = layer_norm(axes = layer_norm_97_axes_0, beta = p_encoder_layers_19_norm_conv_bias_to_fp16, epsilon = const_1283_to_fp16, gamma = p_encoder_layers_19_norm_conv_weight_to_fp16, x = add_132_cast_fp16)[name = string("layer_norm_97_cast_fp16")];
            tensor<int32, [3]> transpose_121_perm_0 = const()[name = string("transpose_121_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_57_pad_type_0 = const()[name = string("conv1d_57_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_57_strides_0 = const()[name = string("conv1d_57_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_57_pad_0 = const()[name = string("conv1d_57_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_57_dilations_0 = const()[name = string("conv1d_57_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_57_groups_0 = const()[name = string("conv1d_57_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_19_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(375329600))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(376902528))))[name = string("p_encoder_layers_19_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_121_cast_fp16 = transpose(perm = transpose_121_perm_0, x = layer_norm_97_cast_fp16)[name = string("transpose_25")];
            tensor<fp16, [1, 2048, 188]> conv1d_57_cast_fp16 = conv(dilations = conv1d_57_dilations_0, groups = conv1d_57_groups_0, pad = conv1d_57_pad_0, pad_type = conv1d_57_pad_type_0, strides = conv1d_57_strides_0, weight = p_encoder_layers_19_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_121_cast_fp16)[name = string("conv1d_57_cast_fp16")];
            int32 glu_19_split_num_splits_0 = const()[name = string("glu_19_split_num_splits_0"), val = int32(2)];
            int32 glu_19_split_axis_0 = const()[name = string("glu_19_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_19_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_19_split_cast_fp16_1 = split(axis = glu_19_split_axis_0, num_splits = glu_19_split_num_splits_0, x = conv1d_57_cast_fp16)[name = string("glu_19_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_19_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_19_split_cast_fp16_1)[name = string("glu_19_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_19_cast_fp16 = mul(x = glu_19_split_cast_fp16_0, y = glu_19_split_1_sigmoid_cast_fp16)[name = string("glu_19_cast_fp16")];
            fp16 const_1289_to_fp16 = const()[name = string("const_1289_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_39_cast_fp16 = select(a = const_1289_to_fp16, b = glu_19_cast_fp16, cond = all_1)[name = string("masked_fill_39_cast_fp16")];
            string conv1d_58_pad_type_0 = const()[name = string("conv1d_58_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_58_pad_0 = const()[name = string("conv1d_58_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_58_groups_0 = const()[name = string("conv1d_58_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_58_strides_0 = const()[name = string("conv1d_58_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_58_dilations_0 = const()[name = string("conv1d_58_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1585_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(376918976))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(376925952))))[name = string("const_1585_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1586_to_fp16 = const()[name = string("const_1586_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(376934208)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_19_cast_fp16 = conv(bias = const_1586_to_fp16, dilations = conv1d_58_dilations_0, groups = conv1d_58_groups_0, pad = conv1d_58_pad_0, pad_type = conv1d_58_pad_type_0, strides = conv1d_58_strides_0, weight = const_1585_to_fp16_palettized, x = masked_fill_39_cast_fp16)[name = string("_native_batch_norm_legit_no_training_19_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_58_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_19_cast_fp16)[name = string("silu_58_cast_fp16")];
            string conv1d_59_pad_type_0 = const()[name = string("conv1d_59_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_59_strides_0 = const()[name = string("conv1d_59_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_59_pad_0 = const()[name = string("conv1d_59_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_59_dilations_0 = const()[name = string("conv1d_59_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_59_groups_0 = const()[name = string("conv1d_59_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_19_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(376936320))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(377722816))))[name = string("p_encoder_layers_19_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_59_cast_fp16 = conv(dilations = conv1d_59_dilations_0, groups = conv1d_59_groups_0, pad = conv1d_59_pad_0, pad_type = conv1d_59_pad_type_0, strides = conv1d_59_strides_0, weight = p_encoder_layers_19_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_58_cast_fp16)[name = string("conv1d_59_cast_fp16")];
            tensor<int32, [3]> transpose_122_perm_0 = const()[name = string("transpose_122_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_122_cast_fp16 = transpose(perm = transpose_122_perm_0, x = conv1d_59_cast_fp16)[name = string("transpose_24")];
            tensor<fp16, [1, 188, 1024]> add_133_cast_fp16 = add(x = add_132_cast_fp16, y = transpose_122_cast_fp16)[name = string("add_133_cast_fp16")];
            tensor<int32, [1]> layer_norm_98_axes_0 = const()[name = string("layer_norm_98_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_19_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_19_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(377731072)))];
            tensor<fp16, [1024]> p_encoder_layers_19_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_19_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(377733184)))];
            fp16 const_1300_to_fp16 = const()[name = string("const_1300_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_98_cast_fp16 = layer_norm(axes = layer_norm_98_axes_0, beta = p_encoder_layers_19_norm_feed_forward2_bias_to_fp16, epsilon = const_1300_to_fp16, gamma = p_encoder_layers_19_norm_feed_forward2_weight_to_fp16, x = add_133_cast_fp16)[name = string("layer_norm_98_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_19_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(377735296))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(380881088))))[name = string("p_encoder_layers_19_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_179_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_19_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_98_cast_fp16)[name = string("linear_179_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_59_cast_fp16 = silu(x = linear_179_cast_fp16)[name = string("silu_59_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_19_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(380913920))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(384059712))))[name = string("p_encoder_layers_19_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_180_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_19_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_59_cast_fp16)[name = string("linear_180_cast_fp16")];
            fp16 const_1302_to_fp16 = const()[name = string("const_1302_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_65_cast_fp16 = mul(x = linear_180_cast_fp16, y = const_1302_to_fp16)[name = string("mul_65_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_134_cast_fp16 = add(x = add_133_cast_fp16, y = mul_65_cast_fp16)[name = string("add_134_cast_fp16")];
            tensor<int32, [1]> layer_norm_99_axes_0 = const()[name = string("layer_norm_99_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_19_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_19_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(384067968)))];
            tensor<fp16, [1024]> p_encoder_layers_19_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_19_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(384070080)))];
            fp16 const_1304_to_fp16 = const()[name = string("const_1304_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_99_cast_fp16 = layer_norm(axes = layer_norm_99_axes_0, beta = p_encoder_layers_19_norm_out_bias_to_fp16, epsilon = const_1304_to_fp16, gamma = p_encoder_layers_19_norm_out_weight_to_fp16, x = add_134_cast_fp16)[name = string("layer_norm_99_cast_fp16")];
            tensor<int32, [1]> layer_norm_100_axes_0 = const()[name = string("layer_norm_100_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_20_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_20_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(384072192)))];
            tensor<fp16, [1024]> p_encoder_layers_20_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_20_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(384074304)))];
            fp16 const_1307_to_fp16 = const()[name = string("const_1307_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_100_cast_fp16 = layer_norm(axes = layer_norm_100_axes_0, beta = p_encoder_layers_20_norm_feed_forward1_bias_to_fp16, epsilon = const_1307_to_fp16, gamma = p_encoder_layers_20_norm_feed_forward1_weight_to_fp16, x = layer_norm_99_cast_fp16)[name = string("layer_norm_100_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_20_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(384076416))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(387222208))))[name = string("p_encoder_layers_20_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_181_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_20_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_100_cast_fp16)[name = string("linear_181_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_60_cast_fp16 = silu(x = linear_181_cast_fp16)[name = string("silu_60_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_20_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(387255040))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(390400832))))[name = string("p_encoder_layers_20_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_182_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_20_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_60_cast_fp16)[name = string("linear_182_cast_fp16")];
            fp16 const_1309_to_fp16 = const()[name = string("const_1309_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_66_cast_fp16 = mul(x = linear_182_cast_fp16, y = const_1309_to_fp16)[name = string("mul_66_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_135_cast_fp16 = add(x = layer_norm_99_cast_fp16, y = mul_66_cast_fp16)[name = string("add_135_cast_fp16")];
            tensor<int32, [1]> layer_norm_101_axes_0 = const()[name = string("layer_norm_101_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_20_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_20_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(390409088)))];
            tensor<fp16, [1024]> p_encoder_layers_20_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_20_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(390411200)))];
            fp16 const_1311_to_fp16 = const()[name = string("const_1311_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_101_cast_fp16 = layer_norm(axes = layer_norm_101_axes_0, beta = p_encoder_layers_20_norm_self_att_bias_to_fp16, epsilon = const_1311_to_fp16, gamma = p_encoder_layers_20_norm_self_att_weight_to_fp16, x = add_135_cast_fp16)[name = string("layer_norm_101_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_20_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(390413312))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(391199808))))[name = string("p_encoder_layers_20_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_183_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_20_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_101_cast_fp16)[name = string("linear_183_cast_fp16")];
            tensor<int32, [4]> const_1313 = const()[name = string("const_1313"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_181_cast_fp16 = reshape(shape = const_1313, x = linear_183_cast_fp16)[name = string("view_181_cast_fp16")];
            tensor<int32, [4]> transpose_123_perm_0 = const()[name = string("transpose_123_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_20_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(391208064))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(391994560))))[name = string("p_encoder_layers_20_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_184_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_20_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_101_cast_fp16)[name = string("linear_184_cast_fp16")];
            tensor<int32, [4]> const_1316 = const()[name = string("const_1316"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_182_cast_fp16 = reshape(shape = const_1316, x = linear_184_cast_fp16)[name = string("view_182_cast_fp16")];
            tensor<int32, [4]> transpose_124_perm_0 = const()[name = string("transpose_124_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_20_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(392002816))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(392789312))))[name = string("p_encoder_layers_20_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_185_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_20_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_101_cast_fp16)[name = string("linear_185_cast_fp16")];
            tensor<int32, [4]> const_1319 = const()[name = string("const_1319"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_183_cast_fp16 = reshape(shape = const_1319, x = linear_185_cast_fp16)[name = string("view_183_cast_fp16")];
            tensor<int32, [4]> transpose_125_perm_0 = const()[name = string("transpose_125_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_184_to_fp16 = const()[name = string("view_184_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(392797568)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_123_cast_fp16 = transpose(perm = transpose_123_perm_0, x = view_181_cast_fp16)[name = string("transpose_23")];
            tensor<fp16, [1, 8, 188, 128]> add_136_cast_fp16 = add(x = transpose_123_cast_fp16, y = view_184_to_fp16)[name = string("add_136_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_185_to_fp16 = const()[name = string("view_185_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(392799680)))];
            tensor<fp16, [1, 8, 188, 128]> add_137_cast_fp16 = add(x = transpose_123_cast_fp16, y = view_185_to_fp16)[name = string("add_137_cast_fp16")];
            bool matmul_21_transpose_x_0 = const()[name = string("matmul_21_transpose_x_0"), val = bool(false)];
            bool matmul_21_transpose_y_0 = const()[name = string("matmul_21_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_20_to_fp16 = const()[name = string("permute_20_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(392801792)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_21_cast_fp16 = matmul(transpose_x = matmul_21_transpose_x_0, transpose_y = matmul_21_transpose_y_0, x = add_137_cast_fp16, y = permute_20_to_fp16)[name = string("matmul_21_cast_fp16")];
            tensor<int32, [8]> pad_20_pad_0 = const()[name = string("pad_20_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_20_mode_0 = const()[name = string("pad_20_mode_0"), val = string("constant")];
            fp16 const_1327_to_fp16 = const()[name = string("const_1327_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_20_cast_fp16 = pad(constant_val = const_1327_to_fp16, mode = pad_20_mode_0, pad = pad_20_pad_0, x = matmul_21_cast_fp16)[name = string("pad_20_cast_fp16")];
            tensor<int32, [4]> const_1328 = const()[name = string("const_1328"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_187_cast_fp16 = reshape(shape = const_1328, x = pad_20_cast_fp16)[name = string("view_187_cast_fp16")];
            tensor<int32, [4]> slice_41_begin_0 = const()[name = string("slice_41_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_41_end_0 = const()[name = string("slice_41_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_41_end_mask_0 = const()[name = string("slice_41_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_41_cast_fp16 = slice_by_index(begin = slice_41_begin_0, end = slice_41_end_0, end_mask = slice_41_end_mask_0, x = view_187_cast_fp16)[name = string("slice_41_cast_fp16")];
            tensor<int32, [4]> const_1332 = const()[name = string("const_1332"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_188_cast_fp16 = reshape(shape = const_1332, x = slice_41_cast_fp16)[name = string("view_188_cast_fp16")];
            tensor<int32, [4]> slice_42_begin_0 = const()[name = string("slice_42_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_42_end_0 = const()[name = string("slice_42_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_42_end_mask_0 = const()[name = string("slice_42_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_42_cast_fp16 = slice_by_index(begin = slice_42_begin_0, end = slice_42_end_0, end_mask = slice_42_end_mask_0, x = view_188_cast_fp16)[name = string("slice_42_cast_fp16")];
            fp16 const_1336_to_fp16 = const()[name = string("const_1336_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_67_cast_fp16 = mul(x = slice_42_cast_fp16, y = const_1336_to_fp16)[name = string("mul_67_cast_fp16")];
            fp16 const_1337_to_fp16 = const()[name = string("const_1337_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_40_cast_fp16 = select(a = const_1337_to_fp16, b = mul_67_cast_fp16, cond = logical_not)[name = string("masked_fill_40_cast_fp16")];
            fp16 const_1338_to_fp16 = const()[name = string("const_1338_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_20_1_cast_fp16 = mul(x = add_136_cast_fp16, y = const_1338_to_fp16)[name = string("mul_20_1_cast_fp16")];
            bool matmul_20_transpose_y_1 = const()[name = string("matmul_20_transpose_y_1"), val = bool(true)];
            bool matmul_20_transpose_x_1 = const()[name = string("matmul_20_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_124_cast_fp16 = transpose(perm = transpose_124_perm_0, x = view_182_cast_fp16)[name = string("transpose_22")];
            tensor<fp16, [1, 8, 188, 188]> matmul_20_1_cast_fp16 = matmul(transpose_x = matmul_20_transpose_x_1, transpose_y = matmul_20_transpose_y_1, x = mul_20_1_cast_fp16, y = transpose_124_cast_fp16)[name = string("matmul_20_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_20_1_cast_fp16 = add(x = matmul_20_1_cast_fp16, y = masked_fill_40_cast_fp16)[name = string("add_20_1_cast_fp16")];
            int32 softmax_20_axis_0 = const()[name = string("softmax_20_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_20_cast_fp16 = softmax(axis = softmax_20_axis_0, x = add_20_1_cast_fp16)[name = string("softmax_20_cast_fp16")];
            bool scaled_dot_product_attention_20_transpose_x_0 = const()[name = string("scaled_dot_product_attention_20_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_20_transpose_y_0 = const()[name = string("scaled_dot_product_attention_20_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_125_cast_fp16 = transpose(perm = transpose_125_perm_0, x = view_183_cast_fp16)[name = string("transpose_21")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_20_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_20_transpose_x_0, transpose_y = scaled_dot_product_attention_20_transpose_y_0, x = softmax_20_cast_fp16, y = transpose_125_cast_fp16)[name = string("scaled_dot_product_attention_20_cast_fp16")];
            tensor<int32, [4]> transpose_126_perm_0 = const()[name = string("transpose_126_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_1341 = const()[name = string("const_1341"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_126_cast_fp16 = transpose(perm = transpose_126_perm_0, x = scaled_dot_product_attention_20_cast_fp16)[name = string("transpose_20")];
            tensor<fp16, [1, 188, 1024]> view_189_cast_fp16 = reshape(shape = const_1341, x = transpose_126_cast_fp16)[name = string("view_189_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_20_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(393569856))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(394356352))))[name = string("p_encoder_layers_20_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_187_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_20_self_attn_o_proj_weight_to_fp16_palettized, x = view_189_cast_fp16)[name = string("linear_187_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_138_cast_fp16 = add(x = add_135_cast_fp16, y = linear_187_cast_fp16)[name = string("add_138_cast_fp16")];
            tensor<int32, [1]> layer_norm_102_axes_0 = const()[name = string("layer_norm_102_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_20_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_20_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(394364608)))];
            tensor<fp16, [1024]> p_encoder_layers_20_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_20_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(394366720)))];
            fp16 const_1343_to_fp16 = const()[name = string("const_1343_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_102_cast_fp16 = layer_norm(axes = layer_norm_102_axes_0, beta = p_encoder_layers_20_norm_conv_bias_to_fp16, epsilon = const_1343_to_fp16, gamma = p_encoder_layers_20_norm_conv_weight_to_fp16, x = add_138_cast_fp16)[name = string("layer_norm_102_cast_fp16")];
            tensor<int32, [3]> transpose_127_perm_0 = const()[name = string("transpose_127_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_60_pad_type_0 = const()[name = string("conv1d_60_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_60_strides_0 = const()[name = string("conv1d_60_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_60_pad_0 = const()[name = string("conv1d_60_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_60_dilations_0 = const()[name = string("conv1d_60_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_60_groups_0 = const()[name = string("conv1d_60_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_20_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(394368832))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395941760))))[name = string("p_encoder_layers_20_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_127_cast_fp16 = transpose(perm = transpose_127_perm_0, x = layer_norm_102_cast_fp16)[name = string("transpose_19")];
            tensor<fp16, [1, 2048, 188]> conv1d_60_cast_fp16 = conv(dilations = conv1d_60_dilations_0, groups = conv1d_60_groups_0, pad = conv1d_60_pad_0, pad_type = conv1d_60_pad_type_0, strides = conv1d_60_strides_0, weight = p_encoder_layers_20_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_127_cast_fp16)[name = string("conv1d_60_cast_fp16")];
            int32 glu_20_split_num_splits_0 = const()[name = string("glu_20_split_num_splits_0"), val = int32(2)];
            int32 glu_20_split_axis_0 = const()[name = string("glu_20_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_20_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_20_split_cast_fp16_1 = split(axis = glu_20_split_axis_0, num_splits = glu_20_split_num_splits_0, x = conv1d_60_cast_fp16)[name = string("glu_20_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_20_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_20_split_cast_fp16_1)[name = string("glu_20_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_20_cast_fp16 = mul(x = glu_20_split_cast_fp16_0, y = glu_20_split_1_sigmoid_cast_fp16)[name = string("glu_20_cast_fp16")];
            fp16 const_1349_to_fp16 = const()[name = string("const_1349_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_41_cast_fp16 = select(a = const_1349_to_fp16, b = glu_20_cast_fp16, cond = all_1)[name = string("masked_fill_41_cast_fp16")];
            string conv1d_61_pad_type_0 = const()[name = string("conv1d_61_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_61_pad_0 = const()[name = string("conv1d_61_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_61_groups_0 = const()[name = string("conv1d_61_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_61_strides_0 = const()[name = string("conv1d_61_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_61_dilations_0 = const()[name = string("conv1d_61_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1587_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395958208))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395965184))))[name = string("const_1587_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1588_to_fp16 = const()[name = string("const_1588_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395973440)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_20_cast_fp16 = conv(bias = const_1588_to_fp16, dilations = conv1d_61_dilations_0, groups = conv1d_61_groups_0, pad = conv1d_61_pad_0, pad_type = conv1d_61_pad_type_0, strides = conv1d_61_strides_0, weight = const_1587_to_fp16_palettized, x = masked_fill_41_cast_fp16)[name = string("_native_batch_norm_legit_no_training_20_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_61_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_20_cast_fp16)[name = string("silu_61_cast_fp16")];
            string conv1d_62_pad_type_0 = const()[name = string("conv1d_62_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_62_strides_0 = const()[name = string("conv1d_62_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_62_pad_0 = const()[name = string("conv1d_62_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_62_dilations_0 = const()[name = string("conv1d_62_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_62_groups_0 = const()[name = string("conv1d_62_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_20_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395975552))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396762048))))[name = string("p_encoder_layers_20_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_62_cast_fp16 = conv(dilations = conv1d_62_dilations_0, groups = conv1d_62_groups_0, pad = conv1d_62_pad_0, pad_type = conv1d_62_pad_type_0, strides = conv1d_62_strides_0, weight = p_encoder_layers_20_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_61_cast_fp16)[name = string("conv1d_62_cast_fp16")];
            tensor<int32, [3]> transpose_128_perm_0 = const()[name = string("transpose_128_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_128_cast_fp16 = transpose(perm = transpose_128_perm_0, x = conv1d_62_cast_fp16)[name = string("transpose_18")];
            tensor<fp16, [1, 188, 1024]> add_139_cast_fp16 = add(x = add_138_cast_fp16, y = transpose_128_cast_fp16)[name = string("add_139_cast_fp16")];
            tensor<int32, [1]> layer_norm_103_axes_0 = const()[name = string("layer_norm_103_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_20_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_20_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396770304)))];
            tensor<fp16, [1024]> p_encoder_layers_20_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_20_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396772416)))];
            fp16 const_1360_to_fp16 = const()[name = string("const_1360_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_103_cast_fp16 = layer_norm(axes = layer_norm_103_axes_0, beta = p_encoder_layers_20_norm_feed_forward2_bias_to_fp16, epsilon = const_1360_to_fp16, gamma = p_encoder_layers_20_norm_feed_forward2_weight_to_fp16, x = add_139_cast_fp16)[name = string("layer_norm_103_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_20_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396774528))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(399920320))))[name = string("p_encoder_layers_20_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_188_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_20_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_103_cast_fp16)[name = string("linear_188_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_62_cast_fp16 = silu(x = linear_188_cast_fp16)[name = string("silu_62_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_20_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(399953152))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(403098944))))[name = string("p_encoder_layers_20_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_189_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_20_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_62_cast_fp16)[name = string("linear_189_cast_fp16")];
            fp16 const_1362_to_fp16 = const()[name = string("const_1362_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_68_cast_fp16 = mul(x = linear_189_cast_fp16, y = const_1362_to_fp16)[name = string("mul_68_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_140_cast_fp16 = add(x = add_139_cast_fp16, y = mul_68_cast_fp16)[name = string("add_140_cast_fp16")];
            tensor<int32, [1]> layer_norm_104_axes_0 = const()[name = string("layer_norm_104_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_20_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_20_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(403107200)))];
            tensor<fp16, [1024]> p_encoder_layers_20_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_20_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(403109312)))];
            fp16 const_1364_to_fp16 = const()[name = string("const_1364_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_104_cast_fp16 = layer_norm(axes = layer_norm_104_axes_0, beta = p_encoder_layers_20_norm_out_bias_to_fp16, epsilon = const_1364_to_fp16, gamma = p_encoder_layers_20_norm_out_weight_to_fp16, x = add_140_cast_fp16)[name = string("layer_norm_104_cast_fp16")];
            tensor<int32, [1]> layer_norm_105_axes_0 = const()[name = string("layer_norm_105_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_21_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_21_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(403111424)))];
            tensor<fp16, [1024]> p_encoder_layers_21_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_21_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(403113536)))];
            fp16 const_1367_to_fp16 = const()[name = string("const_1367_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_105_cast_fp16 = layer_norm(axes = layer_norm_105_axes_0, beta = p_encoder_layers_21_norm_feed_forward1_bias_to_fp16, epsilon = const_1367_to_fp16, gamma = p_encoder_layers_21_norm_feed_forward1_weight_to_fp16, x = layer_norm_104_cast_fp16)[name = string("layer_norm_105_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_21_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(403115648))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(406261440))))[name = string("p_encoder_layers_21_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_190_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_21_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_105_cast_fp16)[name = string("linear_190_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_63_cast_fp16 = silu(x = linear_190_cast_fp16)[name = string("silu_63_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_21_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(406294272))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(409440064))))[name = string("p_encoder_layers_21_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_191_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_21_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_63_cast_fp16)[name = string("linear_191_cast_fp16")];
            fp16 const_1369_to_fp16 = const()[name = string("const_1369_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_69_cast_fp16 = mul(x = linear_191_cast_fp16, y = const_1369_to_fp16)[name = string("mul_69_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_141_cast_fp16 = add(x = layer_norm_104_cast_fp16, y = mul_69_cast_fp16)[name = string("add_141_cast_fp16")];
            tensor<int32, [1]> layer_norm_106_axes_0 = const()[name = string("layer_norm_106_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_21_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_21_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(409448320)))];
            tensor<fp16, [1024]> p_encoder_layers_21_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_21_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(409450432)))];
            fp16 const_1371_to_fp16 = const()[name = string("const_1371_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_106_cast_fp16 = layer_norm(axes = layer_norm_106_axes_0, beta = p_encoder_layers_21_norm_self_att_bias_to_fp16, epsilon = const_1371_to_fp16, gamma = p_encoder_layers_21_norm_self_att_weight_to_fp16, x = add_141_cast_fp16)[name = string("layer_norm_106_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_21_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(409452544))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(410239040))))[name = string("p_encoder_layers_21_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_192_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_21_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_106_cast_fp16)[name = string("linear_192_cast_fp16")];
            tensor<int32, [4]> const_1373 = const()[name = string("const_1373"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_190_cast_fp16 = reshape(shape = const_1373, x = linear_192_cast_fp16)[name = string("view_190_cast_fp16")];
            tensor<int32, [4]> transpose_129_perm_0 = const()[name = string("transpose_129_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_21_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(410247296))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411033792))))[name = string("p_encoder_layers_21_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_193_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_21_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_106_cast_fp16)[name = string("linear_193_cast_fp16")];
            tensor<int32, [4]> const_1376 = const()[name = string("const_1376"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_191_cast_fp16 = reshape(shape = const_1376, x = linear_193_cast_fp16)[name = string("view_191_cast_fp16")];
            tensor<int32, [4]> transpose_130_perm_0 = const()[name = string("transpose_130_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_21_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411042048))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411828544))))[name = string("p_encoder_layers_21_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_194_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_21_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_106_cast_fp16)[name = string("linear_194_cast_fp16")];
            tensor<int32, [4]> const_1379 = const()[name = string("const_1379"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_192_cast_fp16 = reshape(shape = const_1379, x = linear_194_cast_fp16)[name = string("view_192_cast_fp16")];
            tensor<int32, [4]> transpose_131_perm_0 = const()[name = string("transpose_131_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_193_to_fp16 = const()[name = string("view_193_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411836800)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_129_cast_fp16 = transpose(perm = transpose_129_perm_0, x = view_190_cast_fp16)[name = string("transpose_17")];
            tensor<fp16, [1, 8, 188, 128]> add_142_cast_fp16 = add(x = transpose_129_cast_fp16, y = view_193_to_fp16)[name = string("add_142_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_194_to_fp16 = const()[name = string("view_194_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411838912)))];
            tensor<fp16, [1, 8, 188, 128]> add_143_cast_fp16 = add(x = transpose_129_cast_fp16, y = view_194_to_fp16)[name = string("add_143_cast_fp16")];
            bool matmul_22_transpose_x_0 = const()[name = string("matmul_22_transpose_x_0"), val = bool(false)];
            bool matmul_22_transpose_y_0 = const()[name = string("matmul_22_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_21_to_fp16 = const()[name = string("permute_21_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411841024)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_22_cast_fp16 = matmul(transpose_x = matmul_22_transpose_x_0, transpose_y = matmul_22_transpose_y_0, x = add_143_cast_fp16, y = permute_21_to_fp16)[name = string("matmul_22_cast_fp16")];
            tensor<int32, [8]> pad_21_pad_0 = const()[name = string("pad_21_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_21_mode_0 = const()[name = string("pad_21_mode_0"), val = string("constant")];
            fp16 const_1387_to_fp16 = const()[name = string("const_1387_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_21_cast_fp16 = pad(constant_val = const_1387_to_fp16, mode = pad_21_mode_0, pad = pad_21_pad_0, x = matmul_22_cast_fp16)[name = string("pad_21_cast_fp16")];
            tensor<int32, [4]> const_1388 = const()[name = string("const_1388"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_196_cast_fp16 = reshape(shape = const_1388, x = pad_21_cast_fp16)[name = string("view_196_cast_fp16")];
            tensor<int32, [4]> slice_43_begin_0 = const()[name = string("slice_43_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_43_end_0 = const()[name = string("slice_43_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_43_end_mask_0 = const()[name = string("slice_43_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_43_cast_fp16 = slice_by_index(begin = slice_43_begin_0, end = slice_43_end_0, end_mask = slice_43_end_mask_0, x = view_196_cast_fp16)[name = string("slice_43_cast_fp16")];
            tensor<int32, [4]> const_1392 = const()[name = string("const_1392"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_197_cast_fp16 = reshape(shape = const_1392, x = slice_43_cast_fp16)[name = string("view_197_cast_fp16")];
            tensor<int32, [4]> slice_44_begin_0 = const()[name = string("slice_44_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_44_end_0 = const()[name = string("slice_44_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_44_end_mask_0 = const()[name = string("slice_44_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_44_cast_fp16 = slice_by_index(begin = slice_44_begin_0, end = slice_44_end_0, end_mask = slice_44_end_mask_0, x = view_197_cast_fp16)[name = string("slice_44_cast_fp16")];
            fp16 const_1396_to_fp16 = const()[name = string("const_1396_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_70_cast_fp16 = mul(x = slice_44_cast_fp16, y = const_1396_to_fp16)[name = string("mul_70_cast_fp16")];
            fp16 const_1397_to_fp16 = const()[name = string("const_1397_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_42_cast_fp16 = select(a = const_1397_to_fp16, b = mul_70_cast_fp16, cond = logical_not)[name = string("masked_fill_42_cast_fp16")];
            fp16 const_1398_to_fp16 = const()[name = string("const_1398_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_21_1_cast_fp16 = mul(x = add_142_cast_fp16, y = const_1398_to_fp16)[name = string("mul_21_1_cast_fp16")];
            bool matmul_21_transpose_y_1 = const()[name = string("matmul_21_transpose_y_1"), val = bool(true)];
            bool matmul_21_transpose_x_1 = const()[name = string("matmul_21_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_130_cast_fp16 = transpose(perm = transpose_130_perm_0, x = view_191_cast_fp16)[name = string("transpose_16")];
            tensor<fp16, [1, 8, 188, 188]> matmul_21_1_cast_fp16 = matmul(transpose_x = matmul_21_transpose_x_1, transpose_y = matmul_21_transpose_y_1, x = mul_21_1_cast_fp16, y = transpose_130_cast_fp16)[name = string("matmul_21_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_21_1_cast_fp16 = add(x = matmul_21_1_cast_fp16, y = masked_fill_42_cast_fp16)[name = string("add_21_1_cast_fp16")];
            int32 softmax_21_axis_0 = const()[name = string("softmax_21_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_21_cast_fp16 = softmax(axis = softmax_21_axis_0, x = add_21_1_cast_fp16)[name = string("softmax_21_cast_fp16")];
            bool scaled_dot_product_attention_21_transpose_x_0 = const()[name = string("scaled_dot_product_attention_21_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_21_transpose_y_0 = const()[name = string("scaled_dot_product_attention_21_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_131_cast_fp16 = transpose(perm = transpose_131_perm_0, x = view_192_cast_fp16)[name = string("transpose_15")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_21_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_21_transpose_x_0, transpose_y = scaled_dot_product_attention_21_transpose_y_0, x = softmax_21_cast_fp16, y = transpose_131_cast_fp16)[name = string("scaled_dot_product_attention_21_cast_fp16")];
            tensor<int32, [4]> transpose_132_perm_0 = const()[name = string("transpose_132_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_1401 = const()[name = string("const_1401"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_132_cast_fp16 = transpose(perm = transpose_132_perm_0, x = scaled_dot_product_attention_21_cast_fp16)[name = string("transpose_14")];
            tensor<fp16, [1, 188, 1024]> view_198_cast_fp16 = reshape(shape = const_1401, x = transpose_132_cast_fp16)[name = string("view_198_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_21_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412609088))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(413395584))))[name = string("p_encoder_layers_21_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_196_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_21_self_attn_o_proj_weight_to_fp16_palettized, x = view_198_cast_fp16)[name = string("linear_196_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_144_cast_fp16 = add(x = add_141_cast_fp16, y = linear_196_cast_fp16)[name = string("add_144_cast_fp16")];
            tensor<int32, [1]> layer_norm_107_axes_0 = const()[name = string("layer_norm_107_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_21_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_21_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(413403840)))];
            tensor<fp16, [1024]> p_encoder_layers_21_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_21_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(413405952)))];
            fp16 const_1403_to_fp16 = const()[name = string("const_1403_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_107_cast_fp16 = layer_norm(axes = layer_norm_107_axes_0, beta = p_encoder_layers_21_norm_conv_bias_to_fp16, epsilon = const_1403_to_fp16, gamma = p_encoder_layers_21_norm_conv_weight_to_fp16, x = add_144_cast_fp16)[name = string("layer_norm_107_cast_fp16")];
            tensor<int32, [3]> transpose_133_perm_0 = const()[name = string("transpose_133_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_63_pad_type_0 = const()[name = string("conv1d_63_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_63_strides_0 = const()[name = string("conv1d_63_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_63_pad_0 = const()[name = string("conv1d_63_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_63_dilations_0 = const()[name = string("conv1d_63_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_63_groups_0 = const()[name = string("conv1d_63_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_21_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(413408064))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(414980992))))[name = string("p_encoder_layers_21_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_133_cast_fp16 = transpose(perm = transpose_133_perm_0, x = layer_norm_107_cast_fp16)[name = string("transpose_13")];
            tensor<fp16, [1, 2048, 188]> conv1d_63_cast_fp16 = conv(dilations = conv1d_63_dilations_0, groups = conv1d_63_groups_0, pad = conv1d_63_pad_0, pad_type = conv1d_63_pad_type_0, strides = conv1d_63_strides_0, weight = p_encoder_layers_21_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_133_cast_fp16)[name = string("conv1d_63_cast_fp16")];
            int32 glu_21_split_num_splits_0 = const()[name = string("glu_21_split_num_splits_0"), val = int32(2)];
            int32 glu_21_split_axis_0 = const()[name = string("glu_21_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_21_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_21_split_cast_fp16_1 = split(axis = glu_21_split_axis_0, num_splits = glu_21_split_num_splits_0, x = conv1d_63_cast_fp16)[name = string("glu_21_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_21_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_21_split_cast_fp16_1)[name = string("glu_21_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_21_cast_fp16 = mul(x = glu_21_split_cast_fp16_0, y = glu_21_split_1_sigmoid_cast_fp16)[name = string("glu_21_cast_fp16")];
            fp16 const_1409_to_fp16 = const()[name = string("const_1409_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_43_cast_fp16 = select(a = const_1409_to_fp16, b = glu_21_cast_fp16, cond = all_1)[name = string("masked_fill_43_cast_fp16")];
            string conv1d_64_pad_type_0 = const()[name = string("conv1d_64_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_64_pad_0 = const()[name = string("conv1d_64_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_64_groups_0 = const()[name = string("conv1d_64_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_64_strides_0 = const()[name = string("conv1d_64_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_64_dilations_0 = const()[name = string("conv1d_64_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1589_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(414997440))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(415004416))))[name = string("const_1589_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1590_to_fp16 = const()[name = string("const_1590_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(415012672)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_21_cast_fp16 = conv(bias = const_1590_to_fp16, dilations = conv1d_64_dilations_0, groups = conv1d_64_groups_0, pad = conv1d_64_pad_0, pad_type = conv1d_64_pad_type_0, strides = conv1d_64_strides_0, weight = const_1589_to_fp16_palettized, x = masked_fill_43_cast_fp16)[name = string("_native_batch_norm_legit_no_training_21_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_64_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_21_cast_fp16)[name = string("silu_64_cast_fp16")];
            string conv1d_65_pad_type_0 = const()[name = string("conv1d_65_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_65_strides_0 = const()[name = string("conv1d_65_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_65_pad_0 = const()[name = string("conv1d_65_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_65_dilations_0 = const()[name = string("conv1d_65_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_65_groups_0 = const()[name = string("conv1d_65_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_21_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(415014784))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(415801280))))[name = string("p_encoder_layers_21_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_65_cast_fp16 = conv(dilations = conv1d_65_dilations_0, groups = conv1d_65_groups_0, pad = conv1d_65_pad_0, pad_type = conv1d_65_pad_type_0, strides = conv1d_65_strides_0, weight = p_encoder_layers_21_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_64_cast_fp16)[name = string("conv1d_65_cast_fp16")];
            tensor<int32, [3]> transpose_134_perm_0 = const()[name = string("transpose_134_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_134_cast_fp16 = transpose(perm = transpose_134_perm_0, x = conv1d_65_cast_fp16)[name = string("transpose_12")];
            tensor<fp16, [1, 188, 1024]> add_145_cast_fp16 = add(x = add_144_cast_fp16, y = transpose_134_cast_fp16)[name = string("add_145_cast_fp16")];
            tensor<int32, [1]> layer_norm_108_axes_0 = const()[name = string("layer_norm_108_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_21_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_21_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(415809536)))];
            tensor<fp16, [1024]> p_encoder_layers_21_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_21_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(415811648)))];
            fp16 const_1420_to_fp16 = const()[name = string("const_1420_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_108_cast_fp16 = layer_norm(axes = layer_norm_108_axes_0, beta = p_encoder_layers_21_norm_feed_forward2_bias_to_fp16, epsilon = const_1420_to_fp16, gamma = p_encoder_layers_21_norm_feed_forward2_weight_to_fp16, x = add_145_cast_fp16)[name = string("layer_norm_108_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_21_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(415813760))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(418959552))))[name = string("p_encoder_layers_21_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_197_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_21_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_108_cast_fp16)[name = string("linear_197_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_65_cast_fp16 = silu(x = linear_197_cast_fp16)[name = string("silu_65_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_21_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(418992384))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(422138176))))[name = string("p_encoder_layers_21_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_198_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_21_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_65_cast_fp16)[name = string("linear_198_cast_fp16")];
            fp16 const_1422_to_fp16 = const()[name = string("const_1422_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_71_cast_fp16 = mul(x = linear_198_cast_fp16, y = const_1422_to_fp16)[name = string("mul_71_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_146_cast_fp16 = add(x = add_145_cast_fp16, y = mul_71_cast_fp16)[name = string("add_146_cast_fp16")];
            tensor<int32, [1]> layer_norm_109_axes_0 = const()[name = string("layer_norm_109_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_21_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_21_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(422146432)))];
            tensor<fp16, [1024]> p_encoder_layers_21_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_21_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(422148544)))];
            fp16 const_1424_to_fp16 = const()[name = string("const_1424_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_109_cast_fp16 = layer_norm(axes = layer_norm_109_axes_0, beta = p_encoder_layers_21_norm_out_bias_to_fp16, epsilon = const_1424_to_fp16, gamma = p_encoder_layers_21_norm_out_weight_to_fp16, x = add_146_cast_fp16)[name = string("layer_norm_109_cast_fp16")];
            tensor<int32, [1]> layer_norm_110_axes_0 = const()[name = string("layer_norm_110_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_22_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_22_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(422150656)))];
            tensor<fp16, [1024]> p_encoder_layers_22_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_22_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(422152768)))];
            fp16 const_1427_to_fp16 = const()[name = string("const_1427_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_110_cast_fp16 = layer_norm(axes = layer_norm_110_axes_0, beta = p_encoder_layers_22_norm_feed_forward1_bias_to_fp16, epsilon = const_1427_to_fp16, gamma = p_encoder_layers_22_norm_feed_forward1_weight_to_fp16, x = layer_norm_109_cast_fp16)[name = string("layer_norm_110_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_22_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(422154880))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(425300672))))[name = string("p_encoder_layers_22_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_199_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_22_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_110_cast_fp16)[name = string("linear_199_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_66_cast_fp16 = silu(x = linear_199_cast_fp16)[name = string("silu_66_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_22_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(425333504))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(428479296))))[name = string("p_encoder_layers_22_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_200_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_22_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_66_cast_fp16)[name = string("linear_200_cast_fp16")];
            fp16 const_1429_to_fp16 = const()[name = string("const_1429_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_72_cast_fp16 = mul(x = linear_200_cast_fp16, y = const_1429_to_fp16)[name = string("mul_72_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_147_cast_fp16 = add(x = layer_norm_109_cast_fp16, y = mul_72_cast_fp16)[name = string("add_147_cast_fp16")];
            tensor<int32, [1]> layer_norm_111_axes_0 = const()[name = string("layer_norm_111_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_22_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_22_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(428487552)))];
            tensor<fp16, [1024]> p_encoder_layers_22_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_22_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(428489664)))];
            fp16 const_1431_to_fp16 = const()[name = string("const_1431_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_111_cast_fp16 = layer_norm(axes = layer_norm_111_axes_0, beta = p_encoder_layers_22_norm_self_att_bias_to_fp16, epsilon = const_1431_to_fp16, gamma = p_encoder_layers_22_norm_self_att_weight_to_fp16, x = add_147_cast_fp16)[name = string("layer_norm_111_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_22_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(428491776))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(429278272))))[name = string("p_encoder_layers_22_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_201_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_22_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_111_cast_fp16)[name = string("linear_201_cast_fp16")];
            tensor<int32, [4]> const_1433 = const()[name = string("const_1433"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_199_cast_fp16 = reshape(shape = const_1433, x = linear_201_cast_fp16)[name = string("view_199_cast_fp16")];
            tensor<int32, [4]> transpose_135_perm_0 = const()[name = string("transpose_135_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_22_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(429286528))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(430073024))))[name = string("p_encoder_layers_22_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_202_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_22_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_111_cast_fp16)[name = string("linear_202_cast_fp16")];
            tensor<int32, [4]> const_1436 = const()[name = string("const_1436"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_200_cast_fp16 = reshape(shape = const_1436, x = linear_202_cast_fp16)[name = string("view_200_cast_fp16")];
            tensor<int32, [4]> transpose_136_perm_0 = const()[name = string("transpose_136_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_22_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(430081280))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(430867776))))[name = string("p_encoder_layers_22_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_203_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_22_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_111_cast_fp16)[name = string("linear_203_cast_fp16")];
            tensor<int32, [4]> const_1439 = const()[name = string("const_1439"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_201_cast_fp16 = reshape(shape = const_1439, x = linear_203_cast_fp16)[name = string("view_201_cast_fp16")];
            tensor<int32, [4]> transpose_137_perm_0 = const()[name = string("transpose_137_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_202_to_fp16 = const()[name = string("view_202_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(430876032)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_135_cast_fp16 = transpose(perm = transpose_135_perm_0, x = view_199_cast_fp16)[name = string("transpose_11")];
            tensor<fp16, [1, 8, 188, 128]> add_148_cast_fp16 = add(x = transpose_135_cast_fp16, y = view_202_to_fp16)[name = string("add_148_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_203_to_fp16 = const()[name = string("view_203_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(430878144)))];
            tensor<fp16, [1, 8, 188, 128]> add_149_cast_fp16 = add(x = transpose_135_cast_fp16, y = view_203_to_fp16)[name = string("add_149_cast_fp16")];
            bool matmul_23_transpose_x_0 = const()[name = string("matmul_23_transpose_x_0"), val = bool(false)];
            bool matmul_23_transpose_y_0 = const()[name = string("matmul_23_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_22_to_fp16 = const()[name = string("permute_22_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(430880256)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_23_cast_fp16 = matmul(transpose_x = matmul_23_transpose_x_0, transpose_y = matmul_23_transpose_y_0, x = add_149_cast_fp16, y = permute_22_to_fp16)[name = string("matmul_23_cast_fp16")];
            tensor<int32, [8]> pad_22_pad_0 = const()[name = string("pad_22_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_22_mode_0 = const()[name = string("pad_22_mode_0"), val = string("constant")];
            fp16 const_1447_to_fp16 = const()[name = string("const_1447_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_22_cast_fp16 = pad(constant_val = const_1447_to_fp16, mode = pad_22_mode_0, pad = pad_22_pad_0, x = matmul_23_cast_fp16)[name = string("pad_22_cast_fp16")];
            tensor<int32, [4]> const_1448 = const()[name = string("const_1448"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_205_cast_fp16 = reshape(shape = const_1448, x = pad_22_cast_fp16)[name = string("view_205_cast_fp16")];
            tensor<int32, [4]> slice_45_begin_0 = const()[name = string("slice_45_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_45_end_0 = const()[name = string("slice_45_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_45_end_mask_0 = const()[name = string("slice_45_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_45_cast_fp16 = slice_by_index(begin = slice_45_begin_0, end = slice_45_end_0, end_mask = slice_45_end_mask_0, x = view_205_cast_fp16)[name = string("slice_45_cast_fp16")];
            tensor<int32, [4]> const_1452 = const()[name = string("const_1452"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_206_cast_fp16 = reshape(shape = const_1452, x = slice_45_cast_fp16)[name = string("view_206_cast_fp16")];
            tensor<int32, [4]> slice_46_begin_0 = const()[name = string("slice_46_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_46_end_0 = const()[name = string("slice_46_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_46_end_mask_0 = const()[name = string("slice_46_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_46_cast_fp16 = slice_by_index(begin = slice_46_begin_0, end = slice_46_end_0, end_mask = slice_46_end_mask_0, x = view_206_cast_fp16)[name = string("slice_46_cast_fp16")];
            fp16 const_1456_to_fp16 = const()[name = string("const_1456_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_73_cast_fp16 = mul(x = slice_46_cast_fp16, y = const_1456_to_fp16)[name = string("mul_73_cast_fp16")];
            fp16 const_1457_to_fp16 = const()[name = string("const_1457_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_44_cast_fp16 = select(a = const_1457_to_fp16, b = mul_73_cast_fp16, cond = logical_not)[name = string("masked_fill_44_cast_fp16")];
            fp16 const_1458_to_fp16 = const()[name = string("const_1458_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_22_1_cast_fp16 = mul(x = add_148_cast_fp16, y = const_1458_to_fp16)[name = string("mul_22_1_cast_fp16")];
            bool matmul_22_transpose_y_1 = const()[name = string("matmul_22_transpose_y_1"), val = bool(true)];
            bool matmul_22_transpose_x_1 = const()[name = string("matmul_22_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_136_cast_fp16 = transpose(perm = transpose_136_perm_0, x = view_200_cast_fp16)[name = string("transpose_10")];
            tensor<fp16, [1, 8, 188, 188]> matmul_22_1_cast_fp16 = matmul(transpose_x = matmul_22_transpose_x_1, transpose_y = matmul_22_transpose_y_1, x = mul_22_1_cast_fp16, y = transpose_136_cast_fp16)[name = string("matmul_22_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_22_1_cast_fp16 = add(x = matmul_22_1_cast_fp16, y = masked_fill_44_cast_fp16)[name = string("add_22_1_cast_fp16")];
            int32 softmax_22_axis_0 = const()[name = string("softmax_22_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_22_cast_fp16 = softmax(axis = softmax_22_axis_0, x = add_22_1_cast_fp16)[name = string("softmax_22_cast_fp16")];
            bool scaled_dot_product_attention_22_transpose_x_0 = const()[name = string("scaled_dot_product_attention_22_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_22_transpose_y_0 = const()[name = string("scaled_dot_product_attention_22_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_137_cast_fp16 = transpose(perm = transpose_137_perm_0, x = view_201_cast_fp16)[name = string("transpose_9")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_22_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_22_transpose_x_0, transpose_y = scaled_dot_product_attention_22_transpose_y_0, x = softmax_22_cast_fp16, y = transpose_137_cast_fp16)[name = string("scaled_dot_product_attention_22_cast_fp16")];
            tensor<int32, [4]> transpose_138_perm_0 = const()[name = string("transpose_138_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_1461 = const()[name = string("const_1461"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_138_cast_fp16 = transpose(perm = transpose_138_perm_0, x = scaled_dot_product_attention_22_cast_fp16)[name = string("transpose_8")];
            tensor<fp16, [1, 188, 1024]> view_207_cast_fp16 = reshape(shape = const_1461, x = transpose_138_cast_fp16)[name = string("view_207_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_22_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(431648320))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(432434816))))[name = string("p_encoder_layers_22_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_205_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_22_self_attn_o_proj_weight_to_fp16_palettized, x = view_207_cast_fp16)[name = string("linear_205_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_150_cast_fp16 = add(x = add_147_cast_fp16, y = linear_205_cast_fp16)[name = string("add_150_cast_fp16")];
            tensor<int32, [1]> layer_norm_112_axes_0 = const()[name = string("layer_norm_112_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_22_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_22_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(432443072)))];
            tensor<fp16, [1024]> p_encoder_layers_22_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_22_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(432445184)))];
            fp16 const_1463_to_fp16 = const()[name = string("const_1463_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_112_cast_fp16 = layer_norm(axes = layer_norm_112_axes_0, beta = p_encoder_layers_22_norm_conv_bias_to_fp16, epsilon = const_1463_to_fp16, gamma = p_encoder_layers_22_norm_conv_weight_to_fp16, x = add_150_cast_fp16)[name = string("layer_norm_112_cast_fp16")];
            tensor<int32, [3]> transpose_139_perm_0 = const()[name = string("transpose_139_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_66_pad_type_0 = const()[name = string("conv1d_66_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_66_strides_0 = const()[name = string("conv1d_66_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_66_pad_0 = const()[name = string("conv1d_66_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_66_dilations_0 = const()[name = string("conv1d_66_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_66_groups_0 = const()[name = string("conv1d_66_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_22_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(432447296))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434020224))))[name = string("p_encoder_layers_22_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_139_cast_fp16 = transpose(perm = transpose_139_perm_0, x = layer_norm_112_cast_fp16)[name = string("transpose_7")];
            tensor<fp16, [1, 2048, 188]> conv1d_66_cast_fp16 = conv(dilations = conv1d_66_dilations_0, groups = conv1d_66_groups_0, pad = conv1d_66_pad_0, pad_type = conv1d_66_pad_type_0, strides = conv1d_66_strides_0, weight = p_encoder_layers_22_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_139_cast_fp16)[name = string("conv1d_66_cast_fp16")];
            int32 glu_22_split_num_splits_0 = const()[name = string("glu_22_split_num_splits_0"), val = int32(2)];
            int32 glu_22_split_axis_0 = const()[name = string("glu_22_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_22_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_22_split_cast_fp16_1 = split(axis = glu_22_split_axis_0, num_splits = glu_22_split_num_splits_0, x = conv1d_66_cast_fp16)[name = string("glu_22_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_22_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_22_split_cast_fp16_1)[name = string("glu_22_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_22_cast_fp16 = mul(x = glu_22_split_cast_fp16_0, y = glu_22_split_1_sigmoid_cast_fp16)[name = string("glu_22_cast_fp16")];
            fp16 const_1469_to_fp16 = const()[name = string("const_1469_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_45_cast_fp16 = select(a = const_1469_to_fp16, b = glu_22_cast_fp16, cond = all_1)[name = string("masked_fill_45_cast_fp16")];
            string conv1d_67_pad_type_0 = const()[name = string("conv1d_67_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_67_pad_0 = const()[name = string("conv1d_67_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_67_groups_0 = const()[name = string("conv1d_67_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_67_strides_0 = const()[name = string("conv1d_67_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_67_dilations_0 = const()[name = string("conv1d_67_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1591_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434036672))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434043648))))[name = string("const_1591_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1592_to_fp16 = const()[name = string("const_1592_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434051904)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_22_cast_fp16 = conv(bias = const_1592_to_fp16, dilations = conv1d_67_dilations_0, groups = conv1d_67_groups_0, pad = conv1d_67_pad_0, pad_type = conv1d_67_pad_type_0, strides = conv1d_67_strides_0, weight = const_1591_to_fp16_palettized, x = masked_fill_45_cast_fp16)[name = string("_native_batch_norm_legit_no_training_22_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_67_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_22_cast_fp16)[name = string("silu_67_cast_fp16")];
            string conv1d_68_pad_type_0 = const()[name = string("conv1d_68_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_68_strides_0 = const()[name = string("conv1d_68_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_68_pad_0 = const()[name = string("conv1d_68_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_68_dilations_0 = const()[name = string("conv1d_68_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_68_groups_0 = const()[name = string("conv1d_68_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_22_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434054016))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434840512))))[name = string("p_encoder_layers_22_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_68_cast_fp16 = conv(dilations = conv1d_68_dilations_0, groups = conv1d_68_groups_0, pad = conv1d_68_pad_0, pad_type = conv1d_68_pad_type_0, strides = conv1d_68_strides_0, weight = p_encoder_layers_22_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_67_cast_fp16)[name = string("conv1d_68_cast_fp16")];
            tensor<int32, [3]> transpose_140_perm_0 = const()[name = string("transpose_140_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_140_cast_fp16 = transpose(perm = transpose_140_perm_0, x = conv1d_68_cast_fp16)[name = string("transpose_6")];
            tensor<fp16, [1, 188, 1024]> add_151_cast_fp16 = add(x = add_150_cast_fp16, y = transpose_140_cast_fp16)[name = string("add_151_cast_fp16")];
            tensor<int32, [1]> layer_norm_113_axes_0 = const()[name = string("layer_norm_113_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_22_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_22_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434848768)))];
            tensor<fp16, [1024]> p_encoder_layers_22_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_22_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434850880)))];
            fp16 const_1480_to_fp16 = const()[name = string("const_1480_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_113_cast_fp16 = layer_norm(axes = layer_norm_113_axes_0, beta = p_encoder_layers_22_norm_feed_forward2_bias_to_fp16, epsilon = const_1480_to_fp16, gamma = p_encoder_layers_22_norm_feed_forward2_weight_to_fp16, x = add_151_cast_fp16)[name = string("layer_norm_113_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_22_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434852992))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(437998784))))[name = string("p_encoder_layers_22_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_206_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_22_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_113_cast_fp16)[name = string("linear_206_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_68_cast_fp16 = silu(x = linear_206_cast_fp16)[name = string("silu_68_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_22_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438031616))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(441177408))))[name = string("p_encoder_layers_22_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_207_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_22_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_68_cast_fp16)[name = string("linear_207_cast_fp16")];
            fp16 const_1482_to_fp16 = const()[name = string("const_1482_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_74_cast_fp16 = mul(x = linear_207_cast_fp16, y = const_1482_to_fp16)[name = string("mul_74_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_152_cast_fp16 = add(x = add_151_cast_fp16, y = mul_74_cast_fp16)[name = string("add_152_cast_fp16")];
            tensor<int32, [1]> layer_norm_114_axes_0 = const()[name = string("layer_norm_114_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_22_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_22_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(441185664)))];
            tensor<fp16, [1024]> p_encoder_layers_22_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_22_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(441187776)))];
            fp16 const_1484_to_fp16 = const()[name = string("const_1484_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_114_cast_fp16 = layer_norm(axes = layer_norm_114_axes_0, beta = p_encoder_layers_22_norm_out_bias_to_fp16, epsilon = const_1484_to_fp16, gamma = p_encoder_layers_22_norm_out_weight_to_fp16, x = add_152_cast_fp16)[name = string("layer_norm_114_cast_fp16")];
            tensor<int32, [1]> layer_norm_115_axes_0 = const()[name = string("layer_norm_115_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_23_norm_feed_forward1_weight_to_fp16 = const()[name = string("p_encoder_layers_23_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(441189888)))];
            tensor<fp16, [1024]> p_encoder_layers_23_norm_feed_forward1_bias_to_fp16 = const()[name = string("p_encoder_layers_23_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(441192000)))];
            fp16 const_1487_to_fp16 = const()[name = string("const_1487_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_115_cast_fp16 = layer_norm(axes = layer_norm_115_axes_0, beta = p_encoder_layers_23_norm_feed_forward1_bias_to_fp16, epsilon = const_1487_to_fp16, gamma = p_encoder_layers_23_norm_feed_forward1_weight_to_fp16, x = layer_norm_114_cast_fp16)[name = string("layer_norm_115_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_23_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(441194112))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(444339904))))[name = string("p_encoder_layers_23_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_208_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_23_feed_forward1_linear1_weight_to_fp16_palettized, x = layer_norm_115_cast_fp16)[name = string("linear_208_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_69_cast_fp16 = silu(x = linear_208_cast_fp16)[name = string("silu_69_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_23_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(444372736))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447518528))))[name = string("p_encoder_layers_23_feed_forward1_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_209_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_23_feed_forward1_linear2_weight_to_fp16_palettized, x = silu_69_cast_fp16)[name = string("linear_209_cast_fp16")];
            fp16 const_1489_to_fp16 = const()[name = string("const_1489_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_75_cast_fp16 = mul(x = linear_209_cast_fp16, y = const_1489_to_fp16)[name = string("mul_75_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_153_cast_fp16 = add(x = layer_norm_114_cast_fp16, y = mul_75_cast_fp16)[name = string("add_153_cast_fp16")];
            tensor<int32, [1]> layer_norm_116_axes_0 = const()[name = string("layer_norm_116_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_23_norm_self_att_weight_to_fp16 = const()[name = string("p_encoder_layers_23_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447526784)))];
            tensor<fp16, [1024]> p_encoder_layers_23_norm_self_att_bias_to_fp16 = const()[name = string("p_encoder_layers_23_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447528896)))];
            fp16 const_1491_to_fp16 = const()[name = string("const_1491_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_116_cast_fp16 = layer_norm(axes = layer_norm_116_axes_0, beta = p_encoder_layers_23_norm_self_att_bias_to_fp16, epsilon = const_1491_to_fp16, gamma = p_encoder_layers_23_norm_self_att_weight_to_fp16, x = add_153_cast_fp16)[name = string("layer_norm_116_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_23_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447531008))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(448317504))))[name = string("p_encoder_layers_23_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_210_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_23_self_attn_q_proj_weight_to_fp16_palettized, x = layer_norm_116_cast_fp16)[name = string("linear_210_cast_fp16")];
            tensor<int32, [4]> const_1493 = const()[name = string("const_1493"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_208_cast_fp16 = reshape(shape = const_1493, x = linear_210_cast_fp16)[name = string("view_208_cast_fp16")];
            tensor<int32, [4]> transpose_141_perm_0 = const()[name = string("transpose_141_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_23_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(448325760))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(449112256))))[name = string("p_encoder_layers_23_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_211_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_23_self_attn_k_proj_weight_to_fp16_palettized, x = layer_norm_116_cast_fp16)[name = string("linear_211_cast_fp16")];
            tensor<int32, [4]> const_1496 = const()[name = string("const_1496"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_209_cast_fp16 = reshape(shape = const_1496, x = linear_211_cast_fp16)[name = string("view_209_cast_fp16")];
            tensor<int32, [4]> transpose_142_perm_0 = const()[name = string("transpose_142_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1024, 1024]> p_encoder_layers_23_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(449120512))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(449907008))))[name = string("p_encoder_layers_23_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_212_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_23_self_attn_v_proj_weight_to_fp16_palettized, x = layer_norm_116_cast_fp16)[name = string("linear_212_cast_fp16")];
            tensor<int32, [4]> const_1499 = const()[name = string("const_1499"), val = tensor<int32, [4]>([1, 188, -1, 128])];
            tensor<fp16, [1, 188, 8, 128]> view_210_cast_fp16 = reshape(shape = const_1499, x = linear_212_cast_fp16)[name = string("view_210_cast_fp16")];
            tensor<int32, [4]> transpose_143_perm_0 = const()[name = string("transpose_143_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 8, 1, 128]> view_211_to_fp16 = const()[name = string("view_211_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(449915264)))];
            tensor<fp16, [1, 8, 188, 128]> transpose_141_cast_fp16 = transpose(perm = transpose_141_perm_0, x = view_208_cast_fp16)[name = string("transpose_5")];
            tensor<fp16, [1, 8, 188, 128]> add_154_cast_fp16 = add(x = transpose_141_cast_fp16, y = view_211_to_fp16)[name = string("add_154_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> view_212_to_fp16 = const()[name = string("view_212_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(449917376)))];
            tensor<fp16, [1, 8, 188, 128]> add_155_cast_fp16 = add(x = transpose_141_cast_fp16, y = view_212_to_fp16)[name = string("add_155_cast_fp16")];
            bool matmul_24_transpose_x_0 = const()[name = string("matmul_24_transpose_x_0"), val = bool(false)];
            bool matmul_24_transpose_y_0 = const()[name = string("matmul_24_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> permute_23_to_fp16 = const()[name = string("permute_23_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(449919488)))];
            tensor<fp16, [1, 8, 188, 375]> matmul_24_cast_fp16 = matmul(transpose_x = matmul_24_transpose_x_0, transpose_y = matmul_24_transpose_y_0, x = add_155_cast_fp16, y = permute_23_to_fp16)[name = string("matmul_24_cast_fp16")];
            tensor<int32, [8]> pad_23_pad_0 = const()[name = string("pad_23_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string pad_23_mode_0 = const()[name = string("pad_23_mode_0"), val = string("constant")];
            fp16 const_1507_to_fp16 = const()[name = string("const_1507_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> pad_23_cast_fp16 = pad(constant_val = const_1507_to_fp16, mode = pad_23_mode_0, pad = pad_23_pad_0, x = matmul_24_cast_fp16)[name = string("pad_23_cast_fp16")];
            tensor<int32, [4]> const_1508 = const()[name = string("const_1508"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> view_214_cast_fp16 = reshape(shape = const_1508, x = pad_23_cast_fp16)[name = string("view_214_cast_fp16")];
            tensor<int32, [4]> slice_47_begin_0 = const()[name = string("slice_47_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> slice_47_end_0 = const()[name = string("slice_47_end_0"), val = tensor<int32, [4]>([1, 8, 1, 188])];
            tensor<bool, [4]> slice_47_end_mask_0 = const()[name = string("slice_47_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> slice_47_cast_fp16 = slice_by_index(begin = slice_47_begin_0, end = slice_47_end_0, end_mask = slice_47_end_mask_0, x = view_214_cast_fp16)[name = string("slice_47_cast_fp16")];
            tensor<int32, [4]> const_1512 = const()[name = string("const_1512"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> view_215_cast_fp16 = reshape(shape = const_1512, x = slice_47_cast_fp16)[name = string("view_215_cast_fp16")];
            tensor<int32, [4]> slice_48_begin_0 = const()[name = string("slice_48_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_48_end_0 = const()[name = string("slice_48_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> slice_48_end_mask_0 = const()[name = string("slice_48_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> slice_48_cast_fp16 = slice_by_index(begin = slice_48_begin_0, end = slice_48_end_0, end_mask = slice_48_end_mask_0, x = view_215_cast_fp16)[name = string("slice_48_cast_fp16")];
            fp16 const_1516_to_fp16 = const()[name = string("const_1516_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> mul_76_cast_fp16 = mul(x = slice_48_cast_fp16, y = const_1516_to_fp16)[name = string("mul_76_cast_fp16")];
            fp16 const_1517_to_fp16 = const()[name = string("const_1517_to_fp16"), val = fp16(-inf)];
            tensor<fp16, [1, 8, 188, 188]> masked_fill_46_cast_fp16 = select(a = const_1517_to_fp16, b = mul_76_cast_fp16, cond = logical_not)[name = string("masked_fill_46_cast_fp16")];
            fp16 const_1518_to_fp16 = const()[name = string("const_1518_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 128]> mul_23_1_cast_fp16 = mul(x = add_154_cast_fp16, y = const_1518_to_fp16)[name = string("mul_23_1_cast_fp16")];
            bool matmul_23_transpose_y_1 = const()[name = string("matmul_23_transpose_y_1"), val = bool(true)];
            bool matmul_23_transpose_x_1 = const()[name = string("matmul_23_transpose_x_1"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_142_cast_fp16 = transpose(perm = transpose_142_perm_0, x = view_209_cast_fp16)[name = string("transpose_4")];
            tensor<fp16, [1, 8, 188, 188]> matmul_23_1_cast_fp16 = matmul(transpose_x = matmul_23_transpose_x_1, transpose_y = matmul_23_transpose_y_1, x = mul_23_1_cast_fp16, y = transpose_142_cast_fp16)[name = string("matmul_23_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> add_23_1_cast_fp16 = add(x = matmul_23_1_cast_fp16, y = masked_fill_46_cast_fp16)[name = string("add_23_1_cast_fp16")];
            int32 softmax_23_axis_0 = const()[name = string("softmax_23_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> softmax_23_cast_fp16 = softmax(axis = softmax_23_axis_0, x = add_23_1_cast_fp16)[name = string("softmax_23_cast_fp16")];
            bool scaled_dot_product_attention_23_transpose_x_0 = const()[name = string("scaled_dot_product_attention_23_transpose_x_0"), val = bool(false)];
            bool scaled_dot_product_attention_23_transpose_y_0 = const()[name = string("scaled_dot_product_attention_23_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> transpose_143_cast_fp16 = transpose(perm = transpose_143_perm_0, x = view_210_cast_fp16)[name = string("transpose_3")];
            tensor<fp16, [1, 8, 188, 128]> scaled_dot_product_attention_23_cast_fp16 = matmul(transpose_x = scaled_dot_product_attention_23_transpose_x_0, transpose_y = scaled_dot_product_attention_23_transpose_y_0, x = softmax_23_cast_fp16, y = transpose_143_cast_fp16)[name = string("scaled_dot_product_attention_23_cast_fp16")];
            tensor<int32, [4]> transpose_144_perm_0 = const()[name = string("transpose_144_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_1521 = const()[name = string("const_1521"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 8, 128]> transpose_144_cast_fp16 = transpose(perm = transpose_144_perm_0, x = scaled_dot_product_attention_23_cast_fp16)[name = string("transpose_2")];
            tensor<fp16, [1, 188, 1024]> view_216_cast_fp16 = reshape(shape = const_1521, x = transpose_144_cast_fp16)[name = string("view_216_cast_fp16")];
            tensor<fp16, [1024, 1024]> p_encoder_layers_23_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(450687552))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(451474048))))[name = string("p_encoder_layers_23_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_214_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_23_self_attn_o_proj_weight_to_fp16_palettized, x = view_216_cast_fp16)[name = string("linear_214_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_156_cast_fp16 = add(x = add_153_cast_fp16, y = linear_214_cast_fp16)[name = string("add_156_cast_fp16")];
            tensor<int32, [1]> layer_norm_117_axes_0 = const()[name = string("layer_norm_117_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_23_norm_conv_weight_to_fp16 = const()[name = string("p_encoder_layers_23_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(451482304)))];
            tensor<fp16, [1024]> p_encoder_layers_23_norm_conv_bias_to_fp16 = const()[name = string("p_encoder_layers_23_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(451484416)))];
            fp16 const_1523_to_fp16 = const()[name = string("const_1523_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_117_cast_fp16 = layer_norm(axes = layer_norm_117_axes_0, beta = p_encoder_layers_23_norm_conv_bias_to_fp16, epsilon = const_1523_to_fp16, gamma = p_encoder_layers_23_norm_conv_weight_to_fp16, x = add_156_cast_fp16)[name = string("layer_norm_117_cast_fp16")];
            tensor<int32, [3]> transpose_145_perm_0 = const()[name = string("transpose_145_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string conv1d_69_pad_type_0 = const()[name = string("conv1d_69_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_69_strides_0 = const()[name = string("conv1d_69_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_69_pad_0 = const()[name = string("conv1d_69_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_69_dilations_0 = const()[name = string("conv1d_69_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_69_groups_0 = const()[name = string("conv1d_69_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> p_encoder_layers_23_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(451486528))), lut = tensor<fp16, [128, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(453059456))))[name = string("p_encoder_layers_23_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> transpose_145_cast_fp16 = transpose(perm = transpose_145_perm_0, x = layer_norm_117_cast_fp16)[name = string("transpose_1")];
            tensor<fp16, [1, 2048, 188]> conv1d_69_cast_fp16 = conv(dilations = conv1d_69_dilations_0, groups = conv1d_69_groups_0, pad = conv1d_69_pad_0, pad_type = conv1d_69_pad_type_0, strides = conv1d_69_strides_0, weight = p_encoder_layers_23_conv_pointwise_conv1_weight_to_fp16_palettized, x = transpose_145_cast_fp16)[name = string("conv1d_69_cast_fp16")];
            int32 glu_23_split_num_splits_0 = const()[name = string("glu_23_split_num_splits_0"), val = int32(2)];
            int32 glu_23_split_axis_0 = const()[name = string("glu_23_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> glu_23_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> glu_23_split_cast_fp16_1 = split(axis = glu_23_split_axis_0, num_splits = glu_23_split_num_splits_0, x = conv1d_69_cast_fp16)[name = string("glu_23_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_23_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_23_split_cast_fp16_1)[name = string("glu_23_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> glu_23_cast_fp16 = mul(x = glu_23_split_cast_fp16_0, y = glu_23_split_1_sigmoid_cast_fp16)[name = string("glu_23_cast_fp16")];
            fp16 const_1529_to_fp16 = const()[name = string("const_1529_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 188]> masked_fill_47_cast_fp16 = select(a = const_1529_to_fp16, b = glu_23_cast_fp16, cond = all_1)[name = string("masked_fill_47_cast_fp16")];
            string conv1d_70_pad_type_0 = const()[name = string("conv1d_70_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> conv1d_70_pad_0 = const()[name = string("conv1d_70_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 conv1d_70_groups_0 = const()[name = string("conv1d_70_groups_0"), val = int32(1024)];
            tensor<int32, [1]> conv1d_70_strides_0 = const()[name = string("conv1d_70_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_70_dilations_0 = const()[name = string("conv1d_70_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_1593_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(453075904))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(453082880))))[name = string("const_1593_to_fp16_palettized")];
            tensor<fp16, [1024]> const_1594_to_fp16 = const()[name = string("const_1594_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(453091136)))];
            tensor<fp16, [1, 1024, 188]> _native_batch_norm_legit_no_training_23_cast_fp16 = conv(bias = const_1594_to_fp16, dilations = conv1d_70_dilations_0, groups = conv1d_70_groups_0, pad = conv1d_70_pad_0, pad_type = conv1d_70_pad_type_0, strides = conv1d_70_strides_0, weight = const_1593_to_fp16_palettized, x = masked_fill_47_cast_fp16)[name = string("_native_batch_norm_legit_no_training_23_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> silu_70_cast_fp16 = silu(x = _native_batch_norm_legit_no_training_23_cast_fp16)[name = string("silu_70_cast_fp16")];
            string conv1d_71_pad_type_0 = const()[name = string("conv1d_71_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> conv1d_71_strides_0 = const()[name = string("conv1d_71_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_71_pad_0 = const()[name = string("conv1d_71_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_71_dilations_0 = const()[name = string("conv1d_71_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 conv1d_71_groups_0 = const()[name = string("conv1d_71_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> p_encoder_layers_23_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(453093248))), lut = tensor<fp16, [64, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(453879744))))[name = string("p_encoder_layers_23_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 188]> conv1d_71_cast_fp16 = conv(dilations = conv1d_71_dilations_0, groups = conv1d_71_groups_0, pad = conv1d_71_pad_0, pad_type = conv1d_71_pad_type_0, strides = conv1d_71_strides_0, weight = p_encoder_layers_23_conv_pointwise_conv2_weight_to_fp16_palettized, x = silu_70_cast_fp16)[name = string("conv1d_71_cast_fp16")];
            tensor<int32, [3]> transpose_146_perm_0 = const()[name = string("transpose_146_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> transpose_146_cast_fp16 = transpose(perm = transpose_146_perm_0, x = conv1d_71_cast_fp16)[name = string("transpose_0")];
            tensor<fp16, [1, 188, 1024]> add_157_cast_fp16 = add(x = add_156_cast_fp16, y = transpose_146_cast_fp16)[name = string("add_157_cast_fp16")];
            tensor<int32, [1]> layer_norm_118_axes_0 = const()[name = string("layer_norm_118_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_23_norm_feed_forward2_weight_to_fp16 = const()[name = string("p_encoder_layers_23_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(453888000)))];
            tensor<fp16, [1024]> p_encoder_layers_23_norm_feed_forward2_bias_to_fp16 = const()[name = string("p_encoder_layers_23_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(453890112)))];
            fp16 const_1540_to_fp16 = const()[name = string("const_1540_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_118_cast_fp16 = layer_norm(axes = layer_norm_118_axes_0, beta = p_encoder_layers_23_norm_feed_forward2_bias_to_fp16, epsilon = const_1540_to_fp16, gamma = p_encoder_layers_23_norm_feed_forward2_weight_to_fp16, x = add_157_cast_fp16)[name = string("layer_norm_118_cast_fp16")];
            tensor<fp16, [4096, 1024]> p_encoder_layers_23_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(453892224))), lut = tensor<fp16, [256, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(457038016))))[name = string("p_encoder_layers_23_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 4096]> linear_215_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = p_encoder_layers_23_feed_forward2_linear1_weight_to_fp16_palettized, x = layer_norm_118_cast_fp16)[name = string("linear_215_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> silu_71_cast_fp16 = silu(x = linear_215_cast_fp16)[name = string("silu_71_cast_fp16")];
            tensor<fp16, [1024, 4096]> p_encoder_layers_23_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(457070848))), lut = tensor<fp16, [64, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460216640))))[name = string("p_encoder_layers_23_feed_forward2_linear2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 188, 1024]> linear_216_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = p_encoder_layers_23_feed_forward2_linear2_weight_to_fp16_palettized, x = silu_71_cast_fp16)[name = string("linear_216_cast_fp16")];
            fp16 const_1542_to_fp16 = const()[name = string("const_1542_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> mul_77_cast_fp16 = mul(x = linear_216_cast_fp16, y = const_1542_to_fp16)[name = string("mul_77_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> add_158_cast_fp16 = add(x = add_157_cast_fp16, y = mul_77_cast_fp16)[name = string("add_158_cast_fp16")];
            tensor<int32, [1]> layer_norm_119_axes_0 = const()[name = string("layer_norm_119_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> p_encoder_layers_23_norm_out_weight_to_fp16 = const()[name = string("p_encoder_layers_23_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460224896)))];
            tensor<fp16, [1024]> p_encoder_layers_23_norm_out_bias_to_fp16 = const()[name = string("p_encoder_layers_23_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460227008)))];
            fp16 const_1544_to_fp16 = const()[name = string("const_1544_to_fp16"), val = fp16(0x1.8p-23)];
            tensor<fp16, [1, 188, 1024]> layer_norm_119_cast_fp16 = layer_norm(axes = layer_norm_119_axes_0, beta = p_encoder_layers_23_norm_out_bias_to_fp16, epsilon = const_1544_to_fp16, gamma = p_encoder_layers_23_norm_out_weight_to_fp16, x = add_158_cast_fp16)[name = string("layer_norm_119_cast_fp16")];
            tensor<fp16, [640, 1024]> p_projector_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [640, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460229120))), lut = tensor<fp16, [40, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460720704))))[name = string("p_projector_weight_to_fp16_palettized")];
            tensor<fp16, [640]> p_projector_bias_to_fp16 = const()[name = string("p_projector_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460725888)))];
            tensor<fp16, [1, 188, 640]> frames = linear(bias = p_projector_bias_to_fp16, weight = p_projector_weight_to_fp16_palettized, x = layer_norm_119_cast_fp16)[name = string("linear_217_cast_fp16")];
        } -> (frames);
}