program(1.3)
[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})]
{
    func main<ios18>(tensor<fp32, [1, 50, 256]> style_ttl, tensor<int32, [1, 128]> text_ids, tensor<fp32, [1, 1, 128]> text_mask) {
            int32 var_29 = const()[name = string("op_29"), val = int32(-1)];
            int32 x_1_batch_dims_0 = const()[name = string("x_1_batch_dims_0"), val = int32(0)];
            bool x_1_validate_indices_0 = const()[name = string("x_1_validate_indices_0"), val = bool(false)];
            tensor<fp16, [8322, 256]> m_char_embedder_weight_to_fp16 = const()[name = string("m_char_embedder_weight_to_fp16"), val = tensor<fp16, [8322, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
            string text_ids_to_int16_dtype_0 = const()[name = string("text_ids_to_int16_dtype_0"), val = string("int16")];
            string cast_45_dtype_0 = const()[name = string("cast_45_dtype_0"), val = string("int32")];
            int32 greater_equal_0_y_0 = const()[name = string("greater_equal_0_y_0"), val = int32(0)];
            tensor<int16, [1, 128]> text_ids_to_int16 = cast(dtype = text_ids_to_int16_dtype_0, x = text_ids)[name = string("cast_51")];
            tensor<int32, [1, 128]> cast_45 = cast(dtype = cast_45_dtype_0, x = text_ids_to_int16)[name = string("cast_50")];
            tensor<bool, [1, 128]> greater_equal_0 = greater_equal(x = cast_45, y = greater_equal_0_y_0)[name = string("greater_equal_0")];
            int32 slice_by_index_0 = const()[name = string("slice_by_index_0"), val = int32(8322)];
            tensor<int32, [1, 128]> add_0 = add(x = cast_45, y = slice_by_index_0)[name = string("add_0")];
            tensor<int32, [1, 128]> select_0 = select(a = cast_45, b = add_0, cond = greater_equal_0)[name = string("select_0")];
            int32 x_1_cast_fp16_cast_uint16_axis_0 = const()[name = string("x_1_cast_fp16_cast_uint16_axis_0"), val = int32(0)];
            string select_0_to_int16_dtype_0 = const()[name = string("select_0_to_int16_dtype_0"), val = string("int16")];
            tensor<int16, [1, 128]> select_0_to_int16 = cast(dtype = select_0_to_int16_dtype_0, x = select_0)[name = string("cast_49")];
            tensor<fp16, [1, 128, 256]> x_1_cast_fp16_cast_uint16_cast_uint16 = gather(axis = x_1_cast_fp16_cast_uint16_axis_0, batch_dims = x_1_batch_dims_0, indices = select_0_to_int16, validate_indices = x_1_validate_indices_0, x = m_char_embedder_weight_to_fp16)[name = string("x_1_cast_fp16_cast_uint16_cast_uint16")];
            tensor<int32, [3]> var_55_perm_0 = const()[name = string("op_55_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string text_mask_to_fp16_dtype_0 = const()[name = string("text_mask_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 1, 128]> text_mask_to_fp16 = cast(dtype = text_mask_to_fp16_dtype_0, x = text_mask)[name = string("cast_48")];
            tensor<fp16, [1, 256, 128]> var_55_cast_fp16 = transpose(perm = var_55_perm_0, x = x_1_cast_fp16_cast_uint16_cast_uint16)[name = string("transpose_45")];
            tensor<fp16, [1, 256, 128]> x_3_cast_fp16 = mul(x = var_55_cast_fp16, y = text_mask_to_fp16)[name = string("x_3_cast_fp16")];
            tensor<fp16, [1, 256, 128]> input_3_cast_fp16 = mul(x = x_3_cast_fp16, y = text_mask_to_fp16)[name = string("input_3_cast_fp16")];
            tensor<int32, [6]> input_5_pad_0 = const()[name = string("input_5_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_5_mode_0 = const()[name = string("input_5_mode_0"), val = string("replicate")];
            fp16 const_0_to_fp16 = const()[name = string("const_0_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 256, 132]> input_5_cast_fp16 = pad(constant_val = const_0_to_fp16, mode = input_5_mode_0, pad = input_5_pad_0, x = input_3_cast_fp16)[name = string("input_5_cast_fp16")];
            string h_1_pad_type_0 = const()[name = string("h_1_pad_type_0"), val = string("valid")];
            int32 h_1_groups_0 = const()[name = string("h_1_groups_0"), val = int32(256)];
            tensor<int32, [1]> h_1_strides_0 = const()[name = string("h_1_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_1_pad_0 = const()[name = string("h_1_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_1_dilations_0 = const()[name = string("h_1_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [256, 1, 5]> m_convnext_0_dwconv__conv_weight_to_fp16 = const()[name = string("m_convnext_0_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [256, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4260992)))];
            tensor<fp16, [256]> m_convnext_0_dwconv__conv_bias_to_fp16 = const()[name = string("m_convnext_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4263616)))];
            tensor<fp16, [1, 256, 128]> h_1_cast_fp16 = conv(bias = m_convnext_0_dwconv__conv_bias_to_fp16, dilations = h_1_dilations_0, groups = h_1_groups_0, pad = h_1_pad_0, pad_type = h_1_pad_type_0, strides = h_1_strides_0, weight = m_convnext_0_dwconv__conv_weight_to_fp16, x = input_5_cast_fp16)[name = string("h_1_cast_fp16")];
            tensor<fp16, [1, 256, 128]> x_5_cast_fp16 = mul(x = h_1_cast_fp16, y = text_mask_to_fp16)[name = string("x_5_cast_fp16")];
            tensor<int32, [3]> input_7_perm_0 = const()[name = string("input_7_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_79_axes_0 = const()[name = string("op_79_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [256]> m_convnext_0_norm_norm_weight_to_fp16 = const()[name = string("m_convnext_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4264192)))];
            tensor<fp16, [256]> m_convnext_0_norm_norm_bias_to_fp16 = const()[name = string("m_convnext_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4264768)))];
            fp16 var_25_to_fp16 = const()[name = string("op_25_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 128, 256]> input_7_cast_fp16 = transpose(perm = input_7_perm_0, x = x_5_cast_fp16)[name = string("transpose_44")];
            tensor<fp16, [1, 128, 256]> var_79_cast_fp16 = layer_norm(axes = var_79_axes_0, beta = m_convnext_0_norm_norm_bias_to_fp16, epsilon = var_25_to_fp16, gamma = m_convnext_0_norm_norm_weight_to_fp16, x = input_7_cast_fp16)[name = string("op_79_cast_fp16")];
            tensor<int32, [3]> input_9_perm_0 = const()[name = string("input_9_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_3_pad_type_0 = const()[name = string("h_3_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_3_strides_0 = const()[name = string("h_3_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_3_pad_0 = const()[name = string("h_3_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_3_dilations_0 = const()[name = string("h_3_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_3_groups_0 = const()[name = string("h_3_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 256, 1]> m_convnext_0_pwconv1_weight_to_fp16 = const()[name = string("m_convnext_0_pwconv1_weight_to_fp16"), val = tensor<fp16, [1024, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4265344)))];
            tensor<fp16, [1024]> m_convnext_0_pwconv1_bias_to_fp16 = const()[name = string("m_convnext_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4789696)))];
            tensor<fp16, [1, 256, 128]> input_9_cast_fp16 = transpose(perm = input_9_perm_0, x = var_79_cast_fp16)[name = string("transpose_43")];
            tensor<fp16, [1, 1024, 128]> h_3_cast_fp16 = conv(bias = m_convnext_0_pwconv1_bias_to_fp16, dilations = h_3_dilations_0, groups = h_3_groups_0, pad = h_3_pad_0, pad_type = h_3_pad_type_0, strides = h_3_strides_0, weight = m_convnext_0_pwconv1_weight_to_fp16, x = input_9_cast_fp16)[name = string("h_3_cast_fp16")];
            string input_11_mode_0 = const()[name = string("input_11_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 1024, 128]> input_11_cast_fp16 = gelu(mode = input_11_mode_0, x = h_3_cast_fp16)[name = string("input_11_cast_fp16")];
            string h_5_pad_type_0 = const()[name = string("h_5_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_5_strides_0 = const()[name = string("h_5_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_5_pad_0 = const()[name = string("h_5_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_5_dilations_0 = const()[name = string("h_5_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_5_groups_0 = const()[name = string("h_5_groups_0"), val = int32(1)];
            tensor<fp16, [256, 1024, 1]> var_96_weight_0_to_fp16 = const()[name = string("op_96_weight_0_to_fp16"), val = tensor<fp16, [256, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4791808)))];
            tensor<fp16, [256]> var_96_bias_0_to_fp16 = const()[name = string("op_96_bias_0_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5316160)))];
            tensor<fp16, [1, 256, 128]> var_96_cast_fp16 = conv(bias = var_96_bias_0_to_fp16, dilations = h_5_dilations_0, groups = h_5_groups_0, pad = h_5_pad_0, pad_type = h_5_pad_type_0, strides = h_5_strides_0, weight = var_96_weight_0_to_fp16, x = input_11_cast_fp16)[name = string("op_96_cast_fp16")];
            tensor<fp16, [1, 256, 128]> out_1_cast_fp16 = add(x = input_3_cast_fp16, y = var_96_cast_fp16)[name = string("out_1_cast_fp16")];
            tensor<fp16, [1, 256, 128]> x_7_cast_fp16 = mul(x = out_1_cast_fp16, y = text_mask_to_fp16)[name = string("x_7_cast_fp16")];
            tensor<fp16, [1, 256, 128]> input_13_cast_fp16 = mul(x = x_7_cast_fp16, y = text_mask_to_fp16)[name = string("input_13_cast_fp16")];
            tensor<int32, [6]> input_15_pad_0 = const()[name = string("input_15_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_15_mode_0 = const()[name = string("input_15_mode_0"), val = string("replicate")];
            fp16 const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 256, 132]> input_15_cast_fp16 = pad(constant_val = const_1_to_fp16, mode = input_15_mode_0, pad = input_15_pad_0, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")];
            string h_7_pad_type_0 = const()[name = string("h_7_pad_type_0"), val = string("valid")];
            int32 h_7_groups_0 = const()[name = string("h_7_groups_0"), val = int32(256)];
            tensor<int32, [1]> h_7_strides_0 = const()[name = string("h_7_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_7_pad_0 = const()[name = string("h_7_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_7_dilations_0 = const()[name = string("h_7_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [256, 1, 5]> m_convnext_1_dwconv__conv_weight_to_fp16 = const()[name = string("m_convnext_1_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [256, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5316736)))];
            tensor<fp16, [256]> m_convnext_1_dwconv__conv_bias_to_fp16 = const()[name = string("m_convnext_1_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5319360)))];
            tensor<fp16, [1, 256, 128]> h_7_cast_fp16 = conv(bias = m_convnext_1_dwconv__conv_bias_to_fp16, dilations = h_7_dilations_0, groups = h_7_groups_0, pad = h_7_pad_0, pad_type = h_7_pad_type_0, strides = h_7_strides_0, weight = m_convnext_1_dwconv__conv_weight_to_fp16, x = input_15_cast_fp16)[name = string("h_7_cast_fp16")];
            tensor<fp16, [1, 256, 128]> x_9_cast_fp16 = mul(x = h_7_cast_fp16, y = text_mask_to_fp16)[name = string("x_9_cast_fp16")];
            tensor<int32, [3]> input_17_perm_0 = const()[name = string("input_17_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_121_axes_0 = const()[name = string("op_121_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [256]> m_convnext_1_norm_norm_weight_to_fp16 = const()[name = string("m_convnext_1_norm_norm_weight_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5319936)))];
            tensor<fp16, [256]> m_convnext_1_norm_norm_bias_to_fp16 = const()[name = string("m_convnext_1_norm_norm_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5320512)))];
            tensor<fp16, [1, 128, 256]> input_17_cast_fp16 = transpose(perm = input_17_perm_0, x = x_9_cast_fp16)[name = string("transpose_42")];
            tensor<fp16, [1, 128, 256]> var_121_cast_fp16 = layer_norm(axes = var_121_axes_0, beta = m_convnext_1_norm_norm_bias_to_fp16, epsilon = var_25_to_fp16, gamma = m_convnext_1_norm_norm_weight_to_fp16, x = input_17_cast_fp16)[name = string("op_121_cast_fp16")];
            tensor<int32, [3]> input_19_perm_0 = const()[name = string("input_19_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_9_pad_type_0 = const()[name = string("h_9_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_9_strides_0 = const()[name = string("h_9_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_9_pad_0 = const()[name = string("h_9_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_9_dilations_0 = const()[name = string("h_9_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_9_groups_0 = const()[name = string("h_9_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 256, 1]> m_convnext_1_pwconv1_weight_to_fp16 = const()[name = string("m_convnext_1_pwconv1_weight_to_fp16"), val = tensor<fp16, [1024, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5321088)))];
            tensor<fp16, [1024]> m_convnext_1_pwconv1_bias_to_fp16 = const()[name = string("m_convnext_1_pwconv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5845440)))];
            tensor<fp16, [1, 256, 128]> input_19_cast_fp16 = transpose(perm = input_19_perm_0, x = var_121_cast_fp16)[name = string("transpose_41")];
            tensor<fp16, [1, 1024, 128]> h_9_cast_fp16 = conv(bias = m_convnext_1_pwconv1_bias_to_fp16, dilations = h_9_dilations_0, groups = h_9_groups_0, pad = h_9_pad_0, pad_type = h_9_pad_type_0, strides = h_9_strides_0, weight = m_convnext_1_pwconv1_weight_to_fp16, x = input_19_cast_fp16)[name = string("h_9_cast_fp16")];
            string input_21_mode_0 = const()[name = string("input_21_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 1024, 128]> input_21_cast_fp16 = gelu(mode = input_21_mode_0, x = h_9_cast_fp16)[name = string("input_21_cast_fp16")];
            string h_11_pad_type_0 = const()[name = string("h_11_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_11_strides_0 = const()[name = string("h_11_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_11_pad_0 = const()[name = string("h_11_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_11_dilations_0 = const()[name = string("h_11_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_11_groups_0 = const()[name = string("h_11_groups_0"), val = int32(1)];
            tensor<fp16, [256, 1024, 1]> var_138_weight_0_to_fp16 = const()[name = string("op_138_weight_0_to_fp16"), val = tensor<fp16, [256, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5847552)))];
            tensor<fp16, [256]> var_138_bias_0_to_fp16 = const()[name = string("op_138_bias_0_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6371904)))];
            tensor<fp16, [1, 256, 128]> var_138_cast_fp16 = conv(bias = var_138_bias_0_to_fp16, dilations = h_11_dilations_0, groups = h_11_groups_0, pad = h_11_pad_0, pad_type = h_11_pad_type_0, strides = h_11_strides_0, weight = var_138_weight_0_to_fp16, x = input_21_cast_fp16)[name = string("op_138_cast_fp16")];
            tensor<fp16, [1, 256, 128]> out_3_cast_fp16 = add(x = input_13_cast_fp16, y = var_138_cast_fp16)[name = string("out_3_cast_fp16")];
            tensor<fp16, [1, 256, 128]> x_11_cast_fp16 = mul(x = out_3_cast_fp16, y = text_mask_to_fp16)[name = string("x_11_cast_fp16")];
            tensor<fp16, [1, 256, 128]> input_23_cast_fp16 = mul(x = x_11_cast_fp16, y = text_mask_to_fp16)[name = string("input_23_cast_fp16")];
            tensor<int32, [6]> input_25_pad_0 = const()[name = string("input_25_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_25_mode_0 = const()[name = string("input_25_mode_0"), val = string("replicate")];
            fp16 const_2_to_fp16 = const()[name = string("const_2_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 256, 136]> input_25_cast_fp16 = pad(constant_val = const_2_to_fp16, mode = input_25_mode_0, pad = input_25_pad_0, x = input_23_cast_fp16)[name = string("input_25_cast_fp16")];
            string h_13_pad_type_0 = const()[name = string("h_13_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_13_dilations_0 = const()[name = string("h_13_dilations_0"), val = tensor<int32, [1]>([2])];
            int32 h_13_groups_0 = const()[name = string("h_13_groups_0"), val = int32(256)];
            tensor<int32, [1]> h_13_strides_0 = const()[name = string("h_13_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_13_pad_0 = const()[name = string("h_13_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [256, 1, 5]> m_convnext_2_dwconv__conv_weight_to_fp16 = const()[name = string("m_convnext_2_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [256, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6372480)))];
            tensor<fp16, [256]> m_convnext_2_dwconv__conv_bias_to_fp16 = const()[name = string("m_convnext_2_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6375104)))];
            tensor<fp16, [1, 256, 128]> h_13_cast_fp16 = conv(bias = m_convnext_2_dwconv__conv_bias_to_fp16, dilations = h_13_dilations_0, groups = h_13_groups_0, pad = h_13_pad_0, pad_type = h_13_pad_type_0, strides = h_13_strides_0, weight = m_convnext_2_dwconv__conv_weight_to_fp16, x = input_25_cast_fp16)[name = string("h_13_cast_fp16")];
            tensor<fp16, [1, 256, 128]> x_13_cast_fp16 = mul(x = h_13_cast_fp16, y = text_mask_to_fp16)[name = string("x_13_cast_fp16")];
            tensor<int32, [3]> input_27_perm_0 = const()[name = string("input_27_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_163_axes_0 = const()[name = string("op_163_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [256]> m_convnext_2_norm_norm_weight_to_fp16 = const()[name = string("m_convnext_2_norm_norm_weight_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6375680)))];
            tensor<fp16, [256]> m_convnext_2_norm_norm_bias_to_fp16 = const()[name = string("m_convnext_2_norm_norm_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6376256)))];
            tensor<fp16, [1, 128, 256]> input_27_cast_fp16 = transpose(perm = input_27_perm_0, x = x_13_cast_fp16)[name = string("transpose_40")];
            tensor<fp16, [1, 128, 256]> var_163_cast_fp16 = layer_norm(axes = var_163_axes_0, beta = m_convnext_2_norm_norm_bias_to_fp16, epsilon = var_25_to_fp16, gamma = m_convnext_2_norm_norm_weight_to_fp16, x = input_27_cast_fp16)[name = string("op_163_cast_fp16")];
            tensor<int32, [3]> input_29_perm_0 = const()[name = string("input_29_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_15_pad_type_0 = const()[name = string("h_15_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_15_strides_0 = const()[name = string("h_15_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_15_pad_0 = const()[name = string("h_15_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_15_dilations_0 = const()[name = string("h_15_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_15_groups_0 = const()[name = string("h_15_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 256, 1]> m_convnext_2_pwconv1_weight_to_fp16 = const()[name = string("m_convnext_2_pwconv1_weight_to_fp16"), val = tensor<fp16, [1024, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6376832)))];
            tensor<fp16, [1024]> m_convnext_2_pwconv1_bias_to_fp16 = const()[name = string("m_convnext_2_pwconv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6901184)))];
            tensor<fp16, [1, 256, 128]> input_29_cast_fp16 = transpose(perm = input_29_perm_0, x = var_163_cast_fp16)[name = string("transpose_39")];
            tensor<fp16, [1, 1024, 128]> h_15_cast_fp16 = conv(bias = m_convnext_2_pwconv1_bias_to_fp16, dilations = h_15_dilations_0, groups = h_15_groups_0, pad = h_15_pad_0, pad_type = h_15_pad_type_0, strides = h_15_strides_0, weight = m_convnext_2_pwconv1_weight_to_fp16, x = input_29_cast_fp16)[name = string("h_15_cast_fp16")];
            string input_31_mode_0 = const()[name = string("input_31_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 1024, 128]> input_31_cast_fp16 = gelu(mode = input_31_mode_0, x = h_15_cast_fp16)[name = string("input_31_cast_fp16")];
            string h_17_pad_type_0 = const()[name = string("h_17_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_17_strides_0 = const()[name = string("h_17_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_17_pad_0 = const()[name = string("h_17_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_17_dilations_0 = const()[name = string("h_17_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_17_groups_0 = const()[name = string("h_17_groups_0"), val = int32(1)];
            tensor<fp16, [256, 1024, 1]> var_180_weight_0_to_fp16 = const()[name = string("op_180_weight_0_to_fp16"), val = tensor<fp16, [256, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6903296)))];
            tensor<fp16, [256]> var_180_bias_0_to_fp16 = const()[name = string("op_180_bias_0_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7427648)))];
            tensor<fp16, [1, 256, 128]> var_180_cast_fp16 = conv(bias = var_180_bias_0_to_fp16, dilations = h_17_dilations_0, groups = h_17_groups_0, pad = h_17_pad_0, pad_type = h_17_pad_type_0, strides = h_17_strides_0, weight = var_180_weight_0_to_fp16, x = input_31_cast_fp16)[name = string("op_180_cast_fp16")];
            tensor<fp16, [1, 256, 128]> out_5_cast_fp16 = add(x = input_23_cast_fp16, y = var_180_cast_fp16)[name = string("out_5_cast_fp16")];
            tensor<fp16, [1, 256, 128]> x_15_cast_fp16 = mul(x = out_5_cast_fp16, y = text_mask_to_fp16)[name = string("x_15_cast_fp16")];
            tensor<fp16, [1, 256, 128]> input_33_cast_fp16 = mul(x = x_15_cast_fp16, y = text_mask_to_fp16)[name = string("input_33_cast_fp16")];
            tensor<int32, [6]> input_35_pad_0 = const()[name = string("input_35_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_35_mode_0 = const()[name = string("input_35_mode_0"), val = string("replicate")];
            fp16 const_3_to_fp16 = const()[name = string("const_3_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 256, 136]> input_35_cast_fp16 = pad(constant_val = const_3_to_fp16, mode = input_35_mode_0, pad = input_35_pad_0, x = input_33_cast_fp16)[name = string("input_35_cast_fp16")];
            string h_19_pad_type_0 = const()[name = string("h_19_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_19_dilations_0 = const()[name = string("h_19_dilations_0"), val = tensor<int32, [1]>([2])];
            int32 h_19_groups_0 = const()[name = string("h_19_groups_0"), val = int32(256)];
            tensor<int32, [1]> h_19_strides_0 = const()[name = string("h_19_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_19_pad_0 = const()[name = string("h_19_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [256, 1, 5]> m_convnext_3_dwconv__conv_weight_to_fp16 = const()[name = string("m_convnext_3_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [256, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7428224)))];
            tensor<fp16, [256]> m_convnext_3_dwconv__conv_bias_to_fp16 = const()[name = string("m_convnext_3_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7430848)))];
            tensor<fp16, [1, 256, 128]> h_19_cast_fp16 = conv(bias = m_convnext_3_dwconv__conv_bias_to_fp16, dilations = h_19_dilations_0, groups = h_19_groups_0, pad = h_19_pad_0, pad_type = h_19_pad_type_0, strides = h_19_strides_0, weight = m_convnext_3_dwconv__conv_weight_to_fp16, x = input_35_cast_fp16)[name = string("h_19_cast_fp16")];
            tensor<fp16, [1, 256, 128]> x_17_cast_fp16 = mul(x = h_19_cast_fp16, y = text_mask_to_fp16)[name = string("x_17_cast_fp16")];
            tensor<int32, [3]> input_37_perm_0 = const()[name = string("input_37_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_205_axes_0 = const()[name = string("op_205_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [256]> m_convnext_3_norm_norm_weight_to_fp16 = const()[name = string("m_convnext_3_norm_norm_weight_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7431424)))];
            tensor<fp16, [256]> m_convnext_3_norm_norm_bias_to_fp16 = const()[name = string("m_convnext_3_norm_norm_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7432000)))];
            tensor<fp16, [1, 128, 256]> input_37_cast_fp16 = transpose(perm = input_37_perm_0, x = x_17_cast_fp16)[name = string("transpose_38")];
            tensor<fp16, [1, 128, 256]> var_205_cast_fp16 = layer_norm(axes = var_205_axes_0, beta = m_convnext_3_norm_norm_bias_to_fp16, epsilon = var_25_to_fp16, gamma = m_convnext_3_norm_norm_weight_to_fp16, x = input_37_cast_fp16)[name = string("op_205_cast_fp16")];
            tensor<int32, [3]> input_39_perm_0 = const()[name = string("input_39_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_21_pad_type_0 = const()[name = string("h_21_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_21_strides_0 = const()[name = string("h_21_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_21_pad_0 = const()[name = string("h_21_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_21_dilations_0 = const()[name = string("h_21_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_21_groups_0 = const()[name = string("h_21_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 256, 1]> m_convnext_3_pwconv1_weight_to_fp16 = const()[name = string("m_convnext_3_pwconv1_weight_to_fp16"), val = tensor<fp16, [1024, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7432576)))];
            tensor<fp16, [1024]> m_convnext_3_pwconv1_bias_to_fp16 = const()[name = string("m_convnext_3_pwconv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7956928)))];
            tensor<fp16, [1, 256, 128]> input_39_cast_fp16 = transpose(perm = input_39_perm_0, x = var_205_cast_fp16)[name = string("transpose_37")];
            tensor<fp16, [1, 1024, 128]> h_21_cast_fp16 = conv(bias = m_convnext_3_pwconv1_bias_to_fp16, dilations = h_21_dilations_0, groups = h_21_groups_0, pad = h_21_pad_0, pad_type = h_21_pad_type_0, strides = h_21_strides_0, weight = m_convnext_3_pwconv1_weight_to_fp16, x = input_39_cast_fp16)[name = string("h_21_cast_fp16")];
            string input_41_mode_0 = const()[name = string("input_41_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 1024, 128]> input_41_cast_fp16 = gelu(mode = input_41_mode_0, x = h_21_cast_fp16)[name = string("input_41_cast_fp16")];
            string h_23_pad_type_0 = const()[name = string("h_23_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_23_strides_0 = const()[name = string("h_23_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_23_pad_0 = const()[name = string("h_23_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_23_dilations_0 = const()[name = string("h_23_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_23_groups_0 = const()[name = string("h_23_groups_0"), val = int32(1)];
            tensor<fp16, [256, 1024, 1]> var_222_weight_0_to_fp16 = const()[name = string("op_222_weight_0_to_fp16"), val = tensor<fp16, [256, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7959040)))];
            tensor<fp16, [256]> var_222_bias_0_to_fp16 = const()[name = string("op_222_bias_0_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8483392)))];
            tensor<fp16, [1, 256, 128]> var_222_cast_fp16 = conv(bias = var_222_bias_0_to_fp16, dilations = h_23_dilations_0, groups = h_23_groups_0, pad = h_23_pad_0, pad_type = h_23_pad_type_0, strides = h_23_strides_0, weight = var_222_weight_0_to_fp16, x = input_41_cast_fp16)[name = string("op_222_cast_fp16")];
            tensor<fp16, [1, 256, 128]> out_7_cast_fp16 = add(x = input_33_cast_fp16, y = var_222_cast_fp16)[name = string("out_7_cast_fp16")];
            tensor<fp16, [1, 256, 128]> x_19_cast_fp16 = mul(x = out_7_cast_fp16, y = text_mask_to_fp16)[name = string("x_19_cast_fp16")];
            tensor<fp16, [1, 256, 128]> input_43_cast_fp16 = mul(x = x_19_cast_fp16, y = text_mask_to_fp16)[name = string("input_43_cast_fp16")];
            tensor<int32, [6]> input_45_pad_0 = const()[name = string("input_45_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 8, 8])];
            string input_45_mode_0 = const()[name = string("input_45_mode_0"), val = string("replicate")];
            fp16 const_4_to_fp16 = const()[name = string("const_4_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 256, 144]> input_45_cast_fp16 = pad(constant_val = const_4_to_fp16, mode = input_45_mode_0, pad = input_45_pad_0, x = input_43_cast_fp16)[name = string("input_45_cast_fp16")];
            string h_25_pad_type_0 = const()[name = string("h_25_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_25_dilations_0 = const()[name = string("h_25_dilations_0"), val = tensor<int32, [1]>([4])];
            int32 h_25_groups_0 = const()[name = string("h_25_groups_0"), val = int32(256)];
            tensor<int32, [1]> h_25_strides_0 = const()[name = string("h_25_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_25_pad_0 = const()[name = string("h_25_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [256, 1, 5]> m_convnext_4_dwconv__conv_weight_to_fp16 = const()[name = string("m_convnext_4_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [256, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8483968)))];
            tensor<fp16, [256]> m_convnext_4_dwconv__conv_bias_to_fp16 = const()[name = string("m_convnext_4_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8486592)))];
            tensor<fp16, [1, 256, 128]> h_25_cast_fp16 = conv(bias = m_convnext_4_dwconv__conv_bias_to_fp16, dilations = h_25_dilations_0, groups = h_25_groups_0, pad = h_25_pad_0, pad_type = h_25_pad_type_0, strides = h_25_strides_0, weight = m_convnext_4_dwconv__conv_weight_to_fp16, x = input_45_cast_fp16)[name = string("h_25_cast_fp16")];
            tensor<fp16, [1, 256, 128]> x_21_cast_fp16 = mul(x = h_25_cast_fp16, y = text_mask_to_fp16)[name = string("x_21_cast_fp16")];
            tensor<int32, [3]> input_47_perm_0 = const()[name = string("input_47_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_247_axes_0 = const()[name = string("op_247_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [256]> m_convnext_4_norm_norm_weight_to_fp16 = const()[name = string("m_convnext_4_norm_norm_weight_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8487168)))];
            tensor<fp16, [256]> m_convnext_4_norm_norm_bias_to_fp16 = const()[name = string("m_convnext_4_norm_norm_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8487744)))];
            tensor<fp16, [1, 128, 256]> input_47_cast_fp16 = transpose(perm = input_47_perm_0, x = x_21_cast_fp16)[name = string("transpose_36")];
            tensor<fp16, [1, 128, 256]> var_247_cast_fp16 = layer_norm(axes = var_247_axes_0, beta = m_convnext_4_norm_norm_bias_to_fp16, epsilon = var_25_to_fp16, gamma = m_convnext_4_norm_norm_weight_to_fp16, x = input_47_cast_fp16)[name = string("op_247_cast_fp16")];
            tensor<int32, [3]> input_49_perm_0 = const()[name = string("input_49_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_27_pad_type_0 = const()[name = string("h_27_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_27_strides_0 = const()[name = string("h_27_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_27_pad_0 = const()[name = string("h_27_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_27_dilations_0 = const()[name = string("h_27_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_27_groups_0 = const()[name = string("h_27_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 256, 1]> m_convnext_4_pwconv1_weight_to_fp16 = const()[name = string("m_convnext_4_pwconv1_weight_to_fp16"), val = tensor<fp16, [1024, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8488320)))];
            tensor<fp16, [1024]> m_convnext_4_pwconv1_bias_to_fp16 = const()[name = string("m_convnext_4_pwconv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9012672)))];
            tensor<fp16, [1, 256, 128]> input_49_cast_fp16 = transpose(perm = input_49_perm_0, x = var_247_cast_fp16)[name = string("transpose_35")];
            tensor<fp16, [1, 1024, 128]> h_27_cast_fp16 = conv(bias = m_convnext_4_pwconv1_bias_to_fp16, dilations = h_27_dilations_0, groups = h_27_groups_0, pad = h_27_pad_0, pad_type = h_27_pad_type_0, strides = h_27_strides_0, weight = m_convnext_4_pwconv1_weight_to_fp16, x = input_49_cast_fp16)[name = string("h_27_cast_fp16")];
            string input_51_mode_0 = const()[name = string("input_51_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 1024, 128]> input_51_cast_fp16 = gelu(mode = input_51_mode_0, x = h_27_cast_fp16)[name = string("input_51_cast_fp16")];
            string h_29_pad_type_0 = const()[name = string("h_29_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_29_strides_0 = const()[name = string("h_29_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_29_pad_0 = const()[name = string("h_29_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_29_dilations_0 = const()[name = string("h_29_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_29_groups_0 = const()[name = string("h_29_groups_0"), val = int32(1)];
            tensor<fp16, [256, 1024, 1]> var_264_weight_0_to_fp16 = const()[name = string("op_264_weight_0_to_fp16"), val = tensor<fp16, [256, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9014784)))];
            tensor<fp16, [256]> var_264_bias_0_to_fp16 = const()[name = string("op_264_bias_0_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9539136)))];
            tensor<fp16, [1, 256, 128]> var_264_cast_fp16 = conv(bias = var_264_bias_0_to_fp16, dilations = h_29_dilations_0, groups = h_29_groups_0, pad = h_29_pad_0, pad_type = h_29_pad_type_0, strides = h_29_strides_0, weight = var_264_weight_0_to_fp16, x = input_51_cast_fp16)[name = string("op_264_cast_fp16")];
            tensor<fp16, [1, 256, 128]> out_9_cast_fp16 = add(x = input_43_cast_fp16, y = var_264_cast_fp16)[name = string("out_9_cast_fp16")];
            tensor<fp16, [1, 256, 128]> x_23_cast_fp16 = mul(x = out_9_cast_fp16, y = text_mask_to_fp16)[name = string("x_23_cast_fp16")];
            tensor<fp16, [1, 256, 128]> input_53_cast_fp16 = mul(x = x_23_cast_fp16, y = text_mask_to_fp16)[name = string("input_53_cast_fp16")];
            tensor<int32, [6]> input_55_pad_0 = const()[name = string("input_55_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 8, 8])];
            string input_55_mode_0 = const()[name = string("input_55_mode_0"), val = string("replicate")];
            fp16 const_5_to_fp16 = const()[name = string("const_5_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 256, 144]> input_55_cast_fp16 = pad(constant_val = const_5_to_fp16, mode = input_55_mode_0, pad = input_55_pad_0, x = input_53_cast_fp16)[name = string("input_55_cast_fp16")];
            string h_31_pad_type_0 = const()[name = string("h_31_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_31_dilations_0 = const()[name = string("h_31_dilations_0"), val = tensor<int32, [1]>([4])];
            int32 h_31_groups_0 = const()[name = string("h_31_groups_0"), val = int32(256)];
            tensor<int32, [1]> h_31_strides_0 = const()[name = string("h_31_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_31_pad_0 = const()[name = string("h_31_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [256, 1, 5]> m_convnext_5_dwconv__conv_weight_to_fp16 = const()[name = string("m_convnext_5_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [256, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9539712)))];
            tensor<fp16, [256]> m_convnext_5_dwconv__conv_bias_to_fp16 = const()[name = string("m_convnext_5_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9542336)))];
            tensor<fp16, [1, 256, 128]> h_31_cast_fp16 = conv(bias = m_convnext_5_dwconv__conv_bias_to_fp16, dilations = h_31_dilations_0, groups = h_31_groups_0, pad = h_31_pad_0, pad_type = h_31_pad_type_0, strides = h_31_strides_0, weight = m_convnext_5_dwconv__conv_weight_to_fp16, x = input_55_cast_fp16)[name = string("h_31_cast_fp16")];
            tensor<fp16, [1, 256, 128]> x_25_cast_fp16 = mul(x = h_31_cast_fp16, y = text_mask_to_fp16)[name = string("x_25_cast_fp16")];
            tensor<int32, [3]> input_57_perm_0 = const()[name = string("input_57_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_289_axes_0 = const()[name = string("op_289_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [256]> m_convnext_5_norm_norm_weight_to_fp16 = const()[name = string("m_convnext_5_norm_norm_weight_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9542912)))];
            tensor<fp16, [256]> m_convnext_5_norm_norm_bias_to_fp16 = const()[name = string("m_convnext_5_norm_norm_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9543488)))];
            tensor<fp16, [1, 128, 256]> input_57_cast_fp16 = transpose(perm = input_57_perm_0, x = x_25_cast_fp16)[name = string("transpose_34")];
            tensor<fp16, [1, 128, 256]> var_289_cast_fp16 = layer_norm(axes = var_289_axes_0, beta = m_convnext_5_norm_norm_bias_to_fp16, epsilon = var_25_to_fp16, gamma = m_convnext_5_norm_norm_weight_to_fp16, x = input_57_cast_fp16)[name = string("op_289_cast_fp16")];
            tensor<int32, [3]> input_59_perm_0 = const()[name = string("input_59_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_33_pad_type_0 = const()[name = string("h_33_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_33_strides_0 = const()[name = string("h_33_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_33_pad_0 = const()[name = string("h_33_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_33_dilations_0 = const()[name = string("h_33_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_33_groups_0 = const()[name = string("h_33_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 256, 1]> m_convnext_5_pwconv1_weight_to_fp16 = const()[name = string("m_convnext_5_pwconv1_weight_to_fp16"), val = tensor<fp16, [1024, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9544064)))];
            tensor<fp16, [1024]> m_convnext_5_pwconv1_bias_to_fp16 = const()[name = string("m_convnext_5_pwconv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10068416)))];
            tensor<fp16, [1, 256, 128]> input_59_cast_fp16 = transpose(perm = input_59_perm_0, x = var_289_cast_fp16)[name = string("transpose_33")];
            tensor<fp16, [1, 1024, 128]> h_33_cast_fp16 = conv(bias = m_convnext_5_pwconv1_bias_to_fp16, dilations = h_33_dilations_0, groups = h_33_groups_0, pad = h_33_pad_0, pad_type = h_33_pad_type_0, strides = h_33_strides_0, weight = m_convnext_5_pwconv1_weight_to_fp16, x = input_59_cast_fp16)[name = string("h_33_cast_fp16")];
            string input_61_mode_0 = const()[name = string("input_61_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 1024, 128]> input_61_cast_fp16 = gelu(mode = input_61_mode_0, x = h_33_cast_fp16)[name = string("input_61_cast_fp16")];
            string h_35_pad_type_0 = const()[name = string("h_35_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_35_strides_0 = const()[name = string("h_35_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_35_pad_0 = const()[name = string("h_35_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_35_dilations_0 = const()[name = string("h_35_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_35_groups_0 = const()[name = string("h_35_groups_0"), val = int32(1)];
            tensor<fp16, [256, 1024, 1]> var_306_weight_0_to_fp16 = const()[name = string("op_306_weight_0_to_fp16"), val = tensor<fp16, [256, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10070528)))];
            tensor<fp16, [256]> var_306_bias_0_to_fp16 = const()[name = string("op_306_bias_0_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10594880)))];
            tensor<fp16, [1, 256, 128]> var_306_cast_fp16 = conv(bias = var_306_bias_0_to_fp16, dilations = h_35_dilations_0, groups = h_35_groups_0, pad = h_35_pad_0, pad_type = h_35_pad_type_0, strides = h_35_strides_0, weight = var_306_weight_0_to_fp16, x = input_61_cast_fp16)[name = string("op_306_cast_fp16")];
            tensor<fp16, [1, 256, 128]> out_11_cast_fp16 = add(x = input_53_cast_fp16, y = var_306_cast_fp16)[name = string("out_11_cast_fp16")];
            tensor<fp16, [1, 256, 128]> x_27_cast_fp16 = mul(x = out_11_cast_fp16, y = text_mask_to_fp16)[name = string("x_27_cast_fp16")];
            tensor<fp16, [1, 256, 128]> x_29_cast_fp16 = mul(x = x_27_cast_fp16, y = text_mask_to_fp16)[name = string("x_29_cast_fp16")];
            string var_329_pad_type_0 = const()[name = string("op_329_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_329_strides_0 = const()[name = string("op_329_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_329_pad_0 = const()[name = string("op_329_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_329_dilations_0 = const()[name = string("op_329_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_329_groups_0 = const()[name = string("op_329_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1]> m_attn_layers_0_attn_conv_q_weight_to_fp16 = const()[name = string("m_attn_layers_0_attn_conv_q_weight_to_fp16"), val = tensor<fp16, [256, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10595456)))];
            tensor<fp16, [256]> m_attn_layers_0_attn_conv_q_bias_to_fp16 = const()[name = string("m_attn_layers_0_attn_conv_q_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10726592)))];
            tensor<fp16, [1, 256, 128]> var_329_cast_fp16 = conv(bias = m_attn_layers_0_attn_conv_q_bias_to_fp16, dilations = var_329_dilations_0, groups = var_329_groups_0, pad = var_329_pad_0, pad_type = var_329_pad_type_0, strides = var_329_strides_0, weight = m_attn_layers_0_attn_conv_q_weight_to_fp16, x = x_29_cast_fp16)[name = string("op_329_cast_fp16")];
            tensor<int32, [4]> var_330 = const()[name = string("op_330"), val = tensor<int32, [4]>([1, 4, 64, 128])];
            tensor<fp16, [1, 4, 64, 128]> var_331_cast_fp16 = reshape(shape = var_330, x = var_329_cast_fp16)[name = string("op_331_cast_fp16")];
            tensor<int32, [4]> q_1_perm_0 = const()[name = string("q_1_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            string var_339_pad_type_0 = const()[name = string("op_339_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_339_strides_0 = const()[name = string("op_339_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_339_pad_0 = const()[name = string("op_339_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_339_dilations_0 = const()[name = string("op_339_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_339_groups_0 = const()[name = string("op_339_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1]> m_attn_layers_0_attn_conv_k_weight_to_fp16 = const()[name = string("m_attn_layers_0_attn_conv_k_weight_to_fp16"), val = tensor<fp16, [256, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10727168)))];
            tensor<fp16, [256]> m_attn_layers_0_attn_conv_k_bias_to_fp16 = const()[name = string("m_attn_layers_0_attn_conv_k_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10858304)))];
            tensor<fp16, [1, 256, 128]> var_339_cast_fp16 = conv(bias = m_attn_layers_0_attn_conv_k_bias_to_fp16, dilations = var_339_dilations_0, groups = var_339_groups_0, pad = var_339_pad_0, pad_type = var_339_pad_type_0, strides = var_339_strides_0, weight = m_attn_layers_0_attn_conv_k_weight_to_fp16, x = x_29_cast_fp16)[name = string("op_339_cast_fp16")];
            tensor<int32, [4]> var_340 = const()[name = string("op_340"), val = tensor<int32, [4]>([1, 4, 64, 128])];
            tensor<fp16, [1, 4, 64, 128]> var_341_cast_fp16 = reshape(shape = var_340, x = var_339_cast_fp16)[name = string("op_341_cast_fp16")];
            string var_349_pad_type_0 = const()[name = string("op_349_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_349_strides_0 = const()[name = string("op_349_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_349_pad_0 = const()[name = string("op_349_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_349_dilations_0 = const()[name = string("op_349_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_349_groups_0 = const()[name = string("op_349_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1]> m_attn_layers_0_attn_conv_v_weight_to_fp16 = const()[name = string("m_attn_layers_0_attn_conv_v_weight_to_fp16"), val = tensor<fp16, [256, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10858880)))];
            tensor<fp16, [256]> m_attn_layers_0_attn_conv_v_bias_to_fp16 = const()[name = string("m_attn_layers_0_attn_conv_v_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10990016)))];
            tensor<fp16, [1, 256, 128]> var_349_cast_fp16 = conv(bias = m_attn_layers_0_attn_conv_v_bias_to_fp16, dilations = var_349_dilations_0, groups = var_349_groups_0, pad = var_349_pad_0, pad_type = var_349_pad_type_0, strides = var_349_strides_0, weight = m_attn_layers_0_attn_conv_v_weight_to_fp16, x = x_29_cast_fp16)[name = string("op_349_cast_fp16")];
            tensor<int32, [4]> var_350 = const()[name = string("op_350"), val = tensor<int32, [4]>([1, 4, 64, 128])];
            tensor<fp16, [1, 4, 64, 128]> var_351_cast_fp16 = reshape(shape = var_350, x = var_349_cast_fp16)[name = string("op_351_cast_fp16")];
            fp16 var_353_to_fp16 = const()[name = string("op_353_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 4, 128, 64]> q_1_cast_fp16 = transpose(perm = q_1_perm_0, x = var_331_cast_fp16)[name = string("transpose_32")];
            tensor<fp16, [1, 4, 128, 64]> var_354_cast_fp16 = mul(x = q_1_cast_fp16, y = var_353_to_fp16)[name = string("op_354_cast_fp16")];
            bool scores_1_transpose_x_0 = const()[name = string("scores_1_transpose_x_0"), val = bool(false)];
            bool scores_1_transpose_y_0 = const()[name = string("scores_1_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 4, 128, 128]> scores_1_cast_fp16 = matmul(transpose_x = scores_1_transpose_x_0, transpose_y = scores_1_transpose_y_0, x = var_354_cast_fp16, y = var_341_cast_fp16)[name = string("scores_1_cast_fp16")];
            bool x_31_transpose_x_0 = const()[name = string("x_31_transpose_x_0"), val = bool(false)];
            bool x_31_transpose_y_0 = const()[name = string("x_31_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 1, 64, 255]> var_378_to_fp16 = const()[name = string("op_378_to_fp16"), val = tensor<fp16, [1, 1, 64, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10990592)))];
            tensor<fp16, [1, 4, 128, 255]> x_31_cast_fp16 = matmul(transpose_x = x_31_transpose_x_0, transpose_y = x_31_transpose_y_0, x = var_354_cast_fp16, y = var_378_to_fp16)[name = string("x_31_cast_fp16")];
            tensor<int32, [8]> x_33_pad_0 = const()[name = string("x_33_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 1])];
            string x_33_mode_0 = const()[name = string("x_33_mode_0"), val = string("constant")];
            fp16 const_12_to_fp16 = const()[name = string("const_12_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 128, 256]> x_33_cast_fp16 = pad(constant_val = const_12_to_fp16, mode = x_33_mode_0, pad = x_33_pad_0, x = x_31_cast_fp16)[name = string("x_33_cast_fp16")];
            tensor<int32, [3]> var_390 = const()[name = string("op_390"), val = tensor<int32, [3]>([1, 4, 32768])];
            tensor<fp16, [1, 4, 32768]> input_65_cast_fp16 = reshape(shape = var_390, x = x_33_cast_fp16)[name = string("input_65_cast_fp16")];
            tensor<int32, [6]> x_35_pad_0 = const()[name = string("x_35_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 127])];
            string x_35_mode_0 = const()[name = string("x_35_mode_0"), val = string("constant")];
            fp16 const_13_to_fp16 = const()[name = string("const_13_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 32895]> x_35_cast_fp16 = pad(constant_val = const_13_to_fp16, mode = x_35_mode_0, pad = x_35_pad_0, x = input_65_cast_fp16)[name = string("x_35_cast_fp16")];
            tensor<int32, [4]> var_405 = const()[name = string("op_405"), val = tensor<int32, [4]>([1, 4, 129, 255])];
            tensor<fp16, [1, 4, 129, 255]> x_37_cast_fp16 = reshape(shape = var_405, x = x_35_cast_fp16)[name = string("x_37_cast_fp16")];
            tensor<int32, [4]> var_412_begin_0 = const()[name = string("op_412_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_412_end_0 = const()[name = string("op_412_end_0"), val = tensor<int32, [4]>([1, 4, 128, 255])];
            tensor<bool, [4]> var_412_end_mask_0 = const()[name = string("op_412_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])];
            tensor<fp16, [1, 4, 128, 255]> var_412_cast_fp16 = slice_by_index(begin = var_412_begin_0, end = var_412_end_0, end_mask = var_412_end_mask_0, x = x_37_cast_fp16)[name = string("op_412_cast_fp16")];
            tensor<int32, [4]> var_413_begin_0 = const()[name = string("op_413_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 127])];
            tensor<int32, [4]> var_413_end_0 = const()[name = string("op_413_end_0"), val = tensor<int32, [4]>([1, 4, 128, 255])];
            tensor<bool, [4]> var_413_end_mask_0 = const()[name = string("op_413_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 128]> var_413_cast_fp16 = slice_by_index(begin = var_413_begin_0, end = var_413_end_0, end_mask = var_413_end_mask_0, x = var_412_cast_fp16)[name = string("op_413_cast_fp16")];
            tensor<fp16, [1, 4, 128, 128]> scores_3_cast_fp16 = add(x = scores_1_cast_fp16, y = var_413_cast_fp16)[name = string("scores_3_cast_fp16")];
            tensor<int32, [1]> var_415_axes_0 = const()[name = string("op_415_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 1, 1, 128]> var_415_cast_fp16 = expand_dims(axes = var_415_axes_0, x = text_mask_to_fp16)[name = string("op_415_cast_fp16")];
            fp16 var_11_to_fp16 = const()[name = string("op_11_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 1, 1, 128]> var_416_cast_fp16 = sub(x = var_11_to_fp16, y = var_415_cast_fp16)[name = string("op_416_cast_fp16")];
            fp16 var_417_to_fp16 = const()[name = string("op_417_to_fp16"), val = fp16(0x1.388p+13)];
            tensor<fp16, [1, 1, 1, 128]> var_418_cast_fp16 = mul(x = var_416_cast_fp16, y = var_417_to_fp16)[name = string("op_418_cast_fp16")];
            tensor<fp16, [1, 4, 128, 128]> input_67_cast_fp16 = sub(x = scores_3_cast_fp16, y = var_418_cast_fp16)[name = string("input_67_cast_fp16")];
            tensor<fp16, [1, 4, 128, 128]> p_attn_1_cast_fp16 = softmax(axis = var_29, x = input_67_cast_fp16)[name = string("p_attn_1_cast_fp16")];
            bool out_13_transpose_x_1 = const()[name = string("out_13_transpose_x_1"), val = bool(false)];
            bool out_13_transpose_y_1 = const()[name = string("out_13_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 4, 128, 64]> out_13_cast_fp16 = matmul(transpose_x = out_13_transpose_x_1, transpose_y = out_13_transpose_y_1, x = p_attn_1_cast_fp16, y = var_351_cast_fp16)[name = string("out_13_cast_fp16")];
            tensor<int32, [8]> x_39_pad_0 = const()[name = string("x_39_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 127])];
            string x_39_mode_0 = const()[name = string("x_39_mode_0"), val = string("constant")];
            fp16 const_18_to_fp16 = const()[name = string("const_18_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 128, 255]> x_39_cast_fp16 = pad(constant_val = const_18_to_fp16, mode = x_39_mode_0, pad = x_39_pad_0, x = p_attn_1_cast_fp16)[name = string("x_39_cast_fp16")];
            tensor<int32, [3]> var_455 = const()[name = string("op_455"), val = tensor<int32, [3]>([1, 4, 32640])];
            tensor<fp16, [1, 4, 32640]> input_71_cast_fp16 = reshape(shape = var_455, x = x_39_cast_fp16)[name = string("input_71_cast_fp16")];
            tensor<int32, [6]> x_41_pad_0 = const()[name = string("x_41_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 128, 0])];
            string x_41_mode_0 = const()[name = string("x_41_mode_0"), val = string("constant")];
            fp16 const_19_to_fp16 = const()[name = string("const_19_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 32768]> x_41_cast_fp16 = pad(constant_val = const_19_to_fp16, mode = x_41_mode_0, pad = x_41_pad_0, x = input_71_cast_fp16)[name = string("x_41_cast_fp16")];
            tensor<int32, [4]> var_462 = const()[name = string("op_462"), val = tensor<int32, [4]>([1, 4, 128, 256])];
            tensor<fp16, [1, 4, 128, 256]> x_43_cast_fp16 = reshape(shape = var_462, x = x_41_cast_fp16)[name = string("x_43_cast_fp16")];
            tensor<int32, [4]> rel_weights_1_begin_0 = const()[name = string("rel_weights_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1])];
            tensor<int32, [4]> rel_weights_1_end_0 = const()[name = string("rel_weights_1_end_0"), val = tensor<int32, [4]>([1, 4, 128, 256])];
            tensor<bool, [4]> rel_weights_1_end_mask_0 = const()[name = string("rel_weights_1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 255]> rel_weights_1_cast_fp16 = slice_by_index(begin = rel_weights_1_begin_0, end = rel_weights_1_end_0, end_mask = rel_weights_1_end_mask_0, x = x_43_cast_fp16)[name = string("rel_weights_1_cast_fp16")];
            bool var_469_transpose_x_0 = const()[name = string("op_469_transpose_x_0"), val = bool(false)];
            bool var_469_transpose_y_0 = const()[name = string("op_469_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 1, 255, 64]> var_468_to_fp16 = const()[name = string("op_468_to_fp16"), val = tensor<fp16, [1, 1, 255, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11023296)))];
            tensor<fp16, [1, 4, 128, 64]> var_469_cast_fp16 = matmul(transpose_x = var_469_transpose_x_0, transpose_y = var_469_transpose_y_0, x = rel_weights_1_cast_fp16, y = var_468_to_fp16)[name = string("op_469_cast_fp16")];
            tensor<fp16, [1, 4, 128, 64]> out_15_cast_fp16 = add(x = out_13_cast_fp16, y = var_469_cast_fp16)[name = string("out_15_cast_fp16")];
            tensor<int32, [4]> var_471_perm_0 = const()[name = string("op_471_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [3]> var_472 = const()[name = string("op_472"), val = tensor<int32, [3]>([1, 256, 128])];
            tensor<fp16, [1, 4, 64, 128]> var_471_cast_fp16 = transpose(perm = var_471_perm_0, x = out_15_cast_fp16)[name = string("transpose_31")];
            tensor<fp16, [1, 256, 128]> input_73_cast_fp16 = reshape(shape = var_472, x = var_471_cast_fp16)[name = string("input_73_cast_fp16")];
            string y_1_pad_type_0 = const()[name = string("y_1_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> y_1_strides_0 = const()[name = string("y_1_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> y_1_pad_0 = const()[name = string("y_1_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> y_1_dilations_0 = const()[name = string("y_1_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 y_1_groups_0 = const()[name = string("y_1_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1]> m_attn_layers_0_attn_conv_o_weight_to_fp16 = const()[name = string("m_attn_layers_0_attn_conv_o_weight_to_fp16"), val = tensor<fp16, [256, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11056000)))];
            tensor<fp16, [256]> m_attn_layers_0_attn_conv_o_bias_to_fp16 = const()[name = string("m_attn_layers_0_attn_conv_o_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11187136)))];
            tensor<fp16, [1, 256, 128]> y_1_cast_fp16 = conv(bias = m_attn_layers_0_attn_conv_o_bias_to_fp16, dilations = y_1_dilations_0, groups = y_1_groups_0, pad = y_1_pad_0, pad_type = y_1_pad_type_0, strides = y_1_strides_0, weight = m_attn_layers_0_attn_conv_o_weight_to_fp16, x = input_73_cast_fp16)[name = string("y_1_cast_fp16")];
            tensor<fp16, [1, 256, 128]> x_45_cast_fp16 = add(x = x_29_cast_fp16, y = y_1_cast_fp16)[name = string("x_45_cast_fp16")];
            tensor<int32, [3]> input_75_perm_0 = const()[name = string("input_75_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_487_axes_0 = const()[name = string("op_487_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [256]> m_attn_layers_0_norm_1_norm_weight_to_fp16 = const()[name = string("m_attn_layers_0_norm_1_norm_weight_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11187712)))];
            tensor<fp16, [256]> m_attn_layers_0_norm_1_norm_bias_to_fp16 = const()[name = string("m_attn_layers_0_norm_1_norm_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11188288)))];
            tensor<fp16, [1, 128, 256]> input_75_cast_fp16 = transpose(perm = input_75_perm_0, x = x_45_cast_fp16)[name = string("transpose_30")];
            tensor<fp16, [1, 128, 256]> var_487_cast_fp16 = layer_norm(axes = var_487_axes_0, beta = m_attn_layers_0_norm_1_norm_bias_to_fp16, epsilon = var_25_to_fp16, gamma = m_attn_layers_0_norm_1_norm_weight_to_fp16, x = input_75_cast_fp16)[name = string("op_487_cast_fp16")];
            tensor<int32, [3]> x_47_perm_0 = const()[name = string("x_47_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 256, 128]> x_47_cast_fp16 = transpose(perm = x_47_perm_0, x = var_487_cast_fp16)[name = string("transpose_29")];
            tensor<fp16, [1, 256, 128]> input_77_cast_fp16 = mul(x = x_47_cast_fp16, y = text_mask_to_fp16)[name = string("input_77_cast_fp16")];
            string input_79_pad_type_0 = const()[name = string("input_79_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_79_strides_0 = const()[name = string("input_79_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_79_pad_0 = const()[name = string("input_79_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_79_dilations_0 = const()[name = string("input_79_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_79_groups_0 = const()[name = string("input_79_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 256, 1]> m_attn_layers_0_ffn_conv_1_weight_to_fp16 = const()[name = string("m_attn_layers_0_ffn_conv_1_weight_to_fp16"), val = tensor<fp16, [1024, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11188864)))];
            tensor<fp16, [1024]> m_attn_layers_0_ffn_conv_1_bias_to_fp16 = const()[name = string("m_attn_layers_0_ffn_conv_1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11713216)))];
            tensor<fp16, [1, 1024, 128]> input_79_cast_fp16 = conv(bias = m_attn_layers_0_ffn_conv_1_bias_to_fp16, dilations = input_79_dilations_0, groups = input_79_groups_0, pad = input_79_pad_0, pad_type = input_79_pad_type_0, strides = input_79_strides_0, weight = m_attn_layers_0_ffn_conv_1_weight_to_fp16, x = input_77_cast_fp16)[name = string("input_79_cast_fp16")];
            tensor<fp16, [1, 1024, 128]> h_37_cast_fp16 = relu(x = input_79_cast_fp16)[name = string("h_37_cast_fp16")];
            tensor<fp16, [1, 1024, 128]> input_81_cast_fp16 = mul(x = h_37_cast_fp16, y = text_mask_to_fp16)[name = string("input_81_cast_fp16")];
            string h_39_pad_type_0 = const()[name = string("h_39_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_39_strides_0 = const()[name = string("h_39_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_39_pad_0 = const()[name = string("h_39_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_39_dilations_0 = const()[name = string("h_39_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_39_groups_0 = const()[name = string("h_39_groups_0"), val = int32(1)];
            tensor<fp16, [256, 1024, 1]> m_attn_layers_0_ffn_conv_2_weight_to_fp16 = const()[name = string("m_attn_layers_0_ffn_conv_2_weight_to_fp16"), val = tensor<fp16, [256, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11715328)))];
            tensor<fp16, [256]> m_attn_layers_0_ffn_conv_2_bias_to_fp16 = const()[name = string("m_attn_layers_0_ffn_conv_2_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12239680)))];
            tensor<fp16, [1, 256, 128]> h_39_cast_fp16 = conv(bias = m_attn_layers_0_ffn_conv_2_bias_to_fp16, dilations = h_39_dilations_0, groups = h_39_groups_0, pad = h_39_pad_0, pad_type = h_39_pad_type_0, strides = h_39_strides_0, weight = m_attn_layers_0_ffn_conv_2_weight_to_fp16, x = input_81_cast_fp16)[name = string("h_39_cast_fp16")];
            tensor<fp16, [1, 256, 128]> y_3_cast_fp16 = mul(x = h_39_cast_fp16, y = text_mask_to_fp16)[name = string("y_3_cast_fp16")];
            tensor<fp16, [1, 256, 128]> x_49_cast_fp16 = add(x = x_47_cast_fp16, y = y_3_cast_fp16)[name = string("x_49_cast_fp16")];
            tensor<int32, [3]> input_83_perm_0 = const()[name = string("input_83_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_515_axes_0 = const()[name = string("op_515_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [256]> m_attn_layers_0_norm_2_norm_weight_to_fp16 = const()[name = string("m_attn_layers_0_norm_2_norm_weight_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12240256)))];
            tensor<fp16, [256]> m_attn_layers_0_norm_2_norm_bias_to_fp16 = const()[name = string("m_attn_layers_0_norm_2_norm_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12240832)))];
            tensor<fp16, [1, 128, 256]> input_83_cast_fp16 = transpose(perm = input_83_perm_0, x = x_49_cast_fp16)[name = string("transpose_28")];
            tensor<fp16, [1, 128, 256]> var_515_cast_fp16 = layer_norm(axes = var_515_axes_0, beta = m_attn_layers_0_norm_2_norm_bias_to_fp16, epsilon = var_25_to_fp16, gamma = m_attn_layers_0_norm_2_norm_weight_to_fp16, x = input_83_cast_fp16)[name = string("op_515_cast_fp16")];
            tensor<int32, [3]> x_51_perm_0 = const()[name = string("x_51_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 256, 128]> x_51_cast_fp16 = transpose(perm = x_51_perm_0, x = var_515_cast_fp16)[name = string("transpose_27")];
            tensor<fp16, [1, 256, 128]> x_53_cast_fp16 = mul(x = x_51_cast_fp16, y = text_mask_to_fp16)[name = string("x_53_cast_fp16")];
            string var_537_pad_type_0 = const()[name = string("op_537_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_537_strides_0 = const()[name = string("op_537_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_537_pad_0 = const()[name = string("op_537_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_537_dilations_0 = const()[name = string("op_537_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_537_groups_0 = const()[name = string("op_537_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1]> m_attn_layers_1_attn_conv_q_weight_to_fp16 = const()[name = string("m_attn_layers_1_attn_conv_q_weight_to_fp16"), val = tensor<fp16, [256, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12241408)))];
            tensor<fp16, [256]> m_attn_layers_1_attn_conv_q_bias_to_fp16 = const()[name = string("m_attn_layers_1_attn_conv_q_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12372544)))];
            tensor<fp16, [1, 256, 128]> var_537_cast_fp16 = conv(bias = m_attn_layers_1_attn_conv_q_bias_to_fp16, dilations = var_537_dilations_0, groups = var_537_groups_0, pad = var_537_pad_0, pad_type = var_537_pad_type_0, strides = var_537_strides_0, weight = m_attn_layers_1_attn_conv_q_weight_to_fp16, x = x_53_cast_fp16)[name = string("op_537_cast_fp16")];
            tensor<int32, [4]> var_538 = const()[name = string("op_538"), val = tensor<int32, [4]>([1, 4, 64, 128])];
            tensor<fp16, [1, 4, 64, 128]> var_539_cast_fp16 = reshape(shape = var_538, x = var_537_cast_fp16)[name = string("op_539_cast_fp16")];
            tensor<int32, [4]> q_3_perm_0 = const()[name = string("q_3_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            string var_547_pad_type_0 = const()[name = string("op_547_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_547_strides_0 = const()[name = string("op_547_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_547_pad_0 = const()[name = string("op_547_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_547_dilations_0 = const()[name = string("op_547_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_547_groups_0 = const()[name = string("op_547_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1]> m_attn_layers_1_attn_conv_k_weight_to_fp16 = const()[name = string("m_attn_layers_1_attn_conv_k_weight_to_fp16"), val = tensor<fp16, [256, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12373120)))];
            tensor<fp16, [256]> m_attn_layers_1_attn_conv_k_bias_to_fp16 = const()[name = string("m_attn_layers_1_attn_conv_k_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12504256)))];
            tensor<fp16, [1, 256, 128]> var_547_cast_fp16 = conv(bias = m_attn_layers_1_attn_conv_k_bias_to_fp16, dilations = var_547_dilations_0, groups = var_547_groups_0, pad = var_547_pad_0, pad_type = var_547_pad_type_0, strides = var_547_strides_0, weight = m_attn_layers_1_attn_conv_k_weight_to_fp16, x = x_53_cast_fp16)[name = string("op_547_cast_fp16")];
            tensor<int32, [4]> var_548 = const()[name = string("op_548"), val = tensor<int32, [4]>([1, 4, 64, 128])];
            tensor<fp16, [1, 4, 64, 128]> var_549_cast_fp16 = reshape(shape = var_548, x = var_547_cast_fp16)[name = string("op_549_cast_fp16")];
            string var_557_pad_type_0 = const()[name = string("op_557_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_557_strides_0 = const()[name = string("op_557_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_557_pad_0 = const()[name = string("op_557_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_557_dilations_0 = const()[name = string("op_557_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_557_groups_0 = const()[name = string("op_557_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1]> m_attn_layers_1_attn_conv_v_weight_to_fp16 = const()[name = string("m_attn_layers_1_attn_conv_v_weight_to_fp16"), val = tensor<fp16, [256, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12504832)))];
            tensor<fp16, [256]> m_attn_layers_1_attn_conv_v_bias_to_fp16 = const()[name = string("m_attn_layers_1_attn_conv_v_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12635968)))];
            tensor<fp16, [1, 256, 128]> var_557_cast_fp16 = conv(bias = m_attn_layers_1_attn_conv_v_bias_to_fp16, dilations = var_557_dilations_0, groups = var_557_groups_0, pad = var_557_pad_0, pad_type = var_557_pad_type_0, strides = var_557_strides_0, weight = m_attn_layers_1_attn_conv_v_weight_to_fp16, x = x_53_cast_fp16)[name = string("op_557_cast_fp16")];
            tensor<int32, [4]> var_558 = const()[name = string("op_558"), val = tensor<int32, [4]>([1, 4, 64, 128])];
            tensor<fp16, [1, 4, 64, 128]> var_559_cast_fp16 = reshape(shape = var_558, x = var_557_cast_fp16)[name = string("op_559_cast_fp16")];
            fp16 var_561_to_fp16 = const()[name = string("op_561_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 4, 128, 64]> q_3_cast_fp16 = transpose(perm = q_3_perm_0, x = var_539_cast_fp16)[name = string("transpose_26")];
            tensor<fp16, [1, 4, 128, 64]> var_562_cast_fp16 = mul(x = q_3_cast_fp16, y = var_561_to_fp16)[name = string("op_562_cast_fp16")];
            bool scores_5_transpose_x_0 = const()[name = string("scores_5_transpose_x_0"), val = bool(false)];
            bool scores_5_transpose_y_0 = const()[name = string("scores_5_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 4, 128, 128]> scores_5_cast_fp16 = matmul(transpose_x = scores_5_transpose_x_0, transpose_y = scores_5_transpose_y_0, x = var_562_cast_fp16, y = var_549_cast_fp16)[name = string("scores_5_cast_fp16")];
            bool x_55_transpose_x_0 = const()[name = string("x_55_transpose_x_0"), val = bool(false)];
            bool x_55_transpose_y_0 = const()[name = string("x_55_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 1, 64, 255]> var_586_to_fp16 = const()[name = string("op_586_to_fp16"), val = tensor<fp16, [1, 1, 64, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12636544)))];
            tensor<fp16, [1, 4, 128, 255]> x_55_cast_fp16 = matmul(transpose_x = x_55_transpose_x_0, transpose_y = x_55_transpose_y_0, x = var_562_cast_fp16, y = var_586_to_fp16)[name = string("x_55_cast_fp16")];
            tensor<int32, [8]> x_57_pad_0 = const()[name = string("x_57_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 1])];
            string x_57_mode_0 = const()[name = string("x_57_mode_0"), val = string("constant")];
            fp16 const_26_to_fp16 = const()[name = string("const_26_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 128, 256]> x_57_cast_fp16 = pad(constant_val = const_26_to_fp16, mode = x_57_mode_0, pad = x_57_pad_0, x = x_55_cast_fp16)[name = string("x_57_cast_fp16")];
            tensor<int32, [3]> var_598 = const()[name = string("op_598"), val = tensor<int32, [3]>([1, 4, 32768])];
            tensor<fp16, [1, 4, 32768]> input_87_cast_fp16 = reshape(shape = var_598, x = x_57_cast_fp16)[name = string("input_87_cast_fp16")];
            tensor<int32, [6]> x_59_pad_0 = const()[name = string("x_59_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 127])];
            string x_59_mode_0 = const()[name = string("x_59_mode_0"), val = string("constant")];
            fp16 const_27_to_fp16 = const()[name = string("const_27_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 32895]> x_59_cast_fp16 = pad(constant_val = const_27_to_fp16, mode = x_59_mode_0, pad = x_59_pad_0, x = input_87_cast_fp16)[name = string("x_59_cast_fp16")];
            tensor<int32, [4]> var_613 = const()[name = string("op_613"), val = tensor<int32, [4]>([1, 4, 129, 255])];
            tensor<fp16, [1, 4, 129, 255]> x_61_cast_fp16 = reshape(shape = var_613, x = x_59_cast_fp16)[name = string("x_61_cast_fp16")];
            tensor<int32, [4]> var_620_begin_0 = const()[name = string("op_620_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_620_end_0 = const()[name = string("op_620_end_0"), val = tensor<int32, [4]>([1, 4, 128, 255])];
            tensor<bool, [4]> var_620_end_mask_0 = const()[name = string("op_620_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])];
            tensor<fp16, [1, 4, 128, 255]> var_620_cast_fp16 = slice_by_index(begin = var_620_begin_0, end = var_620_end_0, end_mask = var_620_end_mask_0, x = x_61_cast_fp16)[name = string("op_620_cast_fp16")];
            tensor<int32, [4]> var_621_begin_0 = const()[name = string("op_621_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 127])];
            tensor<int32, [4]> var_621_end_0 = const()[name = string("op_621_end_0"), val = tensor<int32, [4]>([1, 4, 128, 255])];
            tensor<bool, [4]> var_621_end_mask_0 = const()[name = string("op_621_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 128]> var_621_cast_fp16 = slice_by_index(begin = var_621_begin_0, end = var_621_end_0, end_mask = var_621_end_mask_0, x = var_620_cast_fp16)[name = string("op_621_cast_fp16")];
            tensor<fp16, [1, 4, 128, 128]> scores_7_cast_fp16 = add(x = scores_5_cast_fp16, y = var_621_cast_fp16)[name = string("scores_7_cast_fp16")];
            tensor<fp16, [1, 4, 128, 128]> input_89_cast_fp16 = sub(x = scores_7_cast_fp16, y = var_418_cast_fp16)[name = string("input_89_cast_fp16")];
            tensor<fp16, [1, 4, 128, 128]> p_attn_3_cast_fp16 = softmax(axis = var_29, x = input_89_cast_fp16)[name = string("p_attn_3_cast_fp16")];
            bool out_17_transpose_x_1 = const()[name = string("out_17_transpose_x_1"), val = bool(false)];
            bool out_17_transpose_y_1 = const()[name = string("out_17_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 4, 128, 64]> out_17_cast_fp16 = matmul(transpose_x = out_17_transpose_x_1, transpose_y = out_17_transpose_y_1, x = p_attn_3_cast_fp16, y = var_559_cast_fp16)[name = string("out_17_cast_fp16")];
            tensor<int32, [8]> x_63_pad_0 = const()[name = string("x_63_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 127])];
            string x_63_mode_0 = const()[name = string("x_63_mode_0"), val = string("constant")];
            fp16 const_32_to_fp16 = const()[name = string("const_32_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 128, 255]> x_63_cast_fp16 = pad(constant_val = const_32_to_fp16, mode = x_63_mode_0, pad = x_63_pad_0, x = p_attn_3_cast_fp16)[name = string("x_63_cast_fp16")];
            tensor<int32, [3]> var_663 = const()[name = string("op_663"), val = tensor<int32, [3]>([1, 4, 32640])];
            tensor<fp16, [1, 4, 32640]> input_93_cast_fp16 = reshape(shape = var_663, x = x_63_cast_fp16)[name = string("input_93_cast_fp16")];
            tensor<int32, [6]> x_65_pad_0 = const()[name = string("x_65_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 128, 0])];
            string x_65_mode_0 = const()[name = string("x_65_mode_0"), val = string("constant")];
            fp16 const_33_to_fp16 = const()[name = string("const_33_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 32768]> x_65_cast_fp16 = pad(constant_val = const_33_to_fp16, mode = x_65_mode_0, pad = x_65_pad_0, x = input_93_cast_fp16)[name = string("x_65_cast_fp16")];
            tensor<int32, [4]> var_670 = const()[name = string("op_670"), val = tensor<int32, [4]>([1, 4, 128, 256])];
            tensor<fp16, [1, 4, 128, 256]> x_67_cast_fp16 = reshape(shape = var_670, x = x_65_cast_fp16)[name = string("x_67_cast_fp16")];
            tensor<int32, [4]> rel_weights_3_begin_0 = const()[name = string("rel_weights_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1])];
            tensor<int32, [4]> rel_weights_3_end_0 = const()[name = string("rel_weights_3_end_0"), val = tensor<int32, [4]>([1, 4, 128, 256])];
            tensor<bool, [4]> rel_weights_3_end_mask_0 = const()[name = string("rel_weights_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 255]> rel_weights_3_cast_fp16 = slice_by_index(begin = rel_weights_3_begin_0, end = rel_weights_3_end_0, end_mask = rel_weights_3_end_mask_0, x = x_67_cast_fp16)[name = string("rel_weights_3_cast_fp16")];
            bool var_677_transpose_x_0 = const()[name = string("op_677_transpose_x_0"), val = bool(false)];
            bool var_677_transpose_y_0 = const()[name = string("op_677_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 1, 255, 64]> var_676_to_fp16 = const()[name = string("op_676_to_fp16"), val = tensor<fp16, [1, 1, 255, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12669248)))];
            tensor<fp16, [1, 4, 128, 64]> var_677_cast_fp16 = matmul(transpose_x = var_677_transpose_x_0, transpose_y = var_677_transpose_y_0, x = rel_weights_3_cast_fp16, y = var_676_to_fp16)[name = string("op_677_cast_fp16")];
            tensor<fp16, [1, 4, 128, 64]> out_19_cast_fp16 = add(x = out_17_cast_fp16, y = var_677_cast_fp16)[name = string("out_19_cast_fp16")];
            tensor<int32, [4]> var_679_perm_0 = const()[name = string("op_679_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [3]> var_680 = const()[name = string("op_680"), val = tensor<int32, [3]>([1, 256, 128])];
            tensor<fp16, [1, 4, 64, 128]> var_679_cast_fp16 = transpose(perm = var_679_perm_0, x = out_19_cast_fp16)[name = string("transpose_25")];
            tensor<fp16, [1, 256, 128]> input_95_cast_fp16 = reshape(shape = var_680, x = var_679_cast_fp16)[name = string("input_95_cast_fp16")];
            string y_5_pad_type_0 = const()[name = string("y_5_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> y_5_strides_0 = const()[name = string("y_5_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> y_5_pad_0 = const()[name = string("y_5_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> y_5_dilations_0 = const()[name = string("y_5_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 y_5_groups_0 = const()[name = string("y_5_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1]> m_attn_layers_1_attn_conv_o_weight_to_fp16 = const()[name = string("m_attn_layers_1_attn_conv_o_weight_to_fp16"), val = tensor<fp16, [256, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12701952)))];
            tensor<fp16, [256]> m_attn_layers_1_attn_conv_o_bias_to_fp16 = const()[name = string("m_attn_layers_1_attn_conv_o_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12833088)))];
            tensor<fp16, [1, 256, 128]> y_5_cast_fp16 = conv(bias = m_attn_layers_1_attn_conv_o_bias_to_fp16, dilations = y_5_dilations_0, groups = y_5_groups_0, pad = y_5_pad_0, pad_type = y_5_pad_type_0, strides = y_5_strides_0, weight = m_attn_layers_1_attn_conv_o_weight_to_fp16, x = input_95_cast_fp16)[name = string("y_5_cast_fp16")];
            tensor<fp16, [1, 256, 128]> x_69_cast_fp16 = add(x = x_53_cast_fp16, y = y_5_cast_fp16)[name = string("x_69_cast_fp16")];
            tensor<int32, [3]> input_97_perm_0 = const()[name = string("input_97_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_695_axes_0 = const()[name = string("op_695_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [256]> m_attn_layers_1_norm_1_norm_weight_to_fp16 = const()[name = string("m_attn_layers_1_norm_1_norm_weight_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12833664)))];
            tensor<fp16, [256]> m_attn_layers_1_norm_1_norm_bias_to_fp16 = const()[name = string("m_attn_layers_1_norm_1_norm_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12834240)))];
            tensor<fp16, [1, 128, 256]> input_97_cast_fp16 = transpose(perm = input_97_perm_0, x = x_69_cast_fp16)[name = string("transpose_24")];
            tensor<fp16, [1, 128, 256]> var_695_cast_fp16 = layer_norm(axes = var_695_axes_0, beta = m_attn_layers_1_norm_1_norm_bias_to_fp16, epsilon = var_25_to_fp16, gamma = m_attn_layers_1_norm_1_norm_weight_to_fp16, x = input_97_cast_fp16)[name = string("op_695_cast_fp16")];
            tensor<int32, [3]> x_71_perm_0 = const()[name = string("x_71_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 256, 128]> x_71_cast_fp16 = transpose(perm = x_71_perm_0, x = var_695_cast_fp16)[name = string("transpose_23")];
            tensor<fp16, [1, 256, 128]> input_99_cast_fp16 = mul(x = x_71_cast_fp16, y = text_mask_to_fp16)[name = string("input_99_cast_fp16")];
            string input_101_pad_type_0 = const()[name = string("input_101_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_101_strides_0 = const()[name = string("input_101_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_101_pad_0 = const()[name = string("input_101_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_101_dilations_0 = const()[name = string("input_101_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_101_groups_0 = const()[name = string("input_101_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 256, 1]> m_attn_layers_1_ffn_conv_1_weight_to_fp16 = const()[name = string("m_attn_layers_1_ffn_conv_1_weight_to_fp16"), val = tensor<fp16, [1024, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12834816)))];
            tensor<fp16, [1024]> m_attn_layers_1_ffn_conv_1_bias_to_fp16 = const()[name = string("m_attn_layers_1_ffn_conv_1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13359168)))];
            tensor<fp16, [1, 1024, 128]> input_101_cast_fp16 = conv(bias = m_attn_layers_1_ffn_conv_1_bias_to_fp16, dilations = input_101_dilations_0, groups = input_101_groups_0, pad = input_101_pad_0, pad_type = input_101_pad_type_0, strides = input_101_strides_0, weight = m_attn_layers_1_ffn_conv_1_weight_to_fp16, x = input_99_cast_fp16)[name = string("input_101_cast_fp16")];
            tensor<fp16, [1, 1024, 128]> h_41_cast_fp16 = relu(x = input_101_cast_fp16)[name = string("h_41_cast_fp16")];
            tensor<fp16, [1, 1024, 128]> input_103_cast_fp16 = mul(x = h_41_cast_fp16, y = text_mask_to_fp16)[name = string("input_103_cast_fp16")];
            string h_43_pad_type_0 = const()[name = string("h_43_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_43_strides_0 = const()[name = string("h_43_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_43_pad_0 = const()[name = string("h_43_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_43_dilations_0 = const()[name = string("h_43_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_43_groups_0 = const()[name = string("h_43_groups_0"), val = int32(1)];
            tensor<fp16, [256, 1024, 1]> m_attn_layers_1_ffn_conv_2_weight_to_fp16 = const()[name = string("m_attn_layers_1_ffn_conv_2_weight_to_fp16"), val = tensor<fp16, [256, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13361280)))];
            tensor<fp16, [256]> m_attn_layers_1_ffn_conv_2_bias_to_fp16 = const()[name = string("m_attn_layers_1_ffn_conv_2_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13885632)))];
            tensor<fp16, [1, 256, 128]> h_43_cast_fp16 = conv(bias = m_attn_layers_1_ffn_conv_2_bias_to_fp16, dilations = h_43_dilations_0, groups = h_43_groups_0, pad = h_43_pad_0, pad_type = h_43_pad_type_0, strides = h_43_strides_0, weight = m_attn_layers_1_ffn_conv_2_weight_to_fp16, x = input_103_cast_fp16)[name = string("h_43_cast_fp16")];
            tensor<fp16, [1, 256, 128]> y_7_cast_fp16 = mul(x = h_43_cast_fp16, y = text_mask_to_fp16)[name = string("y_7_cast_fp16")];
            tensor<fp16, [1, 256, 128]> x_73_cast_fp16 = add(x = x_71_cast_fp16, y = y_7_cast_fp16)[name = string("x_73_cast_fp16")];
            tensor<int32, [3]> input_105_perm_0 = const()[name = string("input_105_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_723_axes_0 = const()[name = string("op_723_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [256]> m_attn_layers_1_norm_2_norm_weight_to_fp16 = const()[name = string("m_attn_layers_1_norm_2_norm_weight_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13886208)))];
            tensor<fp16, [256]> m_attn_layers_1_norm_2_norm_bias_to_fp16 = const()[name = string("m_attn_layers_1_norm_2_norm_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13886784)))];
            tensor<fp16, [1, 128, 256]> input_105_cast_fp16 = transpose(perm = input_105_perm_0, x = x_73_cast_fp16)[name = string("transpose_22")];
            tensor<fp16, [1, 128, 256]> var_723_cast_fp16 = layer_norm(axes = var_723_axes_0, beta = m_attn_layers_1_norm_2_norm_bias_to_fp16, epsilon = var_25_to_fp16, gamma = m_attn_layers_1_norm_2_norm_weight_to_fp16, x = input_105_cast_fp16)[name = string("op_723_cast_fp16")];
            tensor<int32, [3]> x_75_perm_0 = const()[name = string("x_75_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 256, 128]> x_75_cast_fp16 = transpose(perm = x_75_perm_0, x = var_723_cast_fp16)[name = string("transpose_21")];
            tensor<fp16, [1, 256, 128]> x_77_cast_fp16 = mul(x = x_75_cast_fp16, y = text_mask_to_fp16)[name = string("x_77_cast_fp16")];
            string var_745_pad_type_0 = const()[name = string("op_745_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_745_strides_0 = const()[name = string("op_745_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_745_pad_0 = const()[name = string("op_745_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_745_dilations_0 = const()[name = string("op_745_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_745_groups_0 = const()[name = string("op_745_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1]> m_attn_layers_2_attn_conv_q_weight_to_fp16 = const()[name = string("m_attn_layers_2_attn_conv_q_weight_to_fp16"), val = tensor<fp16, [256, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13887360)))];
            tensor<fp16, [256]> m_attn_layers_2_attn_conv_q_bias_to_fp16 = const()[name = string("m_attn_layers_2_attn_conv_q_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14018496)))];
            tensor<fp16, [1, 256, 128]> var_745_cast_fp16 = conv(bias = m_attn_layers_2_attn_conv_q_bias_to_fp16, dilations = var_745_dilations_0, groups = var_745_groups_0, pad = var_745_pad_0, pad_type = var_745_pad_type_0, strides = var_745_strides_0, weight = m_attn_layers_2_attn_conv_q_weight_to_fp16, x = x_77_cast_fp16)[name = string("op_745_cast_fp16")];
            tensor<int32, [4]> var_746 = const()[name = string("op_746"), val = tensor<int32, [4]>([1, 4, 64, 128])];
            tensor<fp16, [1, 4, 64, 128]> var_747_cast_fp16 = reshape(shape = var_746, x = var_745_cast_fp16)[name = string("op_747_cast_fp16")];
            tensor<int32, [4]> q_5_perm_0 = const()[name = string("q_5_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            string var_755_pad_type_0 = const()[name = string("op_755_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_755_strides_0 = const()[name = string("op_755_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_755_pad_0 = const()[name = string("op_755_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_755_dilations_0 = const()[name = string("op_755_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_755_groups_0 = const()[name = string("op_755_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1]> m_attn_layers_2_attn_conv_k_weight_to_fp16 = const()[name = string("m_attn_layers_2_attn_conv_k_weight_to_fp16"), val = tensor<fp16, [256, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14019072)))];
            tensor<fp16, [256]> m_attn_layers_2_attn_conv_k_bias_to_fp16 = const()[name = string("m_attn_layers_2_attn_conv_k_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14150208)))];
            tensor<fp16, [1, 256, 128]> var_755_cast_fp16 = conv(bias = m_attn_layers_2_attn_conv_k_bias_to_fp16, dilations = var_755_dilations_0, groups = var_755_groups_0, pad = var_755_pad_0, pad_type = var_755_pad_type_0, strides = var_755_strides_0, weight = m_attn_layers_2_attn_conv_k_weight_to_fp16, x = x_77_cast_fp16)[name = string("op_755_cast_fp16")];
            tensor<int32, [4]> var_756 = const()[name = string("op_756"), val = tensor<int32, [4]>([1, 4, 64, 128])];
            tensor<fp16, [1, 4, 64, 128]> var_757_cast_fp16 = reshape(shape = var_756, x = var_755_cast_fp16)[name = string("op_757_cast_fp16")];
            string var_765_pad_type_0 = const()[name = string("op_765_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_765_strides_0 = const()[name = string("op_765_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_765_pad_0 = const()[name = string("op_765_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_765_dilations_0 = const()[name = string("op_765_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_765_groups_0 = const()[name = string("op_765_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1]> m_attn_layers_2_attn_conv_v_weight_to_fp16 = const()[name = string("m_attn_layers_2_attn_conv_v_weight_to_fp16"), val = tensor<fp16, [256, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14150784)))];
            tensor<fp16, [256]> m_attn_layers_2_attn_conv_v_bias_to_fp16 = const()[name = string("m_attn_layers_2_attn_conv_v_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14281920)))];
            tensor<fp16, [1, 256, 128]> var_765_cast_fp16 = conv(bias = m_attn_layers_2_attn_conv_v_bias_to_fp16, dilations = var_765_dilations_0, groups = var_765_groups_0, pad = var_765_pad_0, pad_type = var_765_pad_type_0, strides = var_765_strides_0, weight = m_attn_layers_2_attn_conv_v_weight_to_fp16, x = x_77_cast_fp16)[name = string("op_765_cast_fp16")];
            tensor<int32, [4]> var_766 = const()[name = string("op_766"), val = tensor<int32, [4]>([1, 4, 64, 128])];
            tensor<fp16, [1, 4, 64, 128]> var_767_cast_fp16 = reshape(shape = var_766, x = var_765_cast_fp16)[name = string("op_767_cast_fp16")];
            fp16 var_769_to_fp16 = const()[name = string("op_769_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 4, 128, 64]> q_5_cast_fp16 = transpose(perm = q_5_perm_0, x = var_747_cast_fp16)[name = string("transpose_20")];
            tensor<fp16, [1, 4, 128, 64]> var_770_cast_fp16 = mul(x = q_5_cast_fp16, y = var_769_to_fp16)[name = string("op_770_cast_fp16")];
            bool scores_9_transpose_x_0 = const()[name = string("scores_9_transpose_x_0"), val = bool(false)];
            bool scores_9_transpose_y_0 = const()[name = string("scores_9_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 4, 128, 128]> scores_9_cast_fp16 = matmul(transpose_x = scores_9_transpose_x_0, transpose_y = scores_9_transpose_y_0, x = var_770_cast_fp16, y = var_757_cast_fp16)[name = string("scores_9_cast_fp16")];
            bool x_79_transpose_x_0 = const()[name = string("x_79_transpose_x_0"), val = bool(false)];
            bool x_79_transpose_y_0 = const()[name = string("x_79_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 1, 64, 255]> var_794_to_fp16 = const()[name = string("op_794_to_fp16"), val = tensor<fp16, [1, 1, 64, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14282496)))];
            tensor<fp16, [1, 4, 128, 255]> x_79_cast_fp16 = matmul(transpose_x = x_79_transpose_x_0, transpose_y = x_79_transpose_y_0, x = var_770_cast_fp16, y = var_794_to_fp16)[name = string("x_79_cast_fp16")];
            tensor<int32, [8]> x_81_pad_0 = const()[name = string("x_81_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 1])];
            string x_81_mode_0 = const()[name = string("x_81_mode_0"), val = string("constant")];
            fp16 const_40_to_fp16 = const()[name = string("const_40_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 128, 256]> x_81_cast_fp16 = pad(constant_val = const_40_to_fp16, mode = x_81_mode_0, pad = x_81_pad_0, x = x_79_cast_fp16)[name = string("x_81_cast_fp16")];
            tensor<int32, [3]> var_806 = const()[name = string("op_806"), val = tensor<int32, [3]>([1, 4, 32768])];
            tensor<fp16, [1, 4, 32768]> input_109_cast_fp16 = reshape(shape = var_806, x = x_81_cast_fp16)[name = string("input_109_cast_fp16")];
            tensor<int32, [6]> x_83_pad_0 = const()[name = string("x_83_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 127])];
            string x_83_mode_0 = const()[name = string("x_83_mode_0"), val = string("constant")];
            fp16 const_41_to_fp16 = const()[name = string("const_41_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 32895]> x_83_cast_fp16 = pad(constant_val = const_41_to_fp16, mode = x_83_mode_0, pad = x_83_pad_0, x = input_109_cast_fp16)[name = string("x_83_cast_fp16")];
            tensor<int32, [4]> var_821 = const()[name = string("op_821"), val = tensor<int32, [4]>([1, 4, 129, 255])];
            tensor<fp16, [1, 4, 129, 255]> x_85_cast_fp16 = reshape(shape = var_821, x = x_83_cast_fp16)[name = string("x_85_cast_fp16")];
            tensor<int32, [4]> var_828_begin_0 = const()[name = string("op_828_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_828_end_0 = const()[name = string("op_828_end_0"), val = tensor<int32, [4]>([1, 4, 128, 255])];
            tensor<bool, [4]> var_828_end_mask_0 = const()[name = string("op_828_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])];
            tensor<fp16, [1, 4, 128, 255]> var_828_cast_fp16 = slice_by_index(begin = var_828_begin_0, end = var_828_end_0, end_mask = var_828_end_mask_0, x = x_85_cast_fp16)[name = string("op_828_cast_fp16")];
            tensor<int32, [4]> var_829_begin_0 = const()[name = string("op_829_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 127])];
            tensor<int32, [4]> var_829_end_0 = const()[name = string("op_829_end_0"), val = tensor<int32, [4]>([1, 4, 128, 255])];
            tensor<bool, [4]> var_829_end_mask_0 = const()[name = string("op_829_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 128]> var_829_cast_fp16 = slice_by_index(begin = var_829_begin_0, end = var_829_end_0, end_mask = var_829_end_mask_0, x = var_828_cast_fp16)[name = string("op_829_cast_fp16")];
            tensor<fp16, [1, 4, 128, 128]> scores_11_cast_fp16 = add(x = scores_9_cast_fp16, y = var_829_cast_fp16)[name = string("scores_11_cast_fp16")];
            tensor<fp16, [1, 4, 128, 128]> input_111_cast_fp16 = sub(x = scores_11_cast_fp16, y = var_418_cast_fp16)[name = string("input_111_cast_fp16")];
            tensor<fp16, [1, 4, 128, 128]> p_attn_5_cast_fp16 = softmax(axis = var_29, x = input_111_cast_fp16)[name = string("p_attn_5_cast_fp16")];
            bool out_21_transpose_x_1 = const()[name = string("out_21_transpose_x_1"), val = bool(false)];
            bool out_21_transpose_y_1 = const()[name = string("out_21_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 4, 128, 64]> out_21_cast_fp16 = matmul(transpose_x = out_21_transpose_x_1, transpose_y = out_21_transpose_y_1, x = p_attn_5_cast_fp16, y = var_767_cast_fp16)[name = string("out_21_cast_fp16")];
            tensor<int32, [8]> x_87_pad_0 = const()[name = string("x_87_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 127])];
            string x_87_mode_0 = const()[name = string("x_87_mode_0"), val = string("constant")];
            fp16 const_46_to_fp16 = const()[name = string("const_46_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 128, 255]> x_87_cast_fp16 = pad(constant_val = const_46_to_fp16, mode = x_87_mode_0, pad = x_87_pad_0, x = p_attn_5_cast_fp16)[name = string("x_87_cast_fp16")];
            tensor<int32, [3]> var_871 = const()[name = string("op_871"), val = tensor<int32, [3]>([1, 4, 32640])];
            tensor<fp16, [1, 4, 32640]> input_115_cast_fp16 = reshape(shape = var_871, x = x_87_cast_fp16)[name = string("input_115_cast_fp16")];
            tensor<int32, [6]> x_89_pad_0 = const()[name = string("x_89_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 128, 0])];
            string x_89_mode_0 = const()[name = string("x_89_mode_0"), val = string("constant")];
            fp16 const_47_to_fp16 = const()[name = string("const_47_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 32768]> x_89_cast_fp16 = pad(constant_val = const_47_to_fp16, mode = x_89_mode_0, pad = x_89_pad_0, x = input_115_cast_fp16)[name = string("x_89_cast_fp16")];
            tensor<int32, [4]> var_878 = const()[name = string("op_878"), val = tensor<int32, [4]>([1, 4, 128, 256])];
            tensor<fp16, [1, 4, 128, 256]> x_91_cast_fp16 = reshape(shape = var_878, x = x_89_cast_fp16)[name = string("x_91_cast_fp16")];
            tensor<int32, [4]> rel_weights_5_begin_0 = const()[name = string("rel_weights_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1])];
            tensor<int32, [4]> rel_weights_5_end_0 = const()[name = string("rel_weights_5_end_0"), val = tensor<int32, [4]>([1, 4, 128, 256])];
            tensor<bool, [4]> rel_weights_5_end_mask_0 = const()[name = string("rel_weights_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 255]> rel_weights_5_cast_fp16 = slice_by_index(begin = rel_weights_5_begin_0, end = rel_weights_5_end_0, end_mask = rel_weights_5_end_mask_0, x = x_91_cast_fp16)[name = string("rel_weights_5_cast_fp16")];
            bool var_885_transpose_x_0 = const()[name = string("op_885_transpose_x_0"), val = bool(false)];
            bool var_885_transpose_y_0 = const()[name = string("op_885_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 1, 255, 64]> var_884_to_fp16 = const()[name = string("op_884_to_fp16"), val = tensor<fp16, [1, 1, 255, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14315200)))];
            tensor<fp16, [1, 4, 128, 64]> var_885_cast_fp16 = matmul(transpose_x = var_885_transpose_x_0, transpose_y = var_885_transpose_y_0, x = rel_weights_5_cast_fp16, y = var_884_to_fp16)[name = string("op_885_cast_fp16")];
            tensor<fp16, [1, 4, 128, 64]> out_23_cast_fp16 = add(x = out_21_cast_fp16, y = var_885_cast_fp16)[name = string("out_23_cast_fp16")];
            tensor<int32, [4]> var_887_perm_0 = const()[name = string("op_887_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [3]> var_888 = const()[name = string("op_888"), val = tensor<int32, [3]>([1, 256, 128])];
            tensor<fp16, [1, 4, 64, 128]> var_887_cast_fp16 = transpose(perm = var_887_perm_0, x = out_23_cast_fp16)[name = string("transpose_19")];
            tensor<fp16, [1, 256, 128]> input_117_cast_fp16 = reshape(shape = var_888, x = var_887_cast_fp16)[name = string("input_117_cast_fp16")];
            string y_9_pad_type_0 = const()[name = string("y_9_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> y_9_strides_0 = const()[name = string("y_9_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> y_9_pad_0 = const()[name = string("y_9_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> y_9_dilations_0 = const()[name = string("y_9_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 y_9_groups_0 = const()[name = string("y_9_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1]> m_attn_layers_2_attn_conv_o_weight_to_fp16 = const()[name = string("m_attn_layers_2_attn_conv_o_weight_to_fp16"), val = tensor<fp16, [256, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14347904)))];
            tensor<fp16, [256]> m_attn_layers_2_attn_conv_o_bias_to_fp16 = const()[name = string("m_attn_layers_2_attn_conv_o_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14479040)))];
            tensor<fp16, [1, 256, 128]> y_9_cast_fp16 = conv(bias = m_attn_layers_2_attn_conv_o_bias_to_fp16, dilations = y_9_dilations_0, groups = y_9_groups_0, pad = y_9_pad_0, pad_type = y_9_pad_type_0, strides = y_9_strides_0, weight = m_attn_layers_2_attn_conv_o_weight_to_fp16, x = input_117_cast_fp16)[name = string("y_9_cast_fp16")];
            tensor<fp16, [1, 256, 128]> x_93_cast_fp16 = add(x = x_77_cast_fp16, y = y_9_cast_fp16)[name = string("x_93_cast_fp16")];
            tensor<int32, [3]> input_119_perm_0 = const()[name = string("input_119_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_903_axes_0 = const()[name = string("op_903_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [256]> m_attn_layers_2_norm_1_norm_weight_to_fp16 = const()[name = string("m_attn_layers_2_norm_1_norm_weight_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14479616)))];
            tensor<fp16, [256]> m_attn_layers_2_norm_1_norm_bias_to_fp16 = const()[name = string("m_attn_layers_2_norm_1_norm_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14480192)))];
            tensor<fp16, [1, 128, 256]> input_119_cast_fp16 = transpose(perm = input_119_perm_0, x = x_93_cast_fp16)[name = string("transpose_18")];
            tensor<fp16, [1, 128, 256]> var_903_cast_fp16 = layer_norm(axes = var_903_axes_0, beta = m_attn_layers_2_norm_1_norm_bias_to_fp16, epsilon = var_25_to_fp16, gamma = m_attn_layers_2_norm_1_norm_weight_to_fp16, x = input_119_cast_fp16)[name = string("op_903_cast_fp16")];
            tensor<int32, [3]> x_95_perm_0 = const()[name = string("x_95_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 256, 128]> x_95_cast_fp16 = transpose(perm = x_95_perm_0, x = var_903_cast_fp16)[name = string("transpose_17")];
            tensor<fp16, [1, 256, 128]> input_121_cast_fp16 = mul(x = x_95_cast_fp16, y = text_mask_to_fp16)[name = string("input_121_cast_fp16")];
            string input_123_pad_type_0 = const()[name = string("input_123_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_123_strides_0 = const()[name = string("input_123_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_123_pad_0 = const()[name = string("input_123_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_123_dilations_0 = const()[name = string("input_123_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_123_groups_0 = const()[name = string("input_123_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 256, 1]> m_attn_layers_2_ffn_conv_1_weight_to_fp16 = const()[name = string("m_attn_layers_2_ffn_conv_1_weight_to_fp16"), val = tensor<fp16, [1024, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14480768)))];
            tensor<fp16, [1024]> m_attn_layers_2_ffn_conv_1_bias_to_fp16 = const()[name = string("m_attn_layers_2_ffn_conv_1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15005120)))];
            tensor<fp16, [1, 1024, 128]> input_123_cast_fp16 = conv(bias = m_attn_layers_2_ffn_conv_1_bias_to_fp16, dilations = input_123_dilations_0, groups = input_123_groups_0, pad = input_123_pad_0, pad_type = input_123_pad_type_0, strides = input_123_strides_0, weight = m_attn_layers_2_ffn_conv_1_weight_to_fp16, x = input_121_cast_fp16)[name = string("input_123_cast_fp16")];
            tensor<fp16, [1, 1024, 128]> h_45_cast_fp16 = relu(x = input_123_cast_fp16)[name = string("h_45_cast_fp16")];
            tensor<fp16, [1, 1024, 128]> input_125_cast_fp16 = mul(x = h_45_cast_fp16, y = text_mask_to_fp16)[name = string("input_125_cast_fp16")];
            string h_47_pad_type_0 = const()[name = string("h_47_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_47_strides_0 = const()[name = string("h_47_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_47_pad_0 = const()[name = string("h_47_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_47_dilations_0 = const()[name = string("h_47_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_47_groups_0 = const()[name = string("h_47_groups_0"), val = int32(1)];
            tensor<fp16, [256, 1024, 1]> m_attn_layers_2_ffn_conv_2_weight_to_fp16 = const()[name = string("m_attn_layers_2_ffn_conv_2_weight_to_fp16"), val = tensor<fp16, [256, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15007232)))];
            tensor<fp16, [256]> m_attn_layers_2_ffn_conv_2_bias_to_fp16 = const()[name = string("m_attn_layers_2_ffn_conv_2_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15531584)))];
            tensor<fp16, [1, 256, 128]> h_47_cast_fp16 = conv(bias = m_attn_layers_2_ffn_conv_2_bias_to_fp16, dilations = h_47_dilations_0, groups = h_47_groups_0, pad = h_47_pad_0, pad_type = h_47_pad_type_0, strides = h_47_strides_0, weight = m_attn_layers_2_ffn_conv_2_weight_to_fp16, x = input_125_cast_fp16)[name = string("h_47_cast_fp16")];
            tensor<fp16, [1, 256, 128]> y_11_cast_fp16 = mul(x = h_47_cast_fp16, y = text_mask_to_fp16)[name = string("y_11_cast_fp16")];
            tensor<fp16, [1, 256, 128]> x_97_cast_fp16 = add(x = x_95_cast_fp16, y = y_11_cast_fp16)[name = string("x_97_cast_fp16")];
            tensor<int32, [3]> input_127_perm_0 = const()[name = string("input_127_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_931_axes_0 = const()[name = string("op_931_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [256]> m_attn_layers_2_norm_2_norm_weight_to_fp16 = const()[name = string("m_attn_layers_2_norm_2_norm_weight_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15532160)))];
            tensor<fp16, [256]> m_attn_layers_2_norm_2_norm_bias_to_fp16 = const()[name = string("m_attn_layers_2_norm_2_norm_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15532736)))];
            tensor<fp16, [1, 128, 256]> input_127_cast_fp16 = transpose(perm = input_127_perm_0, x = x_97_cast_fp16)[name = string("transpose_16")];
            tensor<fp16, [1, 128, 256]> var_931_cast_fp16 = layer_norm(axes = var_931_axes_0, beta = m_attn_layers_2_norm_2_norm_bias_to_fp16, epsilon = var_25_to_fp16, gamma = m_attn_layers_2_norm_2_norm_weight_to_fp16, x = input_127_cast_fp16)[name = string("op_931_cast_fp16")];
            tensor<int32, [3]> x_99_perm_0 = const()[name = string("x_99_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 256, 128]> x_99_cast_fp16 = transpose(perm = x_99_perm_0, x = var_931_cast_fp16)[name = string("transpose_15")];
            tensor<fp16, [1, 256, 128]> x_101_cast_fp16 = mul(x = x_99_cast_fp16, y = text_mask_to_fp16)[name = string("x_101_cast_fp16")];
            string var_953_pad_type_0 = const()[name = string("op_953_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_953_strides_0 = const()[name = string("op_953_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_953_pad_0 = const()[name = string("op_953_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_953_dilations_0 = const()[name = string("op_953_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_953_groups_0 = const()[name = string("op_953_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1]> m_attn_layers_3_attn_conv_q_weight_to_fp16 = const()[name = string("m_attn_layers_3_attn_conv_q_weight_to_fp16"), val = tensor<fp16, [256, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15533312)))];
            tensor<fp16, [256]> m_attn_layers_3_attn_conv_q_bias_to_fp16 = const()[name = string("m_attn_layers_3_attn_conv_q_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15664448)))];
            tensor<fp16, [1, 256, 128]> var_953_cast_fp16 = conv(bias = m_attn_layers_3_attn_conv_q_bias_to_fp16, dilations = var_953_dilations_0, groups = var_953_groups_0, pad = var_953_pad_0, pad_type = var_953_pad_type_0, strides = var_953_strides_0, weight = m_attn_layers_3_attn_conv_q_weight_to_fp16, x = x_101_cast_fp16)[name = string("op_953_cast_fp16")];
            tensor<int32, [4]> var_954 = const()[name = string("op_954"), val = tensor<int32, [4]>([1, 4, 64, 128])];
            tensor<fp16, [1, 4, 64, 128]> var_955_cast_fp16 = reshape(shape = var_954, x = var_953_cast_fp16)[name = string("op_955_cast_fp16")];
            tensor<int32, [4]> q_7_perm_0 = const()[name = string("q_7_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            string var_963_pad_type_0 = const()[name = string("op_963_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_963_strides_0 = const()[name = string("op_963_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_963_pad_0 = const()[name = string("op_963_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_963_dilations_0 = const()[name = string("op_963_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_963_groups_0 = const()[name = string("op_963_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1]> m_attn_layers_3_attn_conv_k_weight_to_fp16 = const()[name = string("m_attn_layers_3_attn_conv_k_weight_to_fp16"), val = tensor<fp16, [256, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15665024)))];
            tensor<fp16, [256]> m_attn_layers_3_attn_conv_k_bias_to_fp16 = const()[name = string("m_attn_layers_3_attn_conv_k_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15796160)))];
            tensor<fp16, [1, 256, 128]> var_963_cast_fp16 = conv(bias = m_attn_layers_3_attn_conv_k_bias_to_fp16, dilations = var_963_dilations_0, groups = var_963_groups_0, pad = var_963_pad_0, pad_type = var_963_pad_type_0, strides = var_963_strides_0, weight = m_attn_layers_3_attn_conv_k_weight_to_fp16, x = x_101_cast_fp16)[name = string("op_963_cast_fp16")];
            tensor<int32, [4]> var_964 = const()[name = string("op_964"), val = tensor<int32, [4]>([1, 4, 64, 128])];
            tensor<fp16, [1, 4, 64, 128]> var_965_cast_fp16 = reshape(shape = var_964, x = var_963_cast_fp16)[name = string("op_965_cast_fp16")];
            string var_973_pad_type_0 = const()[name = string("op_973_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_973_strides_0 = const()[name = string("op_973_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_973_pad_0 = const()[name = string("op_973_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_973_dilations_0 = const()[name = string("op_973_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_973_groups_0 = const()[name = string("op_973_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1]> m_attn_layers_3_attn_conv_v_weight_to_fp16 = const()[name = string("m_attn_layers_3_attn_conv_v_weight_to_fp16"), val = tensor<fp16, [256, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15796736)))];
            tensor<fp16, [256]> m_attn_layers_3_attn_conv_v_bias_to_fp16 = const()[name = string("m_attn_layers_3_attn_conv_v_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15927872)))];
            tensor<fp16, [1, 256, 128]> var_973_cast_fp16 = conv(bias = m_attn_layers_3_attn_conv_v_bias_to_fp16, dilations = var_973_dilations_0, groups = var_973_groups_0, pad = var_973_pad_0, pad_type = var_973_pad_type_0, strides = var_973_strides_0, weight = m_attn_layers_3_attn_conv_v_weight_to_fp16, x = x_101_cast_fp16)[name = string("op_973_cast_fp16")];
            tensor<int32, [4]> var_974 = const()[name = string("op_974"), val = tensor<int32, [4]>([1, 4, 64, 128])];
            tensor<fp16, [1, 4, 64, 128]> var_975_cast_fp16 = reshape(shape = var_974, x = var_973_cast_fp16)[name = string("op_975_cast_fp16")];
            fp16 var_977_to_fp16 = const()[name = string("op_977_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 4, 128, 64]> q_7_cast_fp16 = transpose(perm = q_7_perm_0, x = var_955_cast_fp16)[name = string("transpose_14")];
            tensor<fp16, [1, 4, 128, 64]> var_978_cast_fp16 = mul(x = q_7_cast_fp16, y = var_977_to_fp16)[name = string("op_978_cast_fp16")];
            bool scores_13_transpose_x_0 = const()[name = string("scores_13_transpose_x_0"), val = bool(false)];
            bool scores_13_transpose_y_0 = const()[name = string("scores_13_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 4, 128, 128]> scores_13_cast_fp16 = matmul(transpose_x = scores_13_transpose_x_0, transpose_y = scores_13_transpose_y_0, x = var_978_cast_fp16, y = var_965_cast_fp16)[name = string("scores_13_cast_fp16")];
            bool x_103_transpose_x_0 = const()[name = string("x_103_transpose_x_0"), val = bool(false)];
            bool x_103_transpose_y_0 = const()[name = string("x_103_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 1, 64, 255]> var_1002_to_fp16 = const()[name = string("op_1002_to_fp16"), val = tensor<fp16, [1, 1, 64, 255]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15928448)))];
            tensor<fp16, [1, 4, 128, 255]> x_103_cast_fp16 = matmul(transpose_x = x_103_transpose_x_0, transpose_y = x_103_transpose_y_0, x = var_978_cast_fp16, y = var_1002_to_fp16)[name = string("x_103_cast_fp16")];
            tensor<int32, [8]> x_105_pad_0 = const()[name = string("x_105_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 1])];
            string x_105_mode_0 = const()[name = string("x_105_mode_0"), val = string("constant")];
            fp16 const_54_to_fp16 = const()[name = string("const_54_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 128, 256]> x_105_cast_fp16 = pad(constant_val = const_54_to_fp16, mode = x_105_mode_0, pad = x_105_pad_0, x = x_103_cast_fp16)[name = string("x_105_cast_fp16")];
            tensor<int32, [3]> var_1014 = const()[name = string("op_1014"), val = tensor<int32, [3]>([1, 4, 32768])];
            tensor<fp16, [1, 4, 32768]> input_131_cast_fp16 = reshape(shape = var_1014, x = x_105_cast_fp16)[name = string("input_131_cast_fp16")];
            tensor<int32, [6]> x_107_pad_0 = const()[name = string("x_107_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 127])];
            string x_107_mode_0 = const()[name = string("x_107_mode_0"), val = string("constant")];
            fp16 const_55_to_fp16 = const()[name = string("const_55_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 32895]> x_107_cast_fp16 = pad(constant_val = const_55_to_fp16, mode = x_107_mode_0, pad = x_107_pad_0, x = input_131_cast_fp16)[name = string("x_107_cast_fp16")];
            tensor<int32, [4]> var_1029 = const()[name = string("op_1029"), val = tensor<int32, [4]>([1, 4, 129, 255])];
            tensor<fp16, [1, 4, 129, 255]> x_109_cast_fp16 = reshape(shape = var_1029, x = x_107_cast_fp16)[name = string("x_109_cast_fp16")];
            tensor<int32, [4]> var_1036_begin_0 = const()[name = string("op_1036_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1036_end_0 = const()[name = string("op_1036_end_0"), val = tensor<int32, [4]>([1, 4, 128, 255])];
            tensor<bool, [4]> var_1036_end_mask_0 = const()[name = string("op_1036_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])];
            tensor<fp16, [1, 4, 128, 255]> var_1036_cast_fp16 = slice_by_index(begin = var_1036_begin_0, end = var_1036_end_0, end_mask = var_1036_end_mask_0, x = x_109_cast_fp16)[name = string("op_1036_cast_fp16")];
            tensor<int32, [4]> var_1037_begin_0 = const()[name = string("op_1037_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 127])];
            tensor<int32, [4]> var_1037_end_0 = const()[name = string("op_1037_end_0"), val = tensor<int32, [4]>([1, 4, 128, 255])];
            tensor<bool, [4]> var_1037_end_mask_0 = const()[name = string("op_1037_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 128]> var_1037_cast_fp16 = slice_by_index(begin = var_1037_begin_0, end = var_1037_end_0, end_mask = var_1037_end_mask_0, x = var_1036_cast_fp16)[name = string("op_1037_cast_fp16")];
            tensor<fp16, [1, 4, 128, 128]> scores_cast_fp16 = add(x = scores_13_cast_fp16, y = var_1037_cast_fp16)[name = string("scores_cast_fp16")];
            tensor<fp16, [1, 4, 128, 128]> input_133_cast_fp16 = sub(x = scores_cast_fp16, y = var_418_cast_fp16)[name = string("input_133_cast_fp16")];
            tensor<fp16, [1, 4, 128, 128]> p_attn_cast_fp16 = softmax(axis = var_29, x = input_133_cast_fp16)[name = string("p_attn_cast_fp16")];
            bool out_25_transpose_x_1 = const()[name = string("out_25_transpose_x_1"), val = bool(false)];
            bool out_25_transpose_y_1 = const()[name = string("out_25_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 4, 128, 64]> out_25_cast_fp16 = matmul(transpose_x = out_25_transpose_x_1, transpose_y = out_25_transpose_y_1, x = p_attn_cast_fp16, y = var_975_cast_fp16)[name = string("out_25_cast_fp16")];
            tensor<int32, [8]> x_111_pad_0 = const()[name = string("x_111_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 127])];
            string x_111_mode_0 = const()[name = string("x_111_mode_0"), val = string("constant")];
            fp16 const_60_to_fp16 = const()[name = string("const_60_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 128, 255]> x_111_cast_fp16 = pad(constant_val = const_60_to_fp16, mode = x_111_mode_0, pad = x_111_pad_0, x = p_attn_cast_fp16)[name = string("x_111_cast_fp16")];
            tensor<int32, [3]> var_1079 = const()[name = string("op_1079"), val = tensor<int32, [3]>([1, 4, 32640])];
            tensor<fp16, [1, 4, 32640]> input_137_cast_fp16 = reshape(shape = var_1079, x = x_111_cast_fp16)[name = string("input_137_cast_fp16")];
            tensor<int32, [6]> x_113_pad_0 = const()[name = string("x_113_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 128, 0])];
            string x_113_mode_0 = const()[name = string("x_113_mode_0"), val = string("constant")];
            fp16 const_61_to_fp16 = const()[name = string("const_61_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 32768]> x_113_cast_fp16 = pad(constant_val = const_61_to_fp16, mode = x_113_mode_0, pad = x_113_pad_0, x = input_137_cast_fp16)[name = string("x_113_cast_fp16")];
            tensor<int32, [4]> var_1086 = const()[name = string("op_1086"), val = tensor<int32, [4]>([1, 4, 128, 256])];
            tensor<fp16, [1, 4, 128, 256]> x_115_cast_fp16 = reshape(shape = var_1086, x = x_113_cast_fp16)[name = string("x_115_cast_fp16")];
            tensor<int32, [4]> rel_weights_begin_0 = const()[name = string("rel_weights_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1])];
            tensor<int32, [4]> rel_weights_end_0 = const()[name = string("rel_weights_end_0"), val = tensor<int32, [4]>([1, 4, 128, 256])];
            tensor<bool, [4]> rel_weights_end_mask_0 = const()[name = string("rel_weights_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 255]> rel_weights_cast_fp16 = slice_by_index(begin = rel_weights_begin_0, end = rel_weights_end_0, end_mask = rel_weights_end_mask_0, x = x_115_cast_fp16)[name = string("rel_weights_cast_fp16")];
            bool var_1093_transpose_x_0 = const()[name = string("op_1093_transpose_x_0"), val = bool(false)];
            bool var_1093_transpose_y_0 = const()[name = string("op_1093_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 1, 255, 64]> var_1092_to_fp16 = const()[name = string("op_1092_to_fp16"), val = tensor<fp16, [1, 1, 255, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15961152)))];
            tensor<fp16, [1, 4, 128, 64]> var_1093_cast_fp16 = matmul(transpose_x = var_1093_transpose_x_0, transpose_y = var_1093_transpose_y_0, x = rel_weights_cast_fp16, y = var_1092_to_fp16)[name = string("op_1093_cast_fp16")];
            tensor<fp16, [1, 4, 128, 64]> out_27_cast_fp16 = add(x = out_25_cast_fp16, y = var_1093_cast_fp16)[name = string("out_27_cast_fp16")];
            tensor<int32, [4]> var_1095_perm_0 = const()[name = string("op_1095_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [3]> var_1096 = const()[name = string("op_1096"), val = tensor<int32, [3]>([1, 256, 128])];
            tensor<fp16, [1, 4, 64, 128]> var_1095_cast_fp16 = transpose(perm = var_1095_perm_0, x = out_27_cast_fp16)[name = string("transpose_13")];
            tensor<fp16, [1, 256, 128]> input_139_cast_fp16 = reshape(shape = var_1096, x = var_1095_cast_fp16)[name = string("input_139_cast_fp16")];
            string y_13_pad_type_0 = const()[name = string("y_13_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> y_13_strides_0 = const()[name = string("y_13_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> y_13_pad_0 = const()[name = string("y_13_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> y_13_dilations_0 = const()[name = string("y_13_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 y_13_groups_0 = const()[name = string("y_13_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1]> m_attn_layers_3_attn_conv_o_weight_to_fp16 = const()[name = string("m_attn_layers_3_attn_conv_o_weight_to_fp16"), val = tensor<fp16, [256, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15993856)))];
            tensor<fp16, [256]> m_attn_layers_3_attn_conv_o_bias_to_fp16 = const()[name = string("m_attn_layers_3_attn_conv_o_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16124992)))];
            tensor<fp16, [1, 256, 128]> y_13_cast_fp16 = conv(bias = m_attn_layers_3_attn_conv_o_bias_to_fp16, dilations = y_13_dilations_0, groups = y_13_groups_0, pad = y_13_pad_0, pad_type = y_13_pad_type_0, strides = y_13_strides_0, weight = m_attn_layers_3_attn_conv_o_weight_to_fp16, x = input_139_cast_fp16)[name = string("y_13_cast_fp16")];
            tensor<fp16, [1, 256, 128]> x_117_cast_fp16 = add(x = x_101_cast_fp16, y = y_13_cast_fp16)[name = string("x_117_cast_fp16")];
            tensor<int32, [3]> input_141_perm_0 = const()[name = string("input_141_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1111_axes_0 = const()[name = string("op_1111_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [256]> m_attn_layers_3_norm_1_norm_weight_to_fp16 = const()[name = string("m_attn_layers_3_norm_1_norm_weight_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16125568)))];
            tensor<fp16, [256]> m_attn_layers_3_norm_1_norm_bias_to_fp16 = const()[name = string("m_attn_layers_3_norm_1_norm_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16126144)))];
            tensor<fp16, [1, 128, 256]> input_141_cast_fp16 = transpose(perm = input_141_perm_0, x = x_117_cast_fp16)[name = string("transpose_12")];
            tensor<fp16, [1, 128, 256]> var_1111_cast_fp16 = layer_norm(axes = var_1111_axes_0, beta = m_attn_layers_3_norm_1_norm_bias_to_fp16, epsilon = var_25_to_fp16, gamma = m_attn_layers_3_norm_1_norm_weight_to_fp16, x = input_141_cast_fp16)[name = string("op_1111_cast_fp16")];
            tensor<int32, [3]> x_119_perm_0 = const()[name = string("x_119_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 256, 128]> x_119_cast_fp16 = transpose(perm = x_119_perm_0, x = var_1111_cast_fp16)[name = string("transpose_11")];
            tensor<fp16, [1, 256, 128]> input_143_cast_fp16 = mul(x = x_119_cast_fp16, y = text_mask_to_fp16)[name = string("input_143_cast_fp16")];
            string input_145_pad_type_0 = const()[name = string("input_145_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_145_strides_0 = const()[name = string("input_145_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_145_pad_0 = const()[name = string("input_145_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_145_dilations_0 = const()[name = string("input_145_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_145_groups_0 = const()[name = string("input_145_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 256, 1]> m_attn_layers_3_ffn_conv_1_weight_to_fp16 = const()[name = string("m_attn_layers_3_ffn_conv_1_weight_to_fp16"), val = tensor<fp16, [1024, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16126720)))];
            tensor<fp16, [1024]> m_attn_layers_3_ffn_conv_1_bias_to_fp16 = const()[name = string("m_attn_layers_3_ffn_conv_1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16651072)))];
            tensor<fp16, [1, 1024, 128]> input_145_cast_fp16 = conv(bias = m_attn_layers_3_ffn_conv_1_bias_to_fp16, dilations = input_145_dilations_0, groups = input_145_groups_0, pad = input_145_pad_0, pad_type = input_145_pad_type_0, strides = input_145_strides_0, weight = m_attn_layers_3_ffn_conv_1_weight_to_fp16, x = input_143_cast_fp16)[name = string("input_145_cast_fp16")];
            tensor<fp16, [1, 1024, 128]> h_49_cast_fp16 = relu(x = input_145_cast_fp16)[name = string("h_49_cast_fp16")];
            tensor<fp16, [1, 1024, 128]> input_147_cast_fp16 = mul(x = h_49_cast_fp16, y = text_mask_to_fp16)[name = string("input_147_cast_fp16")];
            string h_pad_type_0 = const()[name = string("h_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_strides_0 = const()[name = string("h_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_pad_0 = const()[name = string("h_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_dilations_0 = const()[name = string("h_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_groups_0 = const()[name = string("h_groups_0"), val = int32(1)];
            tensor<fp16, [256, 1024, 1]> m_attn_layers_3_ffn_conv_2_weight_to_fp16 = const()[name = string("m_attn_layers_3_ffn_conv_2_weight_to_fp16"), val = tensor<fp16, [256, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16653184)))];
            tensor<fp16, [256]> m_attn_layers_3_ffn_conv_2_bias_to_fp16 = const()[name = string("m_attn_layers_3_ffn_conv_2_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17177536)))];
            tensor<fp16, [1, 256, 128]> h_cast_fp16 = conv(bias = m_attn_layers_3_ffn_conv_2_bias_to_fp16, dilations = h_dilations_0, groups = h_groups_0, pad = h_pad_0, pad_type = h_pad_type_0, strides = h_strides_0, weight = m_attn_layers_3_ffn_conv_2_weight_to_fp16, x = input_147_cast_fp16)[name = string("h_cast_fp16")];
            tensor<fp16, [1, 256, 128]> y_cast_fp16 = mul(x = h_cast_fp16, y = text_mask_to_fp16)[name = string("y_cast_fp16")];
            tensor<fp16, [1, 256, 128]> x_121_cast_fp16 = add(x = x_119_cast_fp16, y = y_cast_fp16)[name = string("x_121_cast_fp16")];
            tensor<int32, [3]> input_149_perm_0 = const()[name = string("input_149_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1139_axes_0 = const()[name = string("op_1139_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [256]> m_attn_layers_3_norm_2_norm_weight_to_fp16 = const()[name = string("m_attn_layers_3_norm_2_norm_weight_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17178112)))];
            tensor<fp16, [256]> m_attn_layers_3_norm_2_norm_bias_to_fp16 = const()[name = string("m_attn_layers_3_norm_2_norm_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17178688)))];
            tensor<fp16, [1, 128, 256]> input_149_cast_fp16 = transpose(perm = input_149_perm_0, x = x_121_cast_fp16)[name = string("transpose_10")];
            tensor<fp16, [1, 128, 256]> var_1139_cast_fp16 = layer_norm(axes = var_1139_axes_0, beta = m_attn_layers_3_norm_2_norm_bias_to_fp16, epsilon = var_25_to_fp16, gamma = m_attn_layers_3_norm_2_norm_weight_to_fp16, x = input_149_cast_fp16)[name = string("op_1139_cast_fp16")];
            tensor<int32, [3]> x_123_perm_0 = const()[name = string("x_123_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 256, 128]> x_123_cast_fp16 = transpose(perm = x_123_perm_0, x = var_1139_cast_fp16)[name = string("transpose_9")];
            tensor<fp16, [1, 256, 128]> x_125_cast_fp16 = mul(x = x_123_cast_fp16, y = text_mask_to_fp16)[name = string("x_125_cast_fp16")];
            tensor<fp16, [1, 256, 128]> var_1142_cast_fp16 = add(x = x_125_cast_fp16, y = x_27_cast_fp16)[name = string("op_1142_cast_fp16")];
            tensor<fp16, [1, 256, 128]> text_1_cast_fp16 = mul(x = var_1142_cast_fp16, y = text_mask_to_fp16)[name = string("text_1_cast_fp16")];
            tensor<int32, [3]> text_3_perm_0 = const()[name = string("text_3_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [256, 256]> m_speech_prompted_attention1_W_query_linear_weight_to_fp16 = const()[name = string("m_speech_prompted_attention1_W_query_linear_weight_to_fp16"), val = tensor<fp16, [256, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17179264)))];
            tensor<fp16, [256]> m_speech_prompted_attention1_W_query_linear_bias_to_fp16 = const()[name = string("m_speech_prompted_attention1_W_query_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17310400)))];
            tensor<fp16, [1, 128, 256]> text_3_cast_fp16 = transpose(perm = text_3_perm_0, x = text_1_cast_fp16)[name = string("transpose_8")];
            tensor<fp16, [1, 128, 256]> linear_0_cast_fp16 = linear(bias = m_speech_prompted_attention1_W_query_linear_bias_to_fp16, weight = m_speech_prompted_attention1_W_query_linear_weight_to_fp16, x = text_3_cast_fp16)[name = string("linear_0_cast_fp16")];
            tensor<int32, [4]> var_1163 = const()[name = string("op_1163"), val = tensor<int32, [4]>([1, 128, 2, 128])];
            tensor<fp16, [1, 128, 2, 128]> var_1164_cast_fp16 = reshape(shape = var_1163, x = linear_0_cast_fp16)[name = string("op_1164_cast_fp16")];
            tensor<int32, [4]> q_9_perm_0 = const()[name = string("q_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            string style_ttl_to_fp16_dtype_0 = const()[name = string("style_ttl_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [256, 256]> m_speech_prompted_attention1_W_value_linear_weight_to_fp16 = const()[name = string("m_speech_prompted_attention1_W_value_linear_weight_to_fp16"), val = tensor<fp16, [256, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17310976)))];
            tensor<fp16, [256]> m_speech_prompted_attention1_W_value_linear_bias_to_fp16 = const()[name = string("m_speech_prompted_attention1_W_value_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17442112)))];
            tensor<fp16, [1, 50, 256]> style_ttl_to_fp16 = cast(dtype = style_ttl_to_fp16_dtype_0, x = style_ttl)[name = string("cast_47")];
            tensor<fp16, [1, 50, 256]> linear_2_cast_fp16 = linear(bias = m_speech_prompted_attention1_W_value_linear_bias_to_fp16, weight = m_speech_prompted_attention1_W_value_linear_weight_to_fp16, x = style_ttl_to_fp16)[name = string("linear_2_cast_fp16")];
            tensor<int32, [4]> var_1177 = const()[name = string("op_1177"), val = tensor<int32, [4]>([1, 50, 2, 128])];
            tensor<fp16, [1, 50, 2, 128]> var_1178_cast_fp16 = reshape(shape = var_1177, x = linear_2_cast_fp16)[name = string("op_1178_cast_fp16")];
            tensor<int32, [4]> v_9_perm_0 = const()[name = string("v_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_1182_transpose_x_0 = const()[name = string("op_1182_transpose_x_0"), val = bool(false)];
            bool var_1182_transpose_y_0 = const()[name = string("op_1182_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 2, 128, 50]> var_1181_to_fp16 = const()[name = string("op_1181_to_fp16"), val = tensor<fp16, [1, 2, 128, 50]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17442688)))];
            tensor<fp16, [1, 2, 128, 128]> q_9_cast_fp16 = transpose(perm = q_9_perm_0, x = var_1164_cast_fp16)[name = string("transpose_7")];
            tensor<fp16, [1, 2, 128, 50]> var_1182_cast_fp16 = matmul(transpose_x = var_1182_transpose_x_0, transpose_y = var_1182_transpose_y_0, x = q_9_cast_fp16, y = var_1181_to_fp16)[name = string("op_1182_cast_fp16")];
            fp16 _inversed_input_151_y_0_to_fp16 = const()[name = string("_inversed_input_151_y_0_to_fp16"), val = fp16(0x1p-4)];
            tensor<fp16, [1, 2, 128, 50]> _inversed_input_151_cast_fp16 = mul(x = var_1182_cast_fp16, y = _inversed_input_151_y_0_to_fp16)[name = string("_inversed_input_151_cast_fp16")];
            tensor<fp16, [1, 2, 128, 50]> attn_1_cast_fp16 = softmax(axis = var_29, x = _inversed_input_151_cast_fp16)[name = string("attn_1_cast_fp16")];
            tensor<int32, [3]> var_1186_perm_0 = const()[name = string("op_1186_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> mask_q_1_axes_0 = const()[name = string("mask_q_1_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 128, 1]> var_1186_cast_fp16 = transpose(perm = var_1186_perm_0, x = text_mask_to_fp16)[name = string("transpose_5")];
            tensor<fp16, [1, 1, 128, 1]> mask_q_1_cast_fp16 = expand_dims(axes = mask_q_1_axes_0, x = var_1186_cast_fp16)[name = string("mask_q_1_cast_fp16")];
            tensor<fp16, [1, 2, 128, 50]> attn_3_cast_fp16 = mul(x = attn_1_cast_fp16, y = mask_q_1_cast_fp16)[name = string("attn_3_cast_fp16")];
            bool out_29_transpose_x_0 = const()[name = string("out_29_transpose_x_0"), val = bool(false)];
            bool out_29_transpose_y_0 = const()[name = string("out_29_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 2, 50, 128]> v_9_cast_fp16 = transpose(perm = v_9_perm_0, x = var_1178_cast_fp16)[name = string("transpose_6")];
            tensor<fp16, [1, 2, 128, 128]> out_29_cast_fp16 = matmul(transpose_x = out_29_transpose_x_0, transpose_y = out_29_transpose_y_0, x = attn_3_cast_fp16, y = v_9_cast_fp16)[name = string("out_29_cast_fp16")];
            tensor<int32, [4]> var_1190_perm_0 = const()[name = string("op_1190_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1191 = const()[name = string("op_1191"), val = tensor<int32, [3]>([1, 128, 256])];
            tensor<fp16, [1, 128, 2, 128]> var_1190_cast_fp16 = transpose(perm = var_1190_perm_0, x = out_29_cast_fp16)[name = string("transpose_4")];
            tensor<fp16, [1, 128, 256]> input_153_cast_fp16 = reshape(shape = var_1191, x = var_1190_cast_fp16)[name = string("input_153_cast_fp16")];
            tensor<fp16, [256, 256]> m_speech_prompted_attention1_out_fc_linear_weight_to_fp16 = const()[name = string("m_speech_prompted_attention1_out_fc_linear_weight_to_fp16"), val = tensor<fp16, [256, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17468352)))];
            tensor<fp16, [256]> m_speech_prompted_attention1_out_fc_linear_bias_to_fp16 = const()[name = string("m_speech_prompted_attention1_out_fc_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17599488)))];
            tensor<fp16, [1, 128, 256]> linear_3_cast_fp16 = linear(bias = m_speech_prompted_attention1_out_fc_linear_bias_to_fp16, weight = m_speech_prompted_attention1_out_fc_linear_weight_to_fp16, x = input_153_cast_fp16)[name = string("linear_3_cast_fp16")];
            tensor<fp16, [1, 128, 256]> attn1_out_cast_fp16 = mul(x = linear_3_cast_fp16, y = var_1186_cast_fp16)[name = string("attn1_out_cast_fp16")];
            tensor<fp16, [1, 128, 256]> text_cast_fp16 = add(x = text_3_cast_fp16, y = attn1_out_cast_fp16)[name = string("text_cast_fp16")];
            tensor<fp16, [256, 256]> m_speech_prompted_attention2_W_query_linear_weight_to_fp16 = const()[name = string("m_speech_prompted_attention2_W_query_linear_weight_to_fp16"), val = tensor<fp16, [256, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17600064)))];
            tensor<fp16, [256]> m_speech_prompted_attention2_W_query_linear_bias_to_fp16 = const()[name = string("m_speech_prompted_attention2_W_query_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17731200)))];
            tensor<fp16, [1, 128, 256]> linear_4_cast_fp16 = linear(bias = m_speech_prompted_attention2_W_query_linear_bias_to_fp16, weight = m_speech_prompted_attention2_W_query_linear_weight_to_fp16, x = text_cast_fp16)[name = string("linear_4_cast_fp16")];
            tensor<int32, [4]> var_1211 = const()[name = string("op_1211"), val = tensor<int32, [4]>([1, 128, 2, 128])];
            tensor<fp16, [1, 128, 2, 128]> var_1212_cast_fp16 = reshape(shape = var_1211, x = linear_4_cast_fp16)[name = string("op_1212_cast_fp16")];
            tensor<int32, [4]> q_perm_0 = const()[name = string("q_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [256, 256]> m_speech_prompted_attention2_W_value_linear_weight_to_fp16 = const()[name = string("m_speech_prompted_attention2_W_value_linear_weight_to_fp16"), val = tensor<fp16, [256, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17731776)))];
            tensor<fp16, [256]> m_speech_prompted_attention2_W_value_linear_bias_to_fp16 = const()[name = string("m_speech_prompted_attention2_W_value_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17862912)))];
            tensor<fp16, [1, 50, 256]> linear_6_cast_fp16 = linear(bias = m_speech_prompted_attention2_W_value_linear_bias_to_fp16, weight = m_speech_prompted_attention2_W_value_linear_weight_to_fp16, x = style_ttl_to_fp16)[name = string("linear_6_cast_fp16")];
            tensor<int32, [4]> var_1225 = const()[name = string("op_1225"), val = tensor<int32, [4]>([1, 50, 2, 128])];
            tensor<fp16, [1, 50, 2, 128]> var_1226_cast_fp16 = reshape(shape = var_1225, x = linear_6_cast_fp16)[name = string("op_1226_cast_fp16")];
            tensor<int32, [4]> v_perm_0 = const()[name = string("v_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_1230_transpose_x_0 = const()[name = string("op_1230_transpose_x_0"), val = bool(false)];
            bool var_1230_transpose_y_0 = const()[name = string("op_1230_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 2, 128, 50]> var_1229_to_fp16 = const()[name = string("op_1229_to_fp16"), val = tensor<fp16, [1, 2, 128, 50]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17863488)))];
            tensor<fp16, [1, 2, 128, 128]> q_cast_fp16 = transpose(perm = q_perm_0, x = var_1212_cast_fp16)[name = string("transpose_3")];
            tensor<fp16, [1, 2, 128, 50]> var_1230_cast_fp16 = matmul(transpose_x = var_1230_transpose_x_0, transpose_y = var_1230_transpose_y_0, x = q_cast_fp16, y = var_1229_to_fp16)[name = string("op_1230_cast_fp16")];
            fp16 _inversed_input_155_y_0_to_fp16 = const()[name = string("_inversed_input_155_y_0_to_fp16"), val = fp16(0x1p-4)];
            tensor<fp16, [1, 2, 128, 50]> _inversed_input_155_cast_fp16 = mul(x = var_1230_cast_fp16, y = _inversed_input_155_y_0_to_fp16)[name = string("_inversed_input_155_cast_fp16")];
            tensor<fp16, [1, 2, 128, 50]> attn_5_cast_fp16 = softmax(axis = var_29, x = _inversed_input_155_cast_fp16)[name = string("attn_5_cast_fp16")];
            tensor<fp16, [1, 2, 128, 50]> attn_7_cast_fp16 = mul(x = attn_5_cast_fp16, y = mask_q_1_cast_fp16)[name = string("attn_7_cast_fp16")];
            bool out_33_transpose_x_0 = const()[name = string("out_33_transpose_x_0"), val = bool(false)];
            bool out_33_transpose_y_0 = const()[name = string("out_33_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 2, 50, 128]> v_cast_fp16 = transpose(perm = v_perm_0, x = var_1226_cast_fp16)[name = string("transpose_2")];
            tensor<fp16, [1, 2, 128, 128]> out_33_cast_fp16 = matmul(transpose_x = out_33_transpose_x_0, transpose_y = out_33_transpose_y_0, x = attn_7_cast_fp16, y = v_cast_fp16)[name = string("out_33_cast_fp16")];
            tensor<int32, [4]> var_1238_perm_0 = const()[name = string("op_1238_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1239 = const()[name = string("op_1239"), val = tensor<int32, [3]>([1, 128, 256])];
            tensor<fp16, [1, 128, 2, 128]> var_1238_cast_fp16 = transpose(perm = var_1238_perm_0, x = out_33_cast_fp16)[name = string("transpose_1")];
            tensor<fp16, [1, 128, 256]> input_157_cast_fp16 = reshape(shape = var_1239, x = var_1238_cast_fp16)[name = string("input_157_cast_fp16")];
            tensor<fp16, [256, 256]> m_speech_prompted_attention2_out_fc_linear_weight_to_fp16 = const()[name = string("m_speech_prompted_attention2_out_fc_linear_weight_to_fp16"), val = tensor<fp16, [256, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17889152)))];
            tensor<fp16, [256]> m_speech_prompted_attention2_out_fc_linear_bias_to_fp16 = const()[name = string("m_speech_prompted_attention2_out_fc_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18020288)))];
            tensor<fp16, [1, 128, 256]> linear_7_cast_fp16 = linear(bias = m_speech_prompted_attention2_out_fc_linear_bias_to_fp16, weight = m_speech_prompted_attention2_out_fc_linear_weight_to_fp16, x = input_157_cast_fp16)[name = string("linear_7_cast_fp16")];
            tensor<fp16, [1, 128, 256]> attn2_out_cast_fp16 = mul(x = linear_7_cast_fp16, y = var_1186_cast_fp16)[name = string("attn2_out_cast_fp16")];
            tensor<fp16, [1, 128, 256]> y_final_cast_fp16 = add(x = text_3_cast_fp16, y = attn2_out_cast_fp16)[name = string("y_final_cast_fp16")];
            tensor<int32, [1]> var_1254_axes_0 = const()[name = string("op_1254_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [256]> m_speech_prompted_norm_norm_weight_to_fp16 = const()[name = string("m_speech_prompted_norm_norm_weight_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18020864)))];
            tensor<fp16, [256]> m_speech_prompted_norm_norm_bias_to_fp16 = const()[name = string("m_speech_prompted_norm_norm_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18021440)))];
            tensor<fp16, [1, 128, 256]> var_1254_cast_fp16 = layer_norm(axes = var_1254_axes_0, beta = m_speech_prompted_norm_norm_bias_to_fp16, epsilon = var_25_to_fp16, gamma = m_speech_prompted_norm_norm_weight_to_fp16, x = y_final_cast_fp16)[name = string("op_1254_cast_fp16")];
            tensor<int32, [3]> x_perm_0 = const()[name = string("x_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 256, 128]> x_cast_fp16 = transpose(perm = x_perm_0, x = var_1254_cast_fp16)[name = string("transpose_0")];
            tensor<fp16, [1, 256, 128]> var_1256_cast_fp16 = mul(x = x_cast_fp16, y = text_mask_to_fp16)[name = string("op_1256_cast_fp16")];
            string var_1256_cast_fp16_to_fp32_dtype_0 = const()[name = string("op_1256_cast_fp16_to_fp32_dtype_0"), val = string("fp32")];
            tensor<fp32, [1, 256, 128]> text_emb = cast(dtype = var_1256_cast_fp16_to_fp32_dtype_0, x = var_1256_cast_fp16)[name = string("cast_46")];
        } -> (text_emb);
}