program(1.3)
[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3600.16.1"}, {"coremlc-version", "3600.25.2"}, {"coremltools-component-torch", "2.10.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})]
{
    func main<ios18>(tensor<fp32, [1]> current_step, tensor<fp32, [1, 1, ?]> latent_mask, tensor<fp32, [1, 144, ?]> noisy_latent, tensor<fp32, [1, 50, 256]> style_ttl, tensor<fp32, [1, 256, ?]> text_emb, tensor<fp32, [1, 1, ?]> text_mask, tensor<fp32, [1]> total_step) [FlexibleShapeInformation = tuple<tuple<string, dict<string, tensor<int32, [?]>>>, tuple<string, dict<string, list<tensor<int32, [2]>, ?>>>>((("DefaultShapes", {{"latent_mask", [1, 1, 24]}, {"noisy_latent", [1, 144, 24]}, {"text_emb", [1, 256, 24]}, {"text_mask", [1, 1, 24]}}), ("RangeDims", {{"latent_mask", [[1, 1], [1, 1], [17, 512]]}, {"noisy_latent", [[1, 1], [144, 144], [17, 512]]}, {"text_emb", [[1, 1], [256, 256], [17, 512]]}, {"text_mask", [[1, 1], [1, 1], [17, 512]]}})))] {
            string text_emb_to_fp16_dtype_0 = const()[name = string("text_emb_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 256, ?]> text_emb_to_fp16 = cast(dtype = text_emb_to_fp16_dtype_0, x = text_emb)[name = string("cast_104")];
            tensor<int32, [3]> var_34_shape_cast_fp16 = shape(x = text_emb_to_fp16)[name = string("op_34_shape_cast_fp16")];
            int32 gather_1_axis_0 = const()[name = string("gather_1_axis_0"), val = int32(0)];
            int32 gather_1_batch_dims_0 = const()[name = string("gather_1_batch_dims_0"), val = int32(0)];
            bool gather_1_validate_indices_0 = const()[name = string("gather_1_validate_indices_0"), val = bool(false)];
            string var_34_shape_cast_fp16_to_int16_dtype_0 = const()[name = string("op_34_shape_cast_fp16_to_int16_dtype_0"), val = string("int16")];
            uint16 gather_1_indices_0_to_uint16 = const()[name = string("gather_1_indices_0_to_uint16"), val = uint16(2)];
            tensor<int16, [3]> var_34_shape_cast_fp16_to_int16 = cast(dtype = var_34_shape_cast_fp16_to_int16_dtype_0, x = var_34_shape_cast_fp16)[name = string("cast_103")];
            int16 gather_1_cast_uint16 = gather(axis = gather_1_axis_0, batch_dims = gather_1_batch_dims_0, indices = gather_1_indices_0_to_uint16, validate_indices = gather_1_validate_indices_0, x = var_34_shape_cast_fp16_to_int16)[name = string("gather_1_cast_uint16")];
            string gather_1_cast_uint16_to_int32_dtype_0 = const()[name = string("gather_1_cast_uint16_to_int32_dtype_0"), val = string("int32")];
            int32 var_38 = const()[name = string("op_38"), val = int32(-1)];
            int32 var_39 = const()[name = string("op_39"), val = int32(0)];
            int32 gather_2 = const()[name = string("gather_2"), val = int32(1)];
            int32 concat_0_axis_0 = const()[name = string("concat_0_axis_0"), val = int32(0)];
            bool concat_0_interleave_0 = const()[name = string("concat_0_interleave_0"), val = bool(false)];
            int32 gather_1_cast_uint16_to_int32 = cast(dtype = gather_1_cast_uint16_to_int32_dtype_0, x = gather_1_cast_uint16)[name = string("cast_102")];
            tensor<int32, [3]> concat_0 = concat(axis = concat_0_axis_0, interleave = concat_0_interleave_0, values = (gather_2, var_38, gather_1_cast_uint16_to_int32))[name = string("concat_0")];
            tensor<int32, [3]> shape_0 = const()[name = string("shape_0"), val = tensor<int32, [3]>([1, 256, 1])];
            int32 equal_0_y_0 = const()[name = string("equal_0_y_0"), val = int32(-1)];
            tensor<bool, [3]> equal_0 = equal(x = concat_0, y = equal_0_y_0)[name = string("equal_0")];
            tensor<int32, [3]> select_0 = select(a = shape_0, b = concat_0, cond = equal_0)[name = string("select_0")];
            tensor<int32, [3]> real_div_0 = real_div(x = select_0, y = shape_0)[name = string("real_div_0")];
            tensor<fp16, [1, 256, 1]> uncond_masker_text_special_token_to_fp16 = const()[name = string("uncond_masker_text_special_token_to_fp16"), val = tensor<fp16, [1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
            tensor<fp16, [?, ?, ?]> text_uncond_cast_fp16 = tile(reps = real_div_0, x = uncond_masker_text_special_token_to_fp16)[name = string("text_uncond_cast_fp16")];
            bool text_emb_interleave_0 = const()[name = string("text_emb_interleave_0"), val = bool(false)];
            tensor<fp16, [?, 256, ?]> text_emb_cast_fp16 = concat(axis = var_39, interleave = text_emb_interleave_0, values = (text_emb_to_fp16, text_uncond_cast_fp16))[name = string("text_emb_cast_fp16")];
            bool input_95_interleave_0 = const()[name = string("input_95_interleave_0"), val = bool(false)];
            string style_ttl_to_fp16_dtype_0 = const()[name = string("style_ttl_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 50, 256]> uncond_masker_style_value_special_token_to_fp16 = const()[name = string("uncond_masker_style_value_special_token_to_fp16"), val = tensor<fp16, [1, 50, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(640)))];
            tensor<fp16, [1, 50, 256]> style_ttl_to_fp16 = cast(dtype = style_ttl_to_fp16_dtype_0, x = style_ttl)[name = string("cast_101")];
            tensor<fp16, [2, 50, 256]> input_95_cast_fp16 = concat(axis = var_39, interleave = input_95_interleave_0, values = (style_ttl_to_fp16, uncond_masker_style_value_special_token_to_fp16))[name = string("input_95_cast_fp16")];
            int32 var_64 = const()[name = string("op_64"), val = int32(0)];
            bool input_5_interleave_0 = const()[name = string("input_5_interleave_0"), val = bool(false)];
            string noisy_latent_to_fp16_dtype_0 = const()[name = string("noisy_latent_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 144, ?]> noisy_latent_to_fp16 = cast(dtype = noisy_latent_to_fp16_dtype_0, x = noisy_latent)[name = string("cast_100")];
            tensor<fp16, [2, 144, ?]> input_5_cast_fp16 = concat(axis = var_64, interleave = input_5_interleave_0, values = (noisy_latent_to_fp16, noisy_latent_to_fp16))[name = string("input_5_cast_fp16")];
            int32 var_67 = const()[name = string("op_67"), val = int32(0)];
            bool latent_mask_b_interleave_0 = const()[name = string("latent_mask_b_interleave_0"), val = bool(false)];
            string latent_mask_to_fp16_dtype_0 = const()[name = string("latent_mask_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 1, ?]> latent_mask_to_fp16 = cast(dtype = latent_mask_to_fp16_dtype_0, x = latent_mask)[name = string("cast_99")];
            tensor<fp16, [2, 1, ?]> latent_mask_b_cast_fp16 = concat(axis = var_67, interleave = latent_mask_b_interleave_0, values = (latent_mask_to_fp16, latent_mask_to_fp16))[name = string("latent_mask_b_cast_fp16")];
            int32 var_70 = const()[name = string("op_70"), val = int32(0)];
            bool text_mask_interleave_0 = const()[name = string("text_mask_interleave_0"), val = bool(false)];
            string text_mask_to_fp16_dtype_0 = const()[name = string("text_mask_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 1, ?]> text_mask_to_fp16 = cast(dtype = text_mask_to_fp16_dtype_0, x = text_mask)[name = string("cast_98")];
            tensor<fp16, [2, 1, ?]> text_mask_cast_fp16 = concat(axis = var_70, interleave = text_mask_interleave_0, values = (text_mask_to_fp16, text_mask_to_fp16))[name = string("text_mask_cast_fp16")];
            string current_step_to_fp16_dtype_0 = const()[name = string("current_step_to_fp16_dtype_0"), val = string("fp16")];
            string total_step_to_fp16_dtype_0 = const()[name = string("total_step_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1]> total_step_to_fp16 = cast(dtype = total_step_to_fp16_dtype_0, x = total_step)[name = string("cast_96")];
            tensor<fp16, [1]> current_step_to_fp16 = cast(dtype = current_step_to_fp16_dtype_0, x = current_step)[name = string("cast_97")];
            tensor<fp16, [1]> t_1_cast_fp16 = real_div(x = current_step_to_fp16, y = total_step_to_fp16)[name = string("t_1_cast_fp16")];
            int32 var_74 = const()[name = string("op_74"), val = int32(0)];
            bool t_3_interleave_0 = const()[name = string("t_3_interleave_0"), val = bool(false)];
            tensor<fp16, [2]> t_3_cast_fp16 = concat(axis = var_74, interleave = t_3_interleave_0, values = (t_1_cast_fp16, t_1_cast_fp16))[name = string("t_3_cast_fp16")];
            int32 var_78 = const()[name = string("op_78"), val = int32(-1)];
            tensor<int32, [2]> var_85 = const()[name = string("op_85"), val = tensor<int32, [2]>([-1, 1])];
            tensor<fp16, [2, 1]> var_86_cast_fp16 = reshape(shape = var_85, x = t_3_cast_fp16)[name = string("op_86_cast_fp16")];
            fp16 time_encoder_time_scale_to_fp16 = const()[name = string("time_encoder_time_scale_to_fp16"), val = fp16(0x1.f4p+9)];
            tensor<fp16, [2, 1]> t_scaled_cast_fp16 = mul(x = var_86_cast_fp16, y = time_encoder_time_scale_to_fp16)[name = string("t_scaled_cast_fp16")];
            tensor<fp16, [1, 32]> time_encoder_omegas_to_fp16 = const()[name = string("time_encoder_omegas_to_fp16"), val = tensor<fp16, [1, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26304)))];
            tensor<fp16, [2, 32]> phase_cast_fp16 = mul(x = t_scaled_cast_fp16, y = time_encoder_omegas_to_fp16)[name = string("phase_cast_fp16")];
            tensor<fp16, [2, 32]> var_89_cast_fp16 = sin(x = phase_cast_fp16)[name = string("op_89_cast_fp16")];
            tensor<fp16, [2, 32]> var_90_cast_fp16 = cos(x = phase_cast_fp16)[name = string("op_90_cast_fp16")];
            bool input_1_interleave_0 = const()[name = string("input_1_interleave_0"), val = bool(false)];
            tensor<fp16, [2, 64]> input_1_cast_fp16 = concat(axis = var_78, interleave = input_1_interleave_0, values = (var_89_cast_fp16, var_90_cast_fp16))[name = string("input_1_cast_fp16")];
            tensor<fp16, [256, 64]> time_encoder_mlp_0_linear_weight_to_fp16 = const()[name = string("time_encoder_mlp_0_linear_weight_to_fp16"), val = tensor<fp16, [256, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26432)))];
            tensor<fp16, [256]> time_encoder_mlp_0_linear_bias_to_fp16 = const()[name = string("time_encoder_mlp_0_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59264)))];
            tensor<fp16, [2, 256]> linear_0_cast_fp16 = linear(bias = time_encoder_mlp_0_linear_bias_to_fp16, weight = time_encoder_mlp_0_linear_weight_to_fp16, x = input_1_cast_fp16)[name = string("linear_0_cast_fp16")];
            tensor<fp16, [2, 256]> var_97_cast_fp16 = softplus(x = linear_0_cast_fp16)[name = string("op_97_cast_fp16")];
            tensor<fp16, [2, 256]> var_98_cast_fp16 = tanh(x = var_97_cast_fp16)[name = string("op_98_cast_fp16")];
            tensor<fp16, [2, 256]> input_3_cast_fp16 = mul(x = linear_0_cast_fp16, y = var_98_cast_fp16)[name = string("input_3_cast_fp16")];
            tensor<fp16, [64, 256]> time_encoder_mlp_2_linear_weight_to_fp16 = const()[name = string("time_encoder_mlp_2_linear_weight_to_fp16"), val = tensor<fp16, [64, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59840)))];
            tensor<fp16, [64]> time_encoder_mlp_2_linear_bias_to_fp16 = const()[name = string("time_encoder_mlp_2_linear_bias_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92672)))];
            tensor<fp16, [2, 64]> linear_1_cast_fp16 = linear(bias = time_encoder_mlp_2_linear_bias_to_fp16, weight = time_encoder_mlp_2_linear_weight_to_fp16, x = input_3_cast_fp16)[name = string("linear_1_cast_fp16")];
            tensor<int32, [1]> time_emb_axes_0 = const()[name = string("time_emb_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2, 64, 1]> time_emb_cast_fp16 = expand_dims(axes = time_emb_axes_0, x = linear_1_cast_fp16)[name = string("time_emb_cast_fp16")];
            string var_115_pad_type_0 = const()[name = string("op_115_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_115_strides_0 = const()[name = string("op_115_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_115_pad_0 = const()[name = string("op_115_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_115_dilations_0 = const()[name = string("op_115_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_115_groups_0 = const()[name = string("op_115_groups_0"), val = int32(1)];
            tensor<fp16, [512, 144, 1]> proj_in_weight_to_fp16 = const()[name = string("proj_in_weight_to_fp16"), val = tensor<fp16, [512, 144, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92864)))];
            tensor<fp16, [2, 512, ?]> var_115_cast_fp16 = conv(dilations = var_115_dilations_0, groups = var_115_groups_0, pad = var_115_pad_0, pad_type = var_115_pad_type_0, strides = var_115_strides_0, weight = proj_in_weight_to_fp16, x = input_5_cast_fp16)[name = string("op_115_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_3_cast_fp16 = mul(x = var_115_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_3_cast_fp16")];
            int32 var_134 = const()[name = string("op_134"), val = int32(-1)];
            tensor<fp16, [2, 512, ?]> x_5_cast_fp16 = mul(x = x_3_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_5_cast_fp16")];
            tensor<int32, [3]> input_7_begin_0 = const()[name = string("input_7_begin_0"), val = tensor<int32, [3]>([0, 0, 1])];
            tensor<int32, [3]> input_7_end_0 = const()[name = string("input_7_end_0"), val = tensor<int32, [3]>([2, 1, 0])];
            tensor<bool, [3]> input_7_end_mask_0 = const()[name = string("input_7_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [2, 1, ?]> input_7_cast_fp16 = slice_by_index(begin = input_7_begin_0, end = input_7_end_0, end_mask = input_7_end_mask_0, x = latent_mask_b_cast_fp16)[name = string("input_7_cast_fp16")];
            tensor<int32, [6]> var_166_pad_0 = const()[name = string("op_166_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 1])];
            string var_166_mode_0 = const()[name = string("op_166_mode_0"), val = string("constant")];
            fp16 const_0_to_fp16 = const()[name = string("const_0_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 1, ?]> var_166_cast_fp16 = pad(constant_val = const_0_to_fp16, mode = var_166_mode_0, pad = var_166_pad_0, x = input_7_cast_fp16)[name = string("op_166_cast_fp16")];
            fp16 var_133_to_fp16 = const()[name = string("op_133_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [2, 1, ?]> var_167_cast_fp16 = sub(x = var_133_to_fp16, y = var_166_cast_fp16)[name = string("op_167_cast_fp16")];
            tensor<fp16, [2, 1, ?]> last_1_cast_fp16 = mul(x = latent_mask_b_cast_fp16, y = var_167_cast_fp16)[name = string("last_1_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_169_cast_fp16 = mul(x = x_5_cast_fp16, y = last_1_cast_fp16)[name = string("op_169_cast_fp16")];
            tensor<int32, [1]> x_last_1_axes_0 = const()[name = string("x_last_1_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_1_keep_dims_0 = const()[name = string("x_last_1_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_1_cast_fp16 = reduce_sum(axes = x_last_1_axes_0, keep_dims = x_last_1_keep_dims_0, x = var_169_cast_fp16)[name = string("x_last_1_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_172_cast_fp16 = mul(x = x_5_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_172_cast_fp16")];
            tensor<fp16, [2, 1, ?]> var_173_cast_fp16 = sub(x = var_133_to_fp16, y = latent_mask_b_cast_fp16)[name = string("op_173_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_174_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_1_cast_fp16)[name = string("op_174_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_9_cast_fp16 = add(x = var_172_cast_fp16, y = var_174_cast_fp16)[name = string("input_9_cast_fp16")];
            tensor<int32, [6]> input_11_pad_0 = const()[name = string("input_11_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_11_mode_0 = const()[name = string("input_11_mode_0"), val = string("replicate")];
            fp16 const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_11_cast_fp16 = pad(constant_val = const_1_to_fp16, mode = input_11_mode_0, pad = input_11_pad_0, x = input_9_cast_fp16)[name = string("input_11_cast_fp16")];
            string h_3_pad_type_0 = const()[name = string("h_3_pad_type_0"), val = string("valid")];
            int32 h_3_groups_0 = const()[name = string("h_3_groups_0"), val = int32(512)];
            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])];
            tensor<fp16, [512, 1, 5]> main_blocks_0_convnext_0_0_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_0_0_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240384)))];
            tensor<fp16, [512]> main_blocks_0_convnext_0_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245568)))];
            tensor<fp16, [2, 512, ?]> h_3_cast_fp16 = conv(bias = main_blocks_0_convnext_0_0_dwconv__conv_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 = main_blocks_0_convnext_0_0_dwconv__conv_weight_to_fp16, x = input_11_cast_fp16)[name = string("h_3_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_7_cast_fp16 = mul(x = h_3_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_7_cast_fp16")];
            tensor<int32, [3]> input_13_perm_0 = const()[name = string("input_13_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_192_axes_0 = const()[name = string("op_192_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_0_convnext_0_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_0_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246656)))];
            tensor<fp16, [512]> main_blocks_0_convnext_0_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247744)))];
            fp16 var_139_to_fp16 = const()[name = string("op_139_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [2, ?, 512]> input_13_cast_fp16 = transpose(perm = input_13_perm_0, x = x_7_cast_fp16)[name = string("transpose_117")];
            tensor<fp16, [2, ?, 512]> var_192_cast_fp16 = layer_norm(axes = var_192_axes_0, beta = main_blocks_0_convnext_0_0_norm_norm_bias_to_fp16, epsilon = var_139_to_fp16, gamma = main_blocks_0_convnext_0_0_norm_norm_weight_to_fp16, x = input_13_cast_fp16)[name = string("op_192_cast_fp16")];
            tensor<int32, [3]> input_15_perm_0 = const()[name = string("input_15_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            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, [2048, 512, 1]> main_blocks_0_convnext_0_0_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_0_0_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248832)))];
            tensor<fp16, [2048]> main_blocks_0_convnext_0_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2346048)))];
            tensor<fp16, [2, 512, ?]> input_15_cast_fp16 = transpose(perm = input_15_perm_0, x = var_192_cast_fp16)[name = string("transpose_116")];
            tensor<fp16, [2, 2048, ?]> h_5_cast_fp16 = conv(bias = main_blocks_0_convnext_0_0_pwconv1_bias_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 = main_blocks_0_convnext_0_0_pwconv1_weight_to_fp16, x = input_15_cast_fp16)[name = string("h_5_cast_fp16")];
            string input_17_mode_0 = const()[name = string("input_17_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_17_cast_fp16 = gelu(mode = input_17_mode_0, x = h_5_cast_fp16)[name = string("input_17_cast_fp16")];
            string h_7_pad_type_0 = const()[name = string("h_7_pad_type_0"), val = string("valid")];
            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])];
            int32 h_7_groups_0 = const()[name = string("h_7_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_209_weight_0_to_fp16 = const()[name = string("op_209_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2350208)))];
            tensor<fp16, [512]> var_209_bias_0_to_fp16 = const()[name = string("op_209_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4447424)))];
            tensor<fp16, [2, 512, ?]> var_209_cast_fp16 = conv(bias = var_209_bias_0_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 = var_209_weight_0_to_fp16, x = input_17_cast_fp16)[name = string("op_209_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_1_cast_fp16 = add(x = x_5_cast_fp16, y = var_209_cast_fp16)[name = string("out_1_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_9_cast_fp16 = mul(x = out_1_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_9_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_11_cast_fp16 = mul(x = x_9_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_11_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_223_cast_fp16 = mul(x = x_11_cast_fp16, y = last_1_cast_fp16)[name = string("op_223_cast_fp16")];
            tensor<int32, [1]> x_last_3_axes_0 = const()[name = string("x_last_3_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_3_keep_dims_0 = const()[name = string("x_last_3_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_3_cast_fp16 = reduce_sum(axes = x_last_3_axes_0, keep_dims = x_last_3_keep_dims_0, x = var_223_cast_fp16)[name = string("x_last_3_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_226_cast_fp16 = mul(x = x_11_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_226_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_228_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_3_cast_fp16)[name = string("op_228_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_21_cast_fp16 = add(x = var_226_cast_fp16, y = var_228_cast_fp16)[name = string("input_21_cast_fp16")];
            tensor<int32, [6]> input_23_pad_0 = const()[name = string("input_23_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_23_mode_0 = const()[name = string("input_23_mode_0"), val = string("replicate")];
            fp16 const_3_to_fp16 = const()[name = string("const_3_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_23_cast_fp16 = pad(constant_val = const_3_to_fp16, mode = input_23_mode_0, pad = input_23_pad_0, x = input_21_cast_fp16)[name = string("input_23_cast_fp16")];
            string h_9_pad_type_0 = const()[name = string("h_9_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_9_dilations_0 = const()[name = string("h_9_dilations_0"), val = tensor<int32, [1]>([2])];
            int32 h_9_groups_0 = const()[name = string("h_9_groups_0"), val = int32(512)];
            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<fp16, [512, 1, 5]> main_blocks_0_convnext_0_1_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_0_1_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4448512)))];
            tensor<fp16, [512]> main_blocks_0_convnext_0_1_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_1_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4453696)))];
            tensor<fp16, [2, 512, ?]> h_9_cast_fp16 = conv(bias = main_blocks_0_convnext_0_1_dwconv__conv_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 = main_blocks_0_convnext_0_1_dwconv__conv_weight_to_fp16, x = input_23_cast_fp16)[name = string("h_9_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_13_cast_fp16 = mul(x = h_9_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_13_cast_fp16")];
            tensor<int32, [3]> input_25_perm_0 = const()[name = string("input_25_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_246_axes_0 = const()[name = string("op_246_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_0_convnext_0_1_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_0_1_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4454784)))];
            tensor<fp16, [512]> main_blocks_0_convnext_0_1_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_1_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4455872)))];
            tensor<fp16, [2, ?, 512]> input_25_cast_fp16 = transpose(perm = input_25_perm_0, x = x_13_cast_fp16)[name = string("transpose_115")];
            tensor<fp16, [2, ?, 512]> var_246_cast_fp16 = layer_norm(axes = var_246_axes_0, beta = main_blocks_0_convnext_0_1_norm_norm_bias_to_fp16, epsilon = var_139_to_fp16, gamma = main_blocks_0_convnext_0_1_norm_norm_weight_to_fp16, x = input_25_cast_fp16)[name = string("op_246_cast_fp16")];
            tensor<int32, [3]> input_27_perm_0 = const()[name = string("input_27_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            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, [2048, 512, 1]> main_blocks_0_convnext_0_1_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_0_1_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4456960)))];
            tensor<fp16, [2048]> main_blocks_0_convnext_0_1_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_1_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6554176)))];
            tensor<fp16, [2, 512, ?]> input_27_cast_fp16 = transpose(perm = input_27_perm_0, x = var_246_cast_fp16)[name = string("transpose_114")];
            tensor<fp16, [2, 2048, ?]> h_11_cast_fp16 = conv(bias = main_blocks_0_convnext_0_1_pwconv1_bias_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 = main_blocks_0_convnext_0_1_pwconv1_weight_to_fp16, x = input_27_cast_fp16)[name = string("h_11_cast_fp16")];
            string input_29_mode_0 = const()[name = string("input_29_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_29_cast_fp16 = gelu(mode = input_29_mode_0, x = h_11_cast_fp16)[name = string("input_29_cast_fp16")];
            string h_13_pad_type_0 = const()[name = string("h_13_pad_type_0"), val = string("valid")];
            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<int32, [1]> h_13_dilations_0 = const()[name = string("h_13_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_13_groups_0 = const()[name = string("h_13_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_263_weight_0_to_fp16 = const()[name = string("op_263_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6558336)))];
            tensor<fp16, [512]> var_263_bias_0_to_fp16 = const()[name = string("op_263_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8655552)))];
            tensor<fp16, [2, 512, ?]> var_263_cast_fp16 = conv(bias = var_263_bias_0_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 = var_263_weight_0_to_fp16, x = input_29_cast_fp16)[name = string("op_263_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_3_cast_fp16 = add(x = x_11_cast_fp16, y = var_263_cast_fp16)[name = string("out_3_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_15_cast_fp16 = mul(x = out_3_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_15_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_17_cast_fp16 = mul(x = x_15_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_17_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_277_cast_fp16 = mul(x = x_17_cast_fp16, y = last_1_cast_fp16)[name = string("op_277_cast_fp16")];
            tensor<int32, [1]> x_last_5_axes_0 = const()[name = string("x_last_5_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_5_keep_dims_0 = const()[name = string("x_last_5_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_5_cast_fp16 = reduce_sum(axes = x_last_5_axes_0, keep_dims = x_last_5_keep_dims_0, x = var_277_cast_fp16)[name = string("x_last_5_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_280_cast_fp16 = mul(x = x_17_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_280_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_282_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_5_cast_fp16)[name = string("op_282_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_33_cast_fp16 = add(x = var_280_cast_fp16, y = var_282_cast_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, 8, 8])];
            string input_35_mode_0 = const()[name = string("input_35_mode_0"), val = string("replicate")];
            fp16 const_5_to_fp16 = const()[name = string("const_5_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_35_cast_fp16 = pad(constant_val = const_5_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_15_pad_type_0 = const()[name = string("h_15_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_15_dilations_0 = const()[name = string("h_15_dilations_0"), val = tensor<int32, [1]>([4])];
            int32 h_15_groups_0 = const()[name = string("h_15_groups_0"), val = int32(512)];
            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<fp16, [512, 1, 5]> main_blocks_0_convnext_0_2_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_0_2_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8656640)))];
            tensor<fp16, [512]> main_blocks_0_convnext_0_2_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_2_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8661824)))];
            tensor<fp16, [2, 512, ?]> h_15_cast_fp16 = conv(bias = main_blocks_0_convnext_0_2_dwconv__conv_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 = main_blocks_0_convnext_0_2_dwconv__conv_weight_to_fp16, x = input_35_cast_fp16)[name = string("h_15_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_19_cast_fp16 = mul(x = h_15_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_19_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_300_axes_0 = const()[name = string("op_300_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_0_convnext_0_2_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_0_2_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8662912)))];
            tensor<fp16, [512]> main_blocks_0_convnext_0_2_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_2_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8664000)))];
            tensor<fp16, [2, ?, 512]> input_37_cast_fp16 = transpose(perm = input_37_perm_0, x = x_19_cast_fp16)[name = string("transpose_113")];
            tensor<fp16, [2, ?, 512]> var_300_cast_fp16 = layer_norm(axes = var_300_axes_0, beta = main_blocks_0_convnext_0_2_norm_norm_bias_to_fp16, epsilon = var_139_to_fp16, gamma = main_blocks_0_convnext_0_2_norm_norm_weight_to_fp16, x = input_37_cast_fp16)[name = string("op_300_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_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, [2048, 512, 1]> main_blocks_0_convnext_0_2_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_0_2_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8665088)))];
            tensor<fp16, [2048]> main_blocks_0_convnext_0_2_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_2_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10762304)))];
            tensor<fp16, [2, 512, ?]> input_39_cast_fp16 = transpose(perm = input_39_perm_0, x = var_300_cast_fp16)[name = string("transpose_112")];
            tensor<fp16, [2, 2048, ?]> h_17_cast_fp16 = conv(bias = main_blocks_0_convnext_0_2_pwconv1_bias_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 = main_blocks_0_convnext_0_2_pwconv1_weight_to_fp16, x = input_39_cast_fp16)[name = string("h_17_cast_fp16")];
            string input_41_mode_0 = const()[name = string("input_41_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_41_cast_fp16 = gelu(mode = input_41_mode_0, x = h_17_cast_fp16)[name = string("input_41_cast_fp16")];
            string h_19_pad_type_0 = const()[name = string("h_19_pad_type_0"), val = string("valid")];
            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<int32, [1]> h_19_dilations_0 = const()[name = string("h_19_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_19_groups_0 = const()[name = string("h_19_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_317_weight_0_to_fp16 = const()[name = string("op_317_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10766464)))];
            tensor<fp16, [512]> var_317_bias_0_to_fp16 = const()[name = string("op_317_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12863680)))];
            tensor<fp16, [2, 512, ?]> var_317_cast_fp16 = conv(bias = var_317_bias_0_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 = var_317_weight_0_to_fp16, x = input_41_cast_fp16)[name = string("op_317_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_5_cast_fp16 = add(x = x_17_cast_fp16, y = var_317_cast_fp16)[name = string("out_5_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_21_cast_fp16 = mul(x = out_5_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_21_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_23_cast_fp16 = mul(x = x_21_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_23_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_331_cast_fp16 = mul(x = x_23_cast_fp16, y = last_1_cast_fp16)[name = string("op_331_cast_fp16")];
            tensor<int32, [1]> x_last_7_axes_0 = const()[name = string("x_last_7_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_7_keep_dims_0 = const()[name = string("x_last_7_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_7_cast_fp16 = reduce_sum(axes = x_last_7_axes_0, keep_dims = x_last_7_keep_dims_0, x = var_331_cast_fp16)[name = string("x_last_7_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_334_cast_fp16 = mul(x = x_23_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_334_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_336_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_7_cast_fp16)[name = string("op_336_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_45_cast_fp16 = add(x = var_334_cast_fp16, y = var_336_cast_fp16)[name = string("input_45_cast_fp16")];
            tensor<int32, [6]> input_47_pad_0 = const()[name = string("input_47_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 16, 16])];
            string input_47_mode_0 = const()[name = string("input_47_mode_0"), val = string("replicate")];
            fp16 const_7_to_fp16 = const()[name = string("const_7_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_47_cast_fp16 = pad(constant_val = const_7_to_fp16, mode = input_47_mode_0, pad = input_47_pad_0, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")];
            string h_21_pad_type_0 = const()[name = string("h_21_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_21_dilations_0 = const()[name = string("h_21_dilations_0"), val = tensor<int32, [1]>([8])];
            int32 h_21_groups_0 = const()[name = string("h_21_groups_0"), val = int32(512)];
            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<fp16, [512, 1, 5]> main_blocks_0_convnext_0_3_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_0_3_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12864768)))];
            tensor<fp16, [512]> main_blocks_0_convnext_0_3_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_3_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12869952)))];
            tensor<fp16, [2, 512, ?]> h_21_cast_fp16 = conv(bias = main_blocks_0_convnext_0_3_dwconv__conv_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 = main_blocks_0_convnext_0_3_dwconv__conv_weight_to_fp16, x = input_47_cast_fp16)[name = string("h_21_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_25_cast_fp16 = mul(x = h_21_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_25_cast_fp16")];
            tensor<int32, [3]> input_49_perm_0 = const()[name = string("input_49_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_354_axes_0 = const()[name = string("op_354_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_0_convnext_0_3_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_0_3_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12871040)))];
            tensor<fp16, [512]> main_blocks_0_convnext_0_3_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_3_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12872128)))];
            tensor<fp16, [2, ?, 512]> input_49_cast_fp16 = transpose(perm = input_49_perm_0, x = x_25_cast_fp16)[name = string("transpose_111")];
            tensor<fp16, [2, ?, 512]> var_354_cast_fp16 = layer_norm(axes = var_354_axes_0, beta = main_blocks_0_convnext_0_3_norm_norm_bias_to_fp16, epsilon = var_139_to_fp16, gamma = main_blocks_0_convnext_0_3_norm_norm_weight_to_fp16, x = input_49_cast_fp16)[name = string("op_354_cast_fp16")];
            tensor<int32, [3]> input_51_perm_0 = const()[name = string("input_51_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            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, [2048, 512, 1]> main_blocks_0_convnext_0_3_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_0_3_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12873216)))];
            tensor<fp16, [2048]> main_blocks_0_convnext_0_3_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_3_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14970432)))];
            tensor<fp16, [2, 512, ?]> input_51_cast_fp16 = transpose(perm = input_51_perm_0, x = var_354_cast_fp16)[name = string("transpose_110")];
            tensor<fp16, [2, 2048, ?]> h_23_cast_fp16 = conv(bias = main_blocks_0_convnext_0_3_pwconv1_bias_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 = main_blocks_0_convnext_0_3_pwconv1_weight_to_fp16, x = input_51_cast_fp16)[name = string("h_23_cast_fp16")];
            string input_53_mode_0 = const()[name = string("input_53_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_53_cast_fp16 = gelu(mode = input_53_mode_0, x = h_23_cast_fp16)[name = string("input_53_cast_fp16")];
            string h_25_pad_type_0 = const()[name = string("h_25_pad_type_0"), val = string("valid")];
            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<int32, [1]> h_25_dilations_0 = const()[name = string("h_25_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_25_groups_0 = const()[name = string("h_25_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_371_weight_0_to_fp16 = const()[name = string("op_371_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14974592)))];
            tensor<fp16, [512]> var_371_bias_0_to_fp16 = const()[name = string("op_371_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17071808)))];
            tensor<fp16, [2, 512, ?]> var_371_cast_fp16 = conv(bias = var_371_bias_0_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 = var_371_weight_0_to_fp16, x = input_53_cast_fp16)[name = string("op_371_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_7_cast_fp16 = add(x = x_23_cast_fp16, y = var_371_cast_fp16)[name = string("out_7_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_27_cast_fp16 = mul(x = out_7_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_27_cast_fp16")];
            tensor<int32, [1]> input_55_axes_0 = const()[name = string("input_55_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2, 64]> input_55_cast_fp16 = squeeze(axes = input_55_axes_0, x = time_emb_cast_fp16)[name = string("input_55_cast_fp16")];
            tensor<fp16, [512, 64]> main_blocks_0_time_cond_linear_linear_weight_to_fp16 = const()[name = string("main_blocks_0_time_cond_linear_linear_weight_to_fp16"), val = tensor<fp16, [512, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17072896)))];
            tensor<fp16, [512]> main_blocks_0_time_cond_linear_linear_bias_to_fp16 = const()[name = string("main_blocks_0_time_cond_linear_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17138496)))];
            tensor<fp16, [2, 512]> linear_2_cast_fp16 = linear(bias = main_blocks_0_time_cond_linear_linear_bias_to_fp16, weight = main_blocks_0_time_cond_linear_linear_weight_to_fp16, x = input_55_cast_fp16)[name = string("linear_2_cast_fp16")];
            tensor<int32, [1]> t_5_axes_0 = const()[name = string("t_5_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2, 512, 1]> t_5_cast_fp16 = expand_dims(axes = t_5_axes_0, x = linear_2_cast_fp16)[name = string("t_5_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_381_cast_fp16 = add(x = x_27_cast_fp16, y = t_5_cast_fp16)[name = string("op_381_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_29_cast_fp16 = mul(x = var_381_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_29_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_31_cast_fp16 = mul(x = x_29_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_31_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_394_cast_fp16 = mul(x = x_31_cast_fp16, y = last_1_cast_fp16)[name = string("op_394_cast_fp16")];
            tensor<int32, [1]> x_last_9_axes_0 = const()[name = string("x_last_9_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_9_keep_dims_0 = const()[name = string("x_last_9_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_9_cast_fp16 = reduce_sum(axes = x_last_9_axes_0, keep_dims = x_last_9_keep_dims_0, x = var_394_cast_fp16)[name = string("x_last_9_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_397_cast_fp16 = mul(x = x_31_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_397_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_399_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_9_cast_fp16)[name = string("op_399_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_59_cast_fp16 = add(x = var_397_cast_fp16, y = var_399_cast_fp16)[name = string("input_59_cast_fp16")];
            tensor<int32, [6]> input_61_pad_0 = const()[name = string("input_61_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_61_mode_0 = const()[name = string("input_61_mode_0"), val = string("replicate")];
            fp16 const_9_to_fp16 = const()[name = string("const_9_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_61_cast_fp16 = pad(constant_val = const_9_to_fp16, mode = input_61_mode_0, pad = input_61_pad_0, x = input_59_cast_fp16)[name = string("input_61_cast_fp16")];
            string h_27_pad_type_0 = const()[name = string("h_27_pad_type_0"), val = string("valid")];
            int32 h_27_groups_0 = const()[name = string("h_27_groups_0"), val = int32(512)];
            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])];
            tensor<fp16, [512, 1, 5]> main_blocks_0_convnext_1_0_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_1_0_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17139584)))];
            tensor<fp16, [512]> main_blocks_0_convnext_1_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_1_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17144768)))];
            tensor<fp16, [2, 512, ?]> h_27_cast_fp16 = conv(bias = main_blocks_0_convnext_1_0_dwconv__conv_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 = main_blocks_0_convnext_1_0_dwconv__conv_weight_to_fp16, x = input_61_cast_fp16)[name = string("h_27_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_33_cast_fp16 = mul(x = h_27_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_33_cast_fp16")];
            tensor<int32, [3]> input_63_perm_0 = const()[name = string("input_63_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_417_axes_0 = const()[name = string("op_417_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_0_convnext_1_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_1_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17145856)))];
            tensor<fp16, [512]> main_blocks_0_convnext_1_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_1_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17146944)))];
            tensor<fp16, [2, ?, 512]> input_63_cast_fp16 = transpose(perm = input_63_perm_0, x = x_33_cast_fp16)[name = string("transpose_109")];
            tensor<fp16, [2, ?, 512]> var_417_cast_fp16 = layer_norm(axes = var_417_axes_0, beta = main_blocks_0_convnext_1_0_norm_norm_bias_to_fp16, epsilon = var_139_to_fp16, gamma = main_blocks_0_convnext_1_0_norm_norm_weight_to_fp16, x = input_63_cast_fp16)[name = string("op_417_cast_fp16")];
            tensor<int32, [3]> input_65_perm_0 = const()[name = string("input_65_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            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, [2048, 512, 1]> main_blocks_0_convnext_1_0_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_1_0_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17148032)))];
            tensor<fp16, [2048]> main_blocks_0_convnext_1_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_1_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19245248)))];
            tensor<fp16, [2, 512, ?]> input_65_cast_fp16 = transpose(perm = input_65_perm_0, x = var_417_cast_fp16)[name = string("transpose_108")];
            tensor<fp16, [2, 2048, ?]> h_29_cast_fp16 = conv(bias = main_blocks_0_convnext_1_0_pwconv1_bias_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 = main_blocks_0_convnext_1_0_pwconv1_weight_to_fp16, x = input_65_cast_fp16)[name = string("h_29_cast_fp16")];
            string input_67_mode_0 = const()[name = string("input_67_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_67_cast_fp16 = gelu(mode = input_67_mode_0, x = h_29_cast_fp16)[name = string("input_67_cast_fp16")];
            string h_31_pad_type_0 = const()[name = string("h_31_pad_type_0"), val = string("valid")];
            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<int32, [1]> h_31_dilations_0 = const()[name = string("h_31_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_31_groups_0 = const()[name = string("h_31_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_434_weight_0_to_fp16 = const()[name = string("op_434_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19249408)))];
            tensor<fp16, [512]> var_434_bias_0_to_fp16 = const()[name = string("op_434_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21346624)))];
            tensor<fp16, [2, 512, ?]> var_434_cast_fp16 = conv(bias = var_434_bias_0_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 = var_434_weight_0_to_fp16, x = input_67_cast_fp16)[name = string("op_434_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_9_cast_fp16 = add(x = x_31_cast_fp16, y = var_434_cast_fp16)[name = string("out_9_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_35_cast_fp16 = mul(x = out_9_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_35_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_m_1_cast_fp16 = mul(x = x_35_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_m_1_cast_fp16")];
            tensor<int32, [3]> var_444_shape_cast_fp16 = shape(x = x_35_cast_fp16)[name = string("op_444_shape_cast_fp16")];
            int32 gather_4_axis_0 = const()[name = string("gather_4_axis_0"), val = int32(0)];
            int32 gather_4_batch_dims_0 = const()[name = string("gather_4_batch_dims_0"), val = int32(0)];
            bool gather_4_validate_indices_0 = const()[name = string("gather_4_validate_indices_0"), val = bool(false)];
            string var_444_shape_cast_fp16_to_int16_dtype_0 = const()[name = string("op_444_shape_cast_fp16_to_int16_dtype_0"), val = string("int16")];
            uint16 gather_4_indices_0_to_uint16 = const()[name = string("gather_4_indices_0_to_uint16"), val = uint16(2)];
            tensor<int16, [3]> var_444_shape_cast_fp16_to_int16 = cast(dtype = var_444_shape_cast_fp16_to_int16_dtype_0, x = var_444_shape_cast_fp16)[name = string("cast_95")];
            int16 gather_4_cast_uint16 = gather(axis = gather_4_axis_0, batch_dims = gather_4_batch_dims_0, indices = gather_4_indices_0_to_uint16, validate_indices = gather_4_validate_indices_0, x = var_444_shape_cast_fp16_to_int16)[name = string("gather_4_cast_uint16")];
            string gather_4_cast_uint16_to_int32_dtype_0 = const()[name = string("gather_4_cast_uint16_to_int32_dtype_0"), val = string("int32")];
            tensor<int32, [3]> var_446_shape_cast_fp16 = shape(x = text_emb_cast_fp16)[name = string("op_446_shape_cast_fp16")];
            int32 gather_5_axis_0 = const()[name = string("gather_5_axis_0"), val = int32(0)];
            int32 gather_5_batch_dims_0 = const()[name = string("gather_5_batch_dims_0"), val = int32(0)];
            bool gather_5_validate_indices_0 = const()[name = string("gather_5_validate_indices_0"), val = bool(false)];
            string var_446_shape_cast_fp16_to_uint16_dtype_0 = const()[name = string("op_446_shape_cast_fp16_to_uint16_dtype_0"), val = string("uint16")];
            uint16 gather_5_indices_0_to_uint16 = const()[name = string("gather_5_indices_0_to_uint16"), val = uint16(2)];
            tensor<uint16, [3]> var_446_shape_cast_fp16_to_uint16 = cast(dtype = var_446_shape_cast_fp16_to_uint16_dtype_0, x = var_446_shape_cast_fp16)[name = string("cast_93")];
            uint16 gather_5_cast_uint16 = gather(axis = gather_5_axis_0, batch_dims = gather_5_batch_dims_0, indices = gather_5_indices_0_to_uint16, validate_indices = gather_5_validate_indices_0, x = var_446_shape_cast_fp16_to_uint16)[name = string("gather_5_cast_uint16")];
            string gather_5_cast_uint16_to_int32_dtype_0 = const()[name = string("gather_5_cast_uint16_to_int32_dtype_0"), val = string("int32")];
            tensor<int32, [3]> input_69_perm_0 = const()[name = string("input_69_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> main_blocks_0_text_attn_W_query_linear_weight_to_fp16 = const()[name = string("main_blocks_0_text_attn_W_query_linear_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21347712)))];
            tensor<fp16, [512]> main_blocks_0_text_attn_W_query_linear_bias_to_fp16 = const()[name = string("main_blocks_0_text_attn_W_query_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21872064)))];
            tensor<fp16, [2, ?, 512]> input_69_cast_fp16 = transpose(perm = input_69_perm_0, x = x_35_cast_fp16)[name = string("transpose_107")];
            tensor<fp16, [2, ?, 512]> linear_3_cast_fp16 = linear(bias = main_blocks_0_text_attn_W_query_linear_bias_to_fp16, weight = main_blocks_0_text_attn_W_query_linear_weight_to_fp16, x = input_69_cast_fp16)[name = string("linear_3_cast_fp16")];
            tensor<int32, [4]> concat_4x = const()[name = string("concat_4x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_453_cast_fp16 = reshape(shape = concat_4x, x = linear_3_cast_fp16)[name = string("op_453_cast_fp16")];
            tensor<int32, [4]> x_37_perm_0 = const()[name = string("x_37_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> input_71_perm_0 = const()[name = string("input_71_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 256]> main_blocks_0_text_attn_W_key_linear_weight_to_fp16 = const()[name = string("main_blocks_0_text_attn_W_key_linear_weight_to_fp16"), val = tensor<fp16, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21873152)))];
            tensor<fp16, [512]> main_blocks_0_text_attn_W_key_linear_bias_to_fp16 = const()[name = string("main_blocks_0_text_attn_W_key_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22135360)))];
            tensor<fp16, [?, ?, 256]> input_71_cast_fp16 = transpose(perm = input_71_perm_0, x = text_emb_cast_fp16)[name = string("transpose_105")];
            tensor<fp16, [?, ?, 512]> linear_4_cast_fp16 = linear(bias = main_blocks_0_text_attn_W_key_linear_bias_to_fp16, weight = main_blocks_0_text_attn_W_key_linear_weight_to_fp16, x = input_71_cast_fp16)[name = string("linear_4_cast_fp16")];
            tensor<int32, [4]> concat_5x = const()[name = string("concat_5x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_461_cast_fp16 = reshape(shape = concat_5x, x = linear_4_cast_fp16)[name = string("op_461_cast_fp16")];
            tensor<int32, [4]> x_39_perm_0 = const()[name = string("x_39_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [512, 256]> main_blocks_0_text_attn_W_value_linear_weight_to_fp16 = const()[name = string("main_blocks_0_text_attn_W_value_linear_weight_to_fp16"), val = tensor<fp16, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22136448)))];
            tensor<fp16, [512]> main_blocks_0_text_attn_W_value_linear_bias_to_fp16 = const()[name = string("main_blocks_0_text_attn_W_value_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22398656)))];
            tensor<fp16, [?, ?, 512]> linear_5_cast_fp16 = linear(bias = main_blocks_0_text_attn_W_value_linear_bias_to_fp16, weight = main_blocks_0_text_attn_W_value_linear_weight_to_fp16, x = input_71_cast_fp16)[name = string("linear_5_cast_fp16")];
            tensor<int32, [4]> concat_6x = const()[name = string("concat_6x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_469_cast_fp16 = reshape(shape = concat_6x, x = linear_5_cast_fp16)[name = string("op_469_cast_fp16")];
            tensor<int32, [4]> v_1_perm_0 = const()[name = string("v_1_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            int32 concat_7_values0_0 = const()[name = string("concat_7_values0_0"), val = int32(1)];
            int32 concat_7_values2_0 = const()[name = string("concat_7_values2_0"), val = int32(1)];
            int32 concat_7_axis_0 = const()[name = string("concat_7_axis_0"), val = int32(0)];
            bool concat_7_interleave_0 = const()[name = string("concat_7_interleave_0"), val = bool(false)];
            int32 gather_4_cast_uint16_to_int32 = cast(dtype = gather_4_cast_uint16_to_int32_dtype_0, x = gather_4_cast_uint16)[name = string("cast_94")];
            tensor<int32, [3]> concat_7 = concat(axis = concat_7_axis_0, interleave = concat_7_interleave_0, values = (concat_7_values0_0, gather_4_cast_uint16_to_int32, concat_7_values2_0))[name = string("concat_7")];
            tensor<int32, [3]> var_472_begin_0 = const()[name = string("op_472_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<bool, [3]> var_472_end_mask_0 = const()[name = string("op_472_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 1000, 1]> main_blocks_0_text_attn_increments_to_fp16 = const()[name = string("main_blocks_0_text_attn_increments_to_fp16"), val = tensor<fp16, [1, 1000, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22399744)))];
            tensor<fp16, [1, ?, 1]> var_472_cast_fp16 = slice_by_index(begin = var_472_begin_0, end = concat_7, end_mask = var_472_end_mask_0, x = main_blocks_0_text_attn_increments_to_fp16)[name = string("op_472_cast_fp16")];
            tensor<int32, [3]> concat_8 = const()[name = string("concat_8"), val = tensor<int32, [3]>([2, -1, -1])];
            tensor<int32, [3]> shape_1_cast_fp16 = shape(x = var_472_cast_fp16)[name = string("shape_1_cast_fp16")];
            tensor<bool, [3]> equal_1 = const()[name = string("equal_1"), val = tensor<bool, [3]>([false, true, true])];
            tensor<int32, [3]> select_1 = select(a = shape_1_cast_fp16, b = concat_8, cond = equal_1)[name = string("select_1")];
            tensor<int32, [3]> real_div_1 = real_div(x = select_1, y = shape_1_cast_fp16)[name = string("real_div_1")];
            tensor<fp16, [?, ?, ?]> var_475_cast_fp16 = tile(reps = real_div_1, x = var_472_cast_fp16)[name = string("op_475_cast_fp16")];
            int32 concat_9_values0_0 = const()[name = string("concat_9_values0_0"), val = int32(1)];
            int32 concat_9_values2_0 = const()[name = string("concat_9_values2_0"), val = int32(1)];
            int32 concat_9_axis_0 = const()[name = string("concat_9_axis_0"), val = int32(0)];
            bool concat_9_interleave_0 = const()[name = string("concat_9_interleave_0"), val = bool(false)];
            int32 gather_5_cast_uint16_to_int32 = cast(dtype = gather_5_cast_uint16_to_int32_dtype_0, x = gather_5_cast_uint16)[name = string("cast_92")];
            tensor<int32, [3]> concat_9 = concat(axis = concat_9_axis_0, interleave = concat_9_interleave_0, values = (concat_9_values0_0, gather_5_cast_uint16_to_int32, concat_9_values2_0))[name = string("concat_9")];
            tensor<int32, [3]> var_478_begin_0 = const()[name = string("op_478_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<bool, [3]> var_478_end_mask_0 = const()[name = string("op_478_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, ?, 1]> var_478_cast_fp16 = slice_by_index(begin = var_478_begin_0, end = concat_9, end_mask = var_478_end_mask_0, x = main_blocks_0_text_attn_increments_to_fp16)[name = string("op_478_cast_fp16")];
            tensor<int32, [3]> concat_10 = const()[name = string("concat_10"), val = tensor<int32, [3]>([2, -1, -1])];
            tensor<int32, [3]> shape_2_cast_fp16 = shape(x = var_478_cast_fp16)[name = string("shape_2_cast_fp16")];
            tensor<bool, [3]> equal_2 = const()[name = string("equal_2"), val = tensor<bool, [3]>([false, true, true])];
            tensor<int32, [3]> select_2 = select(a = shape_2_cast_fp16, b = concat_10, cond = equal_2)[name = string("select_2")];
            tensor<int32, [3]> real_div_2 = real_div(x = select_2, y = shape_2_cast_fp16)[name = string("real_div_2")];
            tensor<fp16, [?, ?, ?]> var_481_cast_fp16 = tile(reps = real_div_2, x = var_478_cast_fp16)[name = string("op_481_cast_fp16")];
            tensor<int32, [2]> divisor_1_axes_0 = const()[name = string("divisor_1_axes_0"), val = tensor<int32, [2]>([1, 2])];
            bool divisor_1_keep_dims_0 = const()[name = string("divisor_1_keep_dims_0"), val = bool(false)];
            tensor<fp16, [2]> divisor_1_cast_fp16 = reduce_sum(axes = divisor_1_axes_0, keep_dims = divisor_1_keep_dims_0, x = latent_mask_b_cast_fp16)[name = string("divisor_1_cast_fp16")];
            tensor<int32, [2]> divisor_3_axes_0 = const()[name = string("divisor_3_axes_0"), val = tensor<int32, [2]>([1, 2])];
            bool divisor_3_keep_dims_0 = const()[name = string("divisor_3_keep_dims_0"), val = bool(false)];
            tensor<fp16, [2]> divisor_3_cast_fp16 = reduce_sum(axes = divisor_3_axes_0, keep_dims = divisor_3_keep_dims_0, x = text_mask_cast_fp16)[name = string("divisor_3_cast_fp16")];
            tensor<int32, [3]> var_487 = const()[name = string("op_487"), val = tensor<int32, [3]>([-1, 1, 1])];
            tensor<fp16, [2, 1, 1]> var_488_cast_fp16 = reshape(shape = var_487, x = divisor_1_cast_fp16)[name = string("op_488_cast_fp16")];
            tensor<fp16, [2, ?, ?]> scaled_1_cast_fp16 = real_div(x = var_475_cast_fp16, y = var_488_cast_fp16)[name = string("scaled_1_cast_fp16")];
            tensor<fp16, [1, 1, 32]> main_blocks_0_text_attn_theta_to_fp16 = const()[name = string("main_blocks_0_text_attn_theta_to_fp16"), val = tensor<fp16, [1, 1, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22401856)))];
            tensor<fp16, [2, ?, 32]> angles_1_cast_fp16 = mul(x = scaled_1_cast_fp16, y = main_blocks_0_text_attn_theta_to_fp16)[name = string("angles_1_cast_fp16")];
            tensor<fp16, [2, ?, 32]> var_491_cast_fp16 = cos(x = angles_1_cast_fp16)[name = string("op_491_cast_fp16")];
            tensor<int32, [1]> cos_1_axes_0 = const()[name = string("cos_1_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, ?, 32]> cos_1_cast_fp16 = expand_dims(axes = cos_1_axes_0, x = var_491_cast_fp16)[name = string("cos_1_cast_fp16")];
            tensor<fp16, [2, ?, 32]> var_493_cast_fp16 = sin(x = angles_1_cast_fp16)[name = string("op_493_cast_fp16")];
            tensor<int32, [1]> sin_1_axes_0 = const()[name = string("sin_1_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, ?, 32]> sin_1_cast_fp16 = expand_dims(axes = sin_1_axes_0, x = var_493_cast_fp16)[name = string("sin_1_cast_fp16")];
            tensor<int32, [4]> x_a_1_begin_0 = const()[name = string("x_a_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x_a_1_end_0 = const()[name = string("x_a_1_end_0"), val = tensor<int32, [4]>([2, 8, 0, 32])];
            tensor<bool, [4]> x_a_1_end_mask_0 = const()[name = string("x_a_1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 8, ?, 64]> x_37_cast_fp16 = transpose(perm = x_37_perm_0, x = var_453_cast_fp16)[name = string("transpose_106")];
            tensor<fp16, [2, 8, ?, 32]> x_a_1_cast_fp16 = slice_by_index(begin = x_a_1_begin_0, end = x_a_1_end_0, end_mask = x_a_1_end_mask_0, x = x_37_cast_fp16)[name = string("x_a_1_cast_fp16")];
            tensor<int32, [4]> x_b_1_begin_0 = const()[name = string("x_b_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x_b_1_end_0 = const()[name = string("x_b_1_end_0"), val = tensor<int32, [4]>([2, 8, 0, 64])];
            tensor<bool, [4]> x_b_1_end_mask_0 = const()[name = string("x_b_1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [2, 8, ?, 32]> x_b_1_cast_fp16 = slice_by_index(begin = x_b_1_begin_0, end = x_b_1_end_0, end_mask = x_b_1_end_mask_0, x = x_37_cast_fp16)[name = string("x_b_1_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_497_cast_fp16 = mul(x = x_a_1_cast_fp16, y = cos_1_cast_fp16)[name = string("op_497_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_498_cast_fp16 = mul(x = x_b_1_cast_fp16, y = sin_1_cast_fp16)[name = string("op_498_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_a_1_cast_fp16 = sub(x = var_497_cast_fp16, y = var_498_cast_fp16)[name = string("rot_a_1_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_500_cast_fp16 = mul(x = x_a_1_cast_fp16, y = sin_1_cast_fp16)[name = string("op_500_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_501_cast_fp16 = mul(x = x_b_1_cast_fp16, y = cos_1_cast_fp16)[name = string("op_501_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_b_1_cast_fp16 = add(x = var_500_cast_fp16, y = var_501_cast_fp16)[name = string("rot_b_1_cast_fp16")];
            bool q_1_interleave_0 = const()[name = string("q_1_interleave_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> q_1_cast_fp16 = concat(axis = var_134, interleave = q_1_interleave_0, values = (rot_a_1_cast_fp16, rot_b_1_cast_fp16))[name = string("q_1_cast_fp16")];
            tensor<int32, [3]> var_505 = const()[name = string("op_505"), val = tensor<int32, [3]>([-1, 1, 1])];
            tensor<fp16, [2, 1, 1]> var_506_cast_fp16 = reshape(shape = var_505, x = divisor_3_cast_fp16)[name = string("op_506_cast_fp16")];
            tensor<fp16, [2, ?, ?]> scaled_3_cast_fp16 = real_div(x = var_481_cast_fp16, y = var_506_cast_fp16)[name = string("scaled_3_cast_fp16")];
            tensor<fp16, [2, ?, 32]> angles_3_cast_fp16 = mul(x = scaled_3_cast_fp16, y = main_blocks_0_text_attn_theta_to_fp16)[name = string("angles_3_cast_fp16")];
            tensor<fp16, [2, ?, 32]> var_509_cast_fp16 = cos(x = angles_3_cast_fp16)[name = string("op_509_cast_fp16")];
            tensor<int32, [1]> cos_3_axes_0 = const()[name = string("cos_3_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, ?, 32]> cos_3_cast_fp16 = expand_dims(axes = cos_3_axes_0, x = var_509_cast_fp16)[name = string("cos_3_cast_fp16")];
            tensor<fp16, [2, ?, 32]> var_511_cast_fp16 = sin(x = angles_3_cast_fp16)[name = string("op_511_cast_fp16")];
            tensor<int32, [1]> sin_3_axes_0 = const()[name = string("sin_3_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, ?, 32]> sin_3_cast_fp16 = expand_dims(axes = sin_3_axes_0, x = var_511_cast_fp16)[name = string("sin_3_cast_fp16")];
            tensor<int32, [4]> x_a_3_begin_0 = const()[name = string("x_a_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x_a_3_end_0 = const()[name = string("x_a_3_end_0"), val = tensor<int32, [4]>([2, 8, 0, 32])];
            tensor<bool, [4]> x_a_3_end_mask_0 = const()[name = string("x_a_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 8, ?, 64]> x_39_cast_fp16 = transpose(perm = x_39_perm_0, x = var_461_cast_fp16)[name = string("transpose_104")];
            tensor<fp16, [2, 8, ?, 32]> x_a_3_cast_fp16 = slice_by_index(begin = x_a_3_begin_0, end = x_a_3_end_0, end_mask = x_a_3_end_mask_0, x = x_39_cast_fp16)[name = string("x_a_3_cast_fp16")];
            tensor<int32, [4]> x_b_3_begin_0 = const()[name = string("x_b_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x_b_3_end_0 = const()[name = string("x_b_3_end_0"), val = tensor<int32, [4]>([2, 8, 0, 64])];
            tensor<bool, [4]> x_b_3_end_mask_0 = const()[name = string("x_b_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [2, 8, ?, 32]> x_b_3_cast_fp16 = slice_by_index(begin = x_b_3_begin_0, end = x_b_3_end_0, end_mask = x_b_3_end_mask_0, x = x_39_cast_fp16)[name = string("x_b_3_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_515_cast_fp16 = mul(x = x_a_3_cast_fp16, y = cos_3_cast_fp16)[name = string("op_515_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_516_cast_fp16 = mul(x = x_b_3_cast_fp16, y = sin_3_cast_fp16)[name = string("op_516_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_a_3_cast_fp16 = sub(x = var_515_cast_fp16, y = var_516_cast_fp16)[name = string("rot_a_3_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_518_cast_fp16 = mul(x = x_a_3_cast_fp16, y = sin_3_cast_fp16)[name = string("op_518_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_519_cast_fp16 = mul(x = x_b_3_cast_fp16, y = cos_3_cast_fp16)[name = string("op_519_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_b_3_cast_fp16 = add(x = var_518_cast_fp16, y = var_519_cast_fp16)[name = string("rot_b_3_cast_fp16")];
            bool k_1_interleave_0 = const()[name = string("k_1_interleave_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> k_1_cast_fp16 = concat(axis = var_134, interleave = k_1_interleave_0, values = (rot_a_3_cast_fp16, rot_b_3_cast_fp16))[name = string("k_1_cast_fp16")];
            bool var_524_transpose_x_1 = const()[name = string("op_524_transpose_x_1"), val = bool(false)];
            bool var_524_transpose_y_1 = const()[name = string("op_524_transpose_y_1"), val = bool(true)];
            tensor<fp16, [2, 8, ?, ?]> var_524_cast_fp16 = matmul(transpose_x = var_524_transpose_x_1, transpose_y = var_524_transpose_y_1, x = q_1_cast_fp16, y = k_1_cast_fp16)[name = string("op_524_cast_fp16")];
            fp16 _inversed_scores_1_y_0_to_fp16 = const()[name = string("_inversed_scores_1_y_0_to_fp16"), val = fp16(0x1p-4)];
            tensor<fp16, [2, 8, ?, ?]> _inversed_scores_1_cast_fp16 = mul(x = var_524_cast_fp16, y = _inversed_scores_1_y_0_to_fp16)[name = string("_inversed_scores_1_cast_fp16")];
            tensor<int32, [1]> var_527_axes_0 = const()[name = string("op_527_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [2, 1, 1, ?]> var_527_cast_fp16 = expand_dims(axes = var_527_axes_0, x = text_mask_cast_fp16)[name = string("op_527_cast_fp16")];
            tensor<fp16, [2, 1, 1, ?]> var_528_cast_fp16 = sub(x = var_133_to_fp16, y = var_527_cast_fp16)[name = string("op_528_cast_fp16")];
            fp16 var_529_to_fp16 = const()[name = string("op_529_to_fp16"), val = fp16(0x1.388p+13)];
            tensor<fp16, [2, 1, 1, ?]> var_530_cast_fp16 = mul(x = var_528_cast_fp16, y = var_529_to_fp16)[name = string("op_530_cast_fp16")];
            tensor<fp16, [2, 8, ?, ?]> input_75_cast_fp16 = sub(x = _inversed_scores_1_cast_fp16, y = var_530_cast_fp16)[name = string("input_75_cast_fp16")];
            tensor<fp16, [2, 8, ?, ?]> attn_1_cast_fp16 = softmax(axis = var_134, x = input_75_cast_fp16)[name = string("attn_1_cast_fp16")];
            bool var_533_transpose_x_0 = const()[name = string("op_533_transpose_x_0"), val = bool(false)];
            bool var_533_transpose_y_0 = const()[name = string("op_533_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> v_1_cast_fp16 = transpose(perm = v_1_perm_0, x = var_469_cast_fp16)[name = string("transpose_103")];
            tensor<fp16, [2, 8, ?, 64]> var_533_cast_fp16 = matmul(transpose_x = var_533_transpose_x_0, transpose_y = var_533_transpose_y_0, x = attn_1_cast_fp16, y = v_1_cast_fp16)[name = string("op_533_cast_fp16")];
            tensor<int32, [4]> var_534_perm_0 = const()[name = string("op_534_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_11x = const()[name = string("concat_11x"), val = tensor<int32, [3]>([2, -1, 512])];
            tensor<fp16, [2, ?, 8, 64]> var_534_cast_fp16 = transpose(perm = var_534_perm_0, x = var_533_cast_fp16)[name = string("transpose_102")];
            tensor<fp16, [2, ?, 512]> input_77_cast_fp16 = reshape(shape = concat_11x, x = var_534_cast_fp16)[name = string("input_77_cast_fp16")];
            tensor<fp16, [512, 512]> main_blocks_0_text_attn_out_fc_linear_weight_to_fp16 = const()[name = string("main_blocks_0_text_attn_out_fc_linear_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22401984)))];
            tensor<fp16, [512]> main_blocks_0_text_attn_out_fc_linear_bias_to_fp16 = const()[name = string("main_blocks_0_text_attn_out_fc_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22926336)))];
            tensor<fp16, [2, ?, 512]> linear_6_cast_fp16 = linear(bias = main_blocks_0_text_attn_out_fc_linear_bias_to_fp16, weight = main_blocks_0_text_attn_out_fc_linear_weight_to_fp16, x = input_77_cast_fp16)[name = string("linear_6_cast_fp16")];
            tensor<int32, [3]> out_11_perm_0 = const()[name = string("out_11_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> out_11_cast_fp16 = transpose(perm = out_11_perm_0, x = linear_6_cast_fp16)[name = string("transpose_101")];
            tensor<fp16, [2, 512, ?]> var_542_cast_fp16 = mul(x = out_11_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_542_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_41_cast_fp16 = add(x = x_m_1_cast_fp16, y = var_542_cast_fp16)[name = string("x_41_cast_fp16")];
            tensor<int32, [3]> input_79_perm_0 = const()[name = string("input_79_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_549_axes_0 = const()[name = string("op_549_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_0_text_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_0_text_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22927424)))];
            tensor<fp16, [512]> main_blocks_0_text_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_0_text_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22928512)))];
            tensor<fp16, [2, ?, 512]> input_79_cast_fp16 = transpose(perm = input_79_perm_0, x = x_41_cast_fp16)[name = string("transpose_100")];
            tensor<fp16, [2, ?, 512]> var_549_cast_fp16 = layer_norm(axes = var_549_axes_0, beta = main_blocks_0_text_norm_norm_bias_to_fp16, epsilon = var_139_to_fp16, gamma = main_blocks_0_text_norm_norm_weight_to_fp16, x = input_79_cast_fp16)[name = string("op_549_cast_fp16")];
            tensor<int32, [3]> var_550_perm_0 = const()[name = string("op_550_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> var_550_cast_fp16 = transpose(perm = var_550_perm_0, x = var_549_cast_fp16)[name = string("transpose_99")];
            tensor<fp16, [2, 512, ?]> x_43_cast_fp16 = mul(x = var_550_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_43_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_45_cast_fp16 = mul(x = x_43_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_45_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_563_cast_fp16 = mul(x = x_45_cast_fp16, y = last_1_cast_fp16)[name = string("op_563_cast_fp16")];
            tensor<int32, [1]> x_last_11_axes_0 = const()[name = string("x_last_11_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_11_keep_dims_0 = const()[name = string("x_last_11_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_11_cast_fp16 = reduce_sum(axes = x_last_11_axes_0, keep_dims = x_last_11_keep_dims_0, x = var_563_cast_fp16)[name = string("x_last_11_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_566_cast_fp16 = mul(x = x_45_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_566_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_568_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_11_cast_fp16)[name = string("op_568_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_83_cast_fp16 = add(x = var_566_cast_fp16, y = var_568_cast_fp16)[name = string("input_83_cast_fp16")];
            tensor<int32, [6]> input_85_pad_0 = const()[name = string("input_85_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_85_mode_0 = const()[name = string("input_85_mode_0"), val = string("replicate")];
            fp16 const_11_to_fp16 = const()[name = string("const_11_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_85_cast_fp16 = pad(constant_val = const_11_to_fp16, mode = input_85_mode_0, pad = input_85_pad_0, x = input_83_cast_fp16)[name = string("input_85_cast_fp16")];
            string h_33_pad_type_0 = const()[name = string("h_33_pad_type_0"), val = string("valid")];
            int32 h_33_groups_0 = const()[name = string("h_33_groups_0"), val = int32(512)];
            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])];
            tensor<fp16, [512, 1, 5]> main_blocks_0_convnext_2_0_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_2_0_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22929600)))];
            tensor<fp16, [512]> main_blocks_0_convnext_2_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_2_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22934784)))];
            tensor<fp16, [2, 512, ?]> h_33_cast_fp16 = conv(bias = main_blocks_0_convnext_2_0_dwconv__conv_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 = main_blocks_0_convnext_2_0_dwconv__conv_weight_to_fp16, x = input_85_cast_fp16)[name = string("h_33_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_47_cast_fp16 = mul(x = h_33_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_47_cast_fp16")];
            tensor<int32, [3]> input_87_perm_0 = const()[name = string("input_87_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_586_axes_0 = const()[name = string("op_586_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_0_convnext_2_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_2_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22935872)))];
            tensor<fp16, [512]> main_blocks_0_convnext_2_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_2_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22936960)))];
            tensor<fp16, [2, ?, 512]> input_87_cast_fp16 = transpose(perm = input_87_perm_0, x = x_47_cast_fp16)[name = string("transpose_98")];
            tensor<fp16, [2, ?, 512]> var_586_cast_fp16 = layer_norm(axes = var_586_axes_0, beta = main_blocks_0_convnext_2_0_norm_norm_bias_to_fp16, epsilon = var_139_to_fp16, gamma = main_blocks_0_convnext_2_0_norm_norm_weight_to_fp16, x = input_87_cast_fp16)[name = string("op_586_cast_fp16")];
            tensor<int32, [3]> input_89_perm_0 = const()[name = string("input_89_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            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, [2048, 512, 1]> main_blocks_0_convnext_2_0_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_2_0_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22938048)))];
            tensor<fp16, [2048]> main_blocks_0_convnext_2_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_2_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25035264)))];
            tensor<fp16, [2, 512, ?]> input_89_cast_fp16 = transpose(perm = input_89_perm_0, x = var_586_cast_fp16)[name = string("transpose_97")];
            tensor<fp16, [2, 2048, ?]> h_35_cast_fp16 = conv(bias = main_blocks_0_convnext_2_0_pwconv1_bias_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 = main_blocks_0_convnext_2_0_pwconv1_weight_to_fp16, x = input_89_cast_fp16)[name = string("h_35_cast_fp16")];
            string input_91_mode_0 = const()[name = string("input_91_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_91_cast_fp16 = gelu(mode = input_91_mode_0, x = h_35_cast_fp16)[name = string("input_91_cast_fp16")];
            string h_37_pad_type_0 = const()[name = string("h_37_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_37_strides_0 = const()[name = string("h_37_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_37_pad_0 = const()[name = string("h_37_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_37_dilations_0 = const()[name = string("h_37_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_37_groups_0 = const()[name = string("h_37_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_603_weight_0_to_fp16 = const()[name = string("op_603_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25039424)))];
            tensor<fp16, [512]> var_603_bias_0_to_fp16 = const()[name = string("op_603_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27136640)))];
            tensor<fp16, [2, 512, ?]> var_603_cast_fp16 = conv(bias = var_603_bias_0_to_fp16, dilations = h_37_dilations_0, groups = h_37_groups_0, pad = h_37_pad_0, pad_type = h_37_pad_type_0, strides = h_37_strides_0, weight = var_603_weight_0_to_fp16, x = input_91_cast_fp16)[name = string("op_603_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_13_cast_fp16 = add(x = x_45_cast_fp16, y = var_603_cast_fp16)[name = string("out_13_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_49_cast_fp16 = mul(x = out_13_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_49_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_m_3_cast_fp16 = mul(x = x_49_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_m_3_cast_fp16")];
            tensor<int32, [3]> input_93_perm_0 = const()[name = string("input_93_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [256, 512]> main_blocks_0_style_attn_W_query_linear_weight_to_fp16 = const()[name = string("main_blocks_0_style_attn_W_query_linear_weight_to_fp16"), val = tensor<fp16, [256, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27137728)))];
            tensor<fp16, [256]> main_blocks_0_style_attn_W_query_linear_bias_to_fp16 = const()[name = string("main_blocks_0_style_attn_W_query_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27399936)))];
            tensor<fp16, [2, ?, 512]> input_93_cast_fp16 = transpose(perm = input_93_perm_0, x = x_49_cast_fp16)[name = string("transpose_96")];
            tensor<fp16, [2, ?, 256]> linear_7_cast_fp16 = linear(bias = main_blocks_0_style_attn_W_query_linear_bias_to_fp16, weight = main_blocks_0_style_attn_W_query_linear_weight_to_fp16, x = input_93_cast_fp16)[name = string("linear_7_cast_fp16")];
            tensor<int32, [4]> concat_12x = const()[name = string("concat_12x"), val = tensor<int32, [4]>([2, -1, 2, 128])];
            tensor<fp16, [2, ?, 2, 128]> var_620_cast_fp16 = reshape(shape = concat_12x, x = linear_7_cast_fp16)[name = string("op_620_cast_fp16")];
            tensor<int32, [4]> q_3_perm_0 = const()[name = string("q_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [256, 256]> main_blocks_0_style_attn_W_value_linear_weight_to_fp16 = const()[name = string("main_blocks_0_style_attn_W_value_linear_weight_to_fp16"), val = tensor<fp16, [256, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27400512)))];
            tensor<fp16, [256]> main_blocks_0_style_attn_W_value_linear_bias_to_fp16 = const()[name = string("main_blocks_0_style_attn_W_value_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27531648)))];
            tensor<fp16, [2, 50, 256]> linear_9_cast_fp16 = linear(bias = main_blocks_0_style_attn_W_value_linear_bias_to_fp16, weight = main_blocks_0_style_attn_W_value_linear_weight_to_fp16, x = input_95_cast_fp16)[name = string("linear_9_cast_fp16")];
            tensor<int32, [4]> var_633 = const()[name = string("op_633"), val = tensor<int32, [4]>([2, 50, 2, 128])];
            tensor<fp16, [2, 50, 2, 128]> var_634_cast_fp16 = reshape(shape = var_633, x = linear_9_cast_fp16)[name = string("op_634_cast_fp16")];
            tensor<int32, [4]> v_3_perm_0 = const()[name = string("v_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_638_transpose_x_0 = const()[name = string("op_638_transpose_x_0"), val = bool(false)];
            bool var_638_transpose_y_0 = const()[name = string("op_638_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 2, 128, 50]> var_637_to_fp16 = const()[name = string("op_637_to_fp16"), val = tensor<fp16, [2, 2, 128, 50]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27532224)))];
            tensor<fp16, [2, 2, ?, 128]> q_3_cast_fp16 = transpose(perm = q_3_perm_0, x = var_620_cast_fp16)[name = string("transpose_95")];
            tensor<fp16, [2, 2, ?, 50]> var_638_cast_fp16 = matmul(transpose_x = var_638_transpose_x_0, transpose_y = var_638_transpose_y_0, x = q_3_cast_fp16, y = var_637_to_fp16)[name = string("op_638_cast_fp16")];
            fp16 _inversed_input_97_y_0_to_fp16 = const()[name = string("_inversed_input_97_y_0_to_fp16"), val = fp16(0x1p-4)];
            tensor<fp16, [2, 2, ?, 50]> _inversed_input_97_cast_fp16 = mul(x = var_638_cast_fp16, y = _inversed_input_97_y_0_to_fp16)[name = string("_inversed_input_97_cast_fp16")];
            tensor<fp16, [2, 2, ?, 50]> attn_3_cast_fp16 = softmax(axis = var_134, x = _inversed_input_97_cast_fp16)[name = string("attn_3_cast_fp16")];
            tensor<int32, [3]> var_642_perm_0 = const()[name = string("op_642_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> mask_1_axes_0 = const()[name = string("mask_1_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, ?, 1]> var_642_cast_fp16 = transpose(perm = var_642_perm_0, x = latent_mask_b_cast_fp16)[name = string("transpose_93")];
            tensor<fp16, [2, 1, ?, 1]> mask_1_cast_fp16 = expand_dims(axes = mask_1_axes_0, x = var_642_cast_fp16)[name = string("mask_1_cast_fp16")];
            tensor<fp16, [2, 2, ?, 50]> attn_5_cast_fp16 = mul(x = attn_3_cast_fp16, y = mask_1_cast_fp16)[name = string("attn_5_cast_fp16")];
            bool var_645_transpose_x_0 = const()[name = string("op_645_transpose_x_0"), val = bool(false)];
            bool var_645_transpose_y_0 = const()[name = string("op_645_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 2, 50, 128]> v_3_cast_fp16 = transpose(perm = v_3_perm_0, x = var_634_cast_fp16)[name = string("transpose_94")];
            tensor<fp16, [2, 2, ?, 128]> var_645_cast_fp16 = matmul(transpose_x = var_645_transpose_x_0, transpose_y = var_645_transpose_y_0, x = attn_5_cast_fp16, y = v_3_cast_fp16)[name = string("op_645_cast_fp16")];
            tensor<int32, [4]> var_646_perm_0 = const()[name = string("op_646_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_13x = const()[name = string("concat_13x"), val = tensor<int32, [3]>([2, -1, 256])];
            tensor<fp16, [2, ?, 2, 128]> var_646_cast_fp16 = transpose(perm = var_646_perm_0, x = var_645_cast_fp16)[name = string("transpose_92")];
            tensor<fp16, [2, ?, 256]> input_99_cast_fp16 = reshape(shape = concat_13x, x = var_646_cast_fp16)[name = string("input_99_cast_fp16")];
            tensor<fp16, [512, 256]> main_blocks_0_style_attn_out_fc_linear_weight_to_fp16 = const()[name = string("main_blocks_0_style_attn_out_fc_linear_weight_to_fp16"), val = tensor<fp16, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27583488)))];
            tensor<fp16, [512]> main_blocks_0_style_attn_out_fc_linear_bias_to_fp16 = const()[name = string("main_blocks_0_style_attn_out_fc_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27845696)))];
            tensor<fp16, [2, ?, 512]> linear_10_cast_fp16 = linear(bias = main_blocks_0_style_attn_out_fc_linear_bias_to_fp16, weight = main_blocks_0_style_attn_out_fc_linear_weight_to_fp16, x = input_99_cast_fp16)[name = string("linear_10_cast_fp16")];
            tensor<int32, [3]> out_15_perm_0 = const()[name = string("out_15_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> out_15_cast_fp16 = transpose(perm = out_15_perm_0, x = linear_10_cast_fp16)[name = string("transpose_91")];
            tensor<fp16, [2, 512, ?]> var_654_cast_fp16 = mul(x = out_15_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_654_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_51_cast_fp16 = add(x = x_m_3_cast_fp16, y = var_654_cast_fp16)[name = string("x_51_cast_fp16")];
            tensor<int32, [3]> input_101_perm_0 = const()[name = string("input_101_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_661_axes_0 = const()[name = string("op_661_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_0_style_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_0_style_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27846784)))];
            tensor<fp16, [512]> main_blocks_0_style_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_0_style_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27847872)))];
            tensor<fp16, [2, ?, 512]> input_101_cast_fp16 = transpose(perm = input_101_perm_0, x = x_51_cast_fp16)[name = string("transpose_90")];
            tensor<fp16, [2, ?, 512]> var_661_cast_fp16 = layer_norm(axes = var_661_axes_0, beta = main_blocks_0_style_norm_norm_bias_to_fp16, epsilon = var_139_to_fp16, gamma = main_blocks_0_style_norm_norm_weight_to_fp16, x = input_101_cast_fp16)[name = string("op_661_cast_fp16")];
            tensor<int32, [3]> var_662_perm_0 = const()[name = string("op_662_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> var_662_cast_fp16 = transpose(perm = var_662_perm_0, x = var_661_cast_fp16)[name = string("transpose_89")];
            tensor<fp16, [2, 512, ?]> x_53_cast_fp16 = mul(x = var_662_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_53_cast_fp16")];
            int32 var_681 = const()[name = string("op_681"), val = int32(-1)];
            tensor<fp16, [2, 512, ?]> x_55_cast_fp16 = mul(x = x_53_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_55_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_716_cast_fp16 = mul(x = x_55_cast_fp16, y = last_1_cast_fp16)[name = string("op_716_cast_fp16")];
            tensor<int32, [1]> x_last_13_axes_0 = const()[name = string("x_last_13_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_13_keep_dims_0 = const()[name = string("x_last_13_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_13_cast_fp16 = reduce_sum(axes = x_last_13_axes_0, keep_dims = x_last_13_keep_dims_0, x = var_716_cast_fp16)[name = string("x_last_13_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_719_cast_fp16 = mul(x = x_55_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_719_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_721_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_13_cast_fp16)[name = string("op_721_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_105_cast_fp16 = add(x = var_719_cast_fp16, y = var_721_cast_fp16)[name = string("input_105_cast_fp16")];
            tensor<int32, [6]> input_107_pad_0 = const()[name = string("input_107_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_107_mode_0 = const()[name = string("input_107_mode_0"), val = string("replicate")];
            fp16 const_14_to_fp16 = const()[name = string("const_14_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_107_cast_fp16 = pad(constant_val = const_14_to_fp16, mode = input_107_mode_0, pad = input_107_pad_0, x = input_105_cast_fp16)[name = string("input_107_cast_fp16")];
            string h_39_pad_type_0 = const()[name = string("h_39_pad_type_0"), val = string("valid")];
            int32 h_39_groups_0 = const()[name = string("h_39_groups_0"), val = int32(512)];
            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])];
            tensor<fp16, [512, 1, 5]> main_blocks_1_convnext_0_0_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_0_0_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27848960)))];
            tensor<fp16, [512]> main_blocks_1_convnext_0_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27854144)))];
            tensor<fp16, [2, 512, ?]> h_39_cast_fp16 = conv(bias = main_blocks_1_convnext_0_0_dwconv__conv_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 = main_blocks_1_convnext_0_0_dwconv__conv_weight_to_fp16, x = input_107_cast_fp16)[name = string("h_39_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_57_cast_fp16 = mul(x = h_39_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_57_cast_fp16")];
            tensor<int32, [3]> input_109_perm_0 = const()[name = string("input_109_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_739_axes_0 = const()[name = string("op_739_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_1_convnext_0_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_0_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27855232)))];
            tensor<fp16, [512]> main_blocks_1_convnext_0_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27856320)))];
            fp16 var_686_to_fp16 = const()[name = string("op_686_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [2, ?, 512]> input_109_cast_fp16 = transpose(perm = input_109_perm_0, x = x_57_cast_fp16)[name = string("transpose_88")];
            tensor<fp16, [2, ?, 512]> var_739_cast_fp16 = layer_norm(axes = var_739_axes_0, beta = main_blocks_1_convnext_0_0_norm_norm_bias_to_fp16, epsilon = var_686_to_fp16, gamma = main_blocks_1_convnext_0_0_norm_norm_weight_to_fp16, x = input_109_cast_fp16)[name = string("op_739_cast_fp16")];
            tensor<int32, [3]> input_111_perm_0 = const()[name = string("input_111_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_41_pad_type_0 = const()[name = string("h_41_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_41_strides_0 = const()[name = string("h_41_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_41_pad_0 = const()[name = string("h_41_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_41_dilations_0 = const()[name = string("h_41_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_41_groups_0 = const()[name = string("h_41_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_1_convnext_0_0_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_0_0_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27857408)))];
            tensor<fp16, [2048]> main_blocks_1_convnext_0_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29954624)))];
            tensor<fp16, [2, 512, ?]> input_111_cast_fp16 = transpose(perm = input_111_perm_0, x = var_739_cast_fp16)[name = string("transpose_87")];
            tensor<fp16, [2, 2048, ?]> h_41_cast_fp16 = conv(bias = main_blocks_1_convnext_0_0_pwconv1_bias_to_fp16, dilations = h_41_dilations_0, groups = h_41_groups_0, pad = h_41_pad_0, pad_type = h_41_pad_type_0, strides = h_41_strides_0, weight = main_blocks_1_convnext_0_0_pwconv1_weight_to_fp16, x = input_111_cast_fp16)[name = string("h_41_cast_fp16")];
            string input_113_mode_0 = const()[name = string("input_113_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_113_cast_fp16 = gelu(mode = input_113_mode_0, x = h_41_cast_fp16)[name = string("input_113_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, [512, 2048, 1]> var_756_weight_0_to_fp16 = const()[name = string("op_756_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29958784)))];
            tensor<fp16, [512]> var_756_bias_0_to_fp16 = const()[name = string("op_756_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32056000)))];
            tensor<fp16, [2, 512, ?]> var_756_cast_fp16 = conv(bias = var_756_bias_0_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 = var_756_weight_0_to_fp16, x = input_113_cast_fp16)[name = string("op_756_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_17_cast_fp16 = add(x = x_55_cast_fp16, y = var_756_cast_fp16)[name = string("out_17_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_59_cast_fp16 = mul(x = out_17_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_59_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_61_cast_fp16 = mul(x = x_59_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_61_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_770_cast_fp16 = mul(x = x_61_cast_fp16, y = last_1_cast_fp16)[name = string("op_770_cast_fp16")];
            tensor<int32, [1]> x_last_15_axes_0 = const()[name = string("x_last_15_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_15_keep_dims_0 = const()[name = string("x_last_15_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_15_cast_fp16 = reduce_sum(axes = x_last_15_axes_0, keep_dims = x_last_15_keep_dims_0, x = var_770_cast_fp16)[name = string("x_last_15_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_773_cast_fp16 = mul(x = x_61_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_773_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_775_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_15_cast_fp16)[name = string("op_775_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_117_cast_fp16 = add(x = var_773_cast_fp16, y = var_775_cast_fp16)[name = string("input_117_cast_fp16")];
            tensor<int32, [6]> input_119_pad_0 = const()[name = string("input_119_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_119_mode_0 = const()[name = string("input_119_mode_0"), val = string("replicate")];
            fp16 const_16_to_fp16 = const()[name = string("const_16_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_119_cast_fp16 = pad(constant_val = const_16_to_fp16, mode = input_119_mode_0, pad = input_119_pad_0, x = input_117_cast_fp16)[name = string("input_119_cast_fp16")];
            string h_45_pad_type_0 = const()[name = string("h_45_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_45_dilations_0 = const()[name = string("h_45_dilations_0"), val = tensor<int32, [1]>([2])];
            int32 h_45_groups_0 = const()[name = string("h_45_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_45_strides_0 = const()[name = string("h_45_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_45_pad_0 = const()[name = string("h_45_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [512, 1, 5]> main_blocks_1_convnext_0_1_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_0_1_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32057088)))];
            tensor<fp16, [512]> main_blocks_1_convnext_0_1_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_1_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32062272)))];
            tensor<fp16, [2, 512, ?]> h_45_cast_fp16 = conv(bias = main_blocks_1_convnext_0_1_dwconv__conv_bias_to_fp16, dilations = h_45_dilations_0, groups = h_45_groups_0, pad = h_45_pad_0, pad_type = h_45_pad_type_0, strides = h_45_strides_0, weight = main_blocks_1_convnext_0_1_dwconv__conv_weight_to_fp16, x = input_119_cast_fp16)[name = string("h_45_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_63_cast_fp16 = mul(x = h_45_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_63_cast_fp16")];
            tensor<int32, [3]> input_121_perm_0 = const()[name = string("input_121_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_793_axes_0 = const()[name = string("op_793_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_1_convnext_0_1_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_0_1_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32063360)))];
            tensor<fp16, [512]> main_blocks_1_convnext_0_1_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_1_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32064448)))];
            tensor<fp16, [2, ?, 512]> input_121_cast_fp16 = transpose(perm = input_121_perm_0, x = x_63_cast_fp16)[name = string("transpose_86")];
            tensor<fp16, [2, ?, 512]> var_793_cast_fp16 = layer_norm(axes = var_793_axes_0, beta = main_blocks_1_convnext_0_1_norm_norm_bias_to_fp16, epsilon = var_686_to_fp16, gamma = main_blocks_1_convnext_0_1_norm_norm_weight_to_fp16, x = input_121_cast_fp16)[name = string("op_793_cast_fp16")];
            tensor<int32, [3]> input_123_perm_0 = const()[name = string("input_123_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            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, [2048, 512, 1]> main_blocks_1_convnext_0_1_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_0_1_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32065536)))];
            tensor<fp16, [2048]> main_blocks_1_convnext_0_1_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_1_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34162752)))];
            tensor<fp16, [2, 512, ?]> input_123_cast_fp16 = transpose(perm = input_123_perm_0, x = var_793_cast_fp16)[name = string("transpose_85")];
            tensor<fp16, [2, 2048, ?]> h_47_cast_fp16 = conv(bias = main_blocks_1_convnext_0_1_pwconv1_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 = main_blocks_1_convnext_0_1_pwconv1_weight_to_fp16, x = input_123_cast_fp16)[name = string("h_47_cast_fp16")];
            string input_125_mode_0 = const()[name = string("input_125_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_125_cast_fp16 = gelu(mode = input_125_mode_0, x = h_47_cast_fp16)[name = string("input_125_cast_fp16")];
            string h_49_pad_type_0 = const()[name = string("h_49_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_49_strides_0 = const()[name = string("h_49_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_49_pad_0 = const()[name = string("h_49_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_49_dilations_0 = const()[name = string("h_49_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_49_groups_0 = const()[name = string("h_49_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_810_weight_0_to_fp16 = const()[name = string("op_810_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34166912)))];
            tensor<fp16, [512]> var_810_bias_0_to_fp16 = const()[name = string("op_810_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36264128)))];
            tensor<fp16, [2, 512, ?]> var_810_cast_fp16 = conv(bias = var_810_bias_0_to_fp16, dilations = h_49_dilations_0, groups = h_49_groups_0, pad = h_49_pad_0, pad_type = h_49_pad_type_0, strides = h_49_strides_0, weight = var_810_weight_0_to_fp16, x = input_125_cast_fp16)[name = string("op_810_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_19_cast_fp16 = add(x = x_61_cast_fp16, y = var_810_cast_fp16)[name = string("out_19_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_65_cast_fp16 = mul(x = out_19_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_65_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_67_cast_fp16 = mul(x = x_65_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_67_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_824_cast_fp16 = mul(x = x_67_cast_fp16, y = last_1_cast_fp16)[name = string("op_824_cast_fp16")];
            tensor<int32, [1]> x_last_17_axes_0 = const()[name = string("x_last_17_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_17_keep_dims_0 = const()[name = string("x_last_17_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_17_cast_fp16 = reduce_sum(axes = x_last_17_axes_0, keep_dims = x_last_17_keep_dims_0, x = var_824_cast_fp16)[name = string("x_last_17_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_827_cast_fp16 = mul(x = x_67_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_827_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_829_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_17_cast_fp16)[name = string("op_829_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_129_cast_fp16 = add(x = var_827_cast_fp16, y = var_829_cast_fp16)[name = string("input_129_cast_fp16")];
            tensor<int32, [6]> input_131_pad_0 = const()[name = string("input_131_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 8, 8])];
            string input_131_mode_0 = const()[name = string("input_131_mode_0"), val = string("replicate")];
            fp16 const_18_to_fp16 = const()[name = string("const_18_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_131_cast_fp16 = pad(constant_val = const_18_to_fp16, mode = input_131_mode_0, pad = input_131_pad_0, x = input_129_cast_fp16)[name = string("input_131_cast_fp16")];
            string h_51_pad_type_0 = const()[name = string("h_51_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_51_dilations_0 = const()[name = string("h_51_dilations_0"), val = tensor<int32, [1]>([4])];
            int32 h_51_groups_0 = const()[name = string("h_51_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_51_strides_0 = const()[name = string("h_51_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_51_pad_0 = const()[name = string("h_51_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [512, 1, 5]> main_blocks_1_convnext_0_2_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_0_2_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36265216)))];
            tensor<fp16, [512]> main_blocks_1_convnext_0_2_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_2_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36270400)))];
            tensor<fp16, [2, 512, ?]> h_51_cast_fp16 = conv(bias = main_blocks_1_convnext_0_2_dwconv__conv_bias_to_fp16, dilations = h_51_dilations_0, groups = h_51_groups_0, pad = h_51_pad_0, pad_type = h_51_pad_type_0, strides = h_51_strides_0, weight = main_blocks_1_convnext_0_2_dwconv__conv_weight_to_fp16, x = input_131_cast_fp16)[name = string("h_51_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_69_cast_fp16 = mul(x = h_51_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_69_cast_fp16")];
            tensor<int32, [3]> input_133_perm_0 = const()[name = string("input_133_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_847_axes_0 = const()[name = string("op_847_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_1_convnext_0_2_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_0_2_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36271488)))];
            tensor<fp16, [512]> main_blocks_1_convnext_0_2_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_2_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36272576)))];
            tensor<fp16, [2, ?, 512]> input_133_cast_fp16 = transpose(perm = input_133_perm_0, x = x_69_cast_fp16)[name = string("transpose_84")];
            tensor<fp16, [2, ?, 512]> var_847_cast_fp16 = layer_norm(axes = var_847_axes_0, beta = main_blocks_1_convnext_0_2_norm_norm_bias_to_fp16, epsilon = var_686_to_fp16, gamma = main_blocks_1_convnext_0_2_norm_norm_weight_to_fp16, x = input_133_cast_fp16)[name = string("op_847_cast_fp16")];
            tensor<int32, [3]> input_135_perm_0 = const()[name = string("input_135_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_53_pad_type_0 = const()[name = string("h_53_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_53_strides_0 = const()[name = string("h_53_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_53_pad_0 = const()[name = string("h_53_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_53_dilations_0 = const()[name = string("h_53_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_53_groups_0 = const()[name = string("h_53_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_1_convnext_0_2_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_0_2_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36273664)))];
            tensor<fp16, [2048]> main_blocks_1_convnext_0_2_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_2_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38370880)))];
            tensor<fp16, [2, 512, ?]> input_135_cast_fp16 = transpose(perm = input_135_perm_0, x = var_847_cast_fp16)[name = string("transpose_83")];
            tensor<fp16, [2, 2048, ?]> h_53_cast_fp16 = conv(bias = main_blocks_1_convnext_0_2_pwconv1_bias_to_fp16, dilations = h_53_dilations_0, groups = h_53_groups_0, pad = h_53_pad_0, pad_type = h_53_pad_type_0, strides = h_53_strides_0, weight = main_blocks_1_convnext_0_2_pwconv1_weight_to_fp16, x = input_135_cast_fp16)[name = string("h_53_cast_fp16")];
            string input_137_mode_0 = const()[name = string("input_137_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_137_cast_fp16 = gelu(mode = input_137_mode_0, x = h_53_cast_fp16)[name = string("input_137_cast_fp16")];
            string h_55_pad_type_0 = const()[name = string("h_55_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_55_strides_0 = const()[name = string("h_55_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_55_pad_0 = const()[name = string("h_55_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_55_dilations_0 = const()[name = string("h_55_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_55_groups_0 = const()[name = string("h_55_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_864_weight_0_to_fp16 = const()[name = string("op_864_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38375040)))];
            tensor<fp16, [512]> var_864_bias_0_to_fp16 = const()[name = string("op_864_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40472256)))];
            tensor<fp16, [2, 512, ?]> var_864_cast_fp16 = conv(bias = var_864_bias_0_to_fp16, dilations = h_55_dilations_0, groups = h_55_groups_0, pad = h_55_pad_0, pad_type = h_55_pad_type_0, strides = h_55_strides_0, weight = var_864_weight_0_to_fp16, x = input_137_cast_fp16)[name = string("op_864_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_21_cast_fp16 = add(x = x_67_cast_fp16, y = var_864_cast_fp16)[name = string("out_21_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_71_cast_fp16 = mul(x = out_21_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_71_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_73_cast_fp16 = mul(x = x_71_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_73_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_878_cast_fp16 = mul(x = x_73_cast_fp16, y = last_1_cast_fp16)[name = string("op_878_cast_fp16")];
            tensor<int32, [1]> x_last_19_axes_0 = const()[name = string("x_last_19_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_19_keep_dims_0 = const()[name = string("x_last_19_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_19_cast_fp16 = reduce_sum(axes = x_last_19_axes_0, keep_dims = x_last_19_keep_dims_0, x = var_878_cast_fp16)[name = string("x_last_19_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_881_cast_fp16 = mul(x = x_73_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_881_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_883_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_19_cast_fp16)[name = string("op_883_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_141_cast_fp16 = add(x = var_881_cast_fp16, y = var_883_cast_fp16)[name = string("input_141_cast_fp16")];
            tensor<int32, [6]> input_143_pad_0 = const()[name = string("input_143_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 16, 16])];
            string input_143_mode_0 = const()[name = string("input_143_mode_0"), val = string("replicate")];
            fp16 const_20_to_fp16 = const()[name = string("const_20_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_143_cast_fp16 = pad(constant_val = const_20_to_fp16, mode = input_143_mode_0, pad = input_143_pad_0, x = input_141_cast_fp16)[name = string("input_143_cast_fp16")];
            string h_57_pad_type_0 = const()[name = string("h_57_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_57_dilations_0 = const()[name = string("h_57_dilations_0"), val = tensor<int32, [1]>([8])];
            int32 h_57_groups_0 = const()[name = string("h_57_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_57_strides_0 = const()[name = string("h_57_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_57_pad_0 = const()[name = string("h_57_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [512, 1, 5]> main_blocks_1_convnext_0_3_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_0_3_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40473344)))];
            tensor<fp16, [512]> main_blocks_1_convnext_0_3_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_3_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40478528)))];
            tensor<fp16, [2, 512, ?]> h_57_cast_fp16 = conv(bias = main_blocks_1_convnext_0_3_dwconv__conv_bias_to_fp16, dilations = h_57_dilations_0, groups = h_57_groups_0, pad = h_57_pad_0, pad_type = h_57_pad_type_0, strides = h_57_strides_0, weight = main_blocks_1_convnext_0_3_dwconv__conv_weight_to_fp16, x = input_143_cast_fp16)[name = string("h_57_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_75_cast_fp16 = mul(x = h_57_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_75_cast_fp16")];
            tensor<int32, [3]> input_145_perm_0 = const()[name = string("input_145_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_901_axes_0 = const()[name = string("op_901_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_1_convnext_0_3_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_0_3_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40479616)))];
            tensor<fp16, [512]> main_blocks_1_convnext_0_3_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_3_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40480704)))];
            tensor<fp16, [2, ?, 512]> input_145_cast_fp16 = transpose(perm = input_145_perm_0, x = x_75_cast_fp16)[name = string("transpose_82")];
            tensor<fp16, [2, ?, 512]> var_901_cast_fp16 = layer_norm(axes = var_901_axes_0, beta = main_blocks_1_convnext_0_3_norm_norm_bias_to_fp16, epsilon = var_686_to_fp16, gamma = main_blocks_1_convnext_0_3_norm_norm_weight_to_fp16, x = input_145_cast_fp16)[name = string("op_901_cast_fp16")];
            tensor<int32, [3]> input_147_perm_0 = const()[name = string("input_147_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_59_pad_type_0 = const()[name = string("h_59_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_59_strides_0 = const()[name = string("h_59_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_59_pad_0 = const()[name = string("h_59_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_59_dilations_0 = const()[name = string("h_59_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_59_groups_0 = const()[name = string("h_59_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_1_convnext_0_3_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_0_3_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40481792)))];
            tensor<fp16, [2048]> main_blocks_1_convnext_0_3_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_3_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42579008)))];
            tensor<fp16, [2, 512, ?]> input_147_cast_fp16 = transpose(perm = input_147_perm_0, x = var_901_cast_fp16)[name = string("transpose_81")];
            tensor<fp16, [2, 2048, ?]> h_59_cast_fp16 = conv(bias = main_blocks_1_convnext_0_3_pwconv1_bias_to_fp16, dilations = h_59_dilations_0, groups = h_59_groups_0, pad = h_59_pad_0, pad_type = h_59_pad_type_0, strides = h_59_strides_0, weight = main_blocks_1_convnext_0_3_pwconv1_weight_to_fp16, x = input_147_cast_fp16)[name = string("h_59_cast_fp16")];
            string input_149_mode_0 = const()[name = string("input_149_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_149_cast_fp16 = gelu(mode = input_149_mode_0, x = h_59_cast_fp16)[name = string("input_149_cast_fp16")];
            string h_61_pad_type_0 = const()[name = string("h_61_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_61_strides_0 = const()[name = string("h_61_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_61_pad_0 = const()[name = string("h_61_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_61_dilations_0 = const()[name = string("h_61_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_61_groups_0 = const()[name = string("h_61_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_918_weight_0_to_fp16 = const()[name = string("op_918_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42583168)))];
            tensor<fp16, [512]> var_918_bias_0_to_fp16 = const()[name = string("op_918_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44680384)))];
            tensor<fp16, [2, 512, ?]> var_918_cast_fp16 = conv(bias = var_918_bias_0_to_fp16, dilations = h_61_dilations_0, groups = h_61_groups_0, pad = h_61_pad_0, pad_type = h_61_pad_type_0, strides = h_61_strides_0, weight = var_918_weight_0_to_fp16, x = input_149_cast_fp16)[name = string("op_918_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_23_cast_fp16 = add(x = x_73_cast_fp16, y = var_918_cast_fp16)[name = string("out_23_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_77_cast_fp16 = mul(x = out_23_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_77_cast_fp16")];
            tensor<fp16, [512, 64]> main_blocks_1_time_cond_linear_linear_weight_to_fp16 = const()[name = string("main_blocks_1_time_cond_linear_linear_weight_to_fp16"), val = tensor<fp16, [512, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44681472)))];
            tensor<fp16, [512]> main_blocks_1_time_cond_linear_linear_bias_to_fp16 = const()[name = string("main_blocks_1_time_cond_linear_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44747072)))];
            tensor<fp16, [2, 512]> linear_11_cast_fp16 = linear(bias = main_blocks_1_time_cond_linear_linear_bias_to_fp16, weight = main_blocks_1_time_cond_linear_linear_weight_to_fp16, x = input_55_cast_fp16)[name = string("linear_11_cast_fp16")];
            tensor<int32, [1]> t_7_axes_0 = const()[name = string("t_7_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2, 512, 1]> t_7_cast_fp16 = expand_dims(axes = t_7_axes_0, x = linear_11_cast_fp16)[name = string("t_7_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_928_cast_fp16 = add(x = x_77_cast_fp16, y = t_7_cast_fp16)[name = string("op_928_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_79_cast_fp16 = mul(x = var_928_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_79_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_81_cast_fp16 = mul(x = x_79_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_81_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_941_cast_fp16 = mul(x = x_81_cast_fp16, y = last_1_cast_fp16)[name = string("op_941_cast_fp16")];
            tensor<int32, [1]> x_last_21_axes_0 = const()[name = string("x_last_21_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_21_keep_dims_0 = const()[name = string("x_last_21_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_21_cast_fp16 = reduce_sum(axes = x_last_21_axes_0, keep_dims = x_last_21_keep_dims_0, x = var_941_cast_fp16)[name = string("x_last_21_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_944_cast_fp16 = mul(x = x_81_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_944_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_946_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_21_cast_fp16)[name = string("op_946_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_155_cast_fp16 = add(x = var_944_cast_fp16, y = var_946_cast_fp16)[name = string("input_155_cast_fp16")];
            tensor<int32, [6]> input_157_pad_0 = const()[name = string("input_157_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_157_mode_0 = const()[name = string("input_157_mode_0"), val = string("replicate")];
            fp16 const_22_to_fp16 = const()[name = string("const_22_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_157_cast_fp16 = pad(constant_val = const_22_to_fp16, mode = input_157_mode_0, pad = input_157_pad_0, x = input_155_cast_fp16)[name = string("input_157_cast_fp16")];
            string h_63_pad_type_0 = const()[name = string("h_63_pad_type_0"), val = string("valid")];
            int32 h_63_groups_0 = const()[name = string("h_63_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_63_strides_0 = const()[name = string("h_63_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_63_pad_0 = const()[name = string("h_63_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_63_dilations_0 = const()[name = string("h_63_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> main_blocks_1_convnext_1_0_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_1_0_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44748160)))];
            tensor<fp16, [512]> main_blocks_1_convnext_1_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_1_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44753344)))];
            tensor<fp16, [2, 512, ?]> h_63_cast_fp16 = conv(bias = main_blocks_1_convnext_1_0_dwconv__conv_bias_to_fp16, dilations = h_63_dilations_0, groups = h_63_groups_0, pad = h_63_pad_0, pad_type = h_63_pad_type_0, strides = h_63_strides_0, weight = main_blocks_1_convnext_1_0_dwconv__conv_weight_to_fp16, x = input_157_cast_fp16)[name = string("h_63_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_83_cast_fp16 = mul(x = h_63_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_83_cast_fp16")];
            tensor<int32, [3]> input_159_perm_0 = const()[name = string("input_159_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_964_axes_0 = const()[name = string("op_964_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_1_convnext_1_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_1_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44754432)))];
            tensor<fp16, [512]> main_blocks_1_convnext_1_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_1_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44755520)))];
            tensor<fp16, [2, ?, 512]> input_159_cast_fp16 = transpose(perm = input_159_perm_0, x = x_83_cast_fp16)[name = string("transpose_80")];
            tensor<fp16, [2, ?, 512]> var_964_cast_fp16 = layer_norm(axes = var_964_axes_0, beta = main_blocks_1_convnext_1_0_norm_norm_bias_to_fp16, epsilon = var_686_to_fp16, gamma = main_blocks_1_convnext_1_0_norm_norm_weight_to_fp16, x = input_159_cast_fp16)[name = string("op_964_cast_fp16")];
            tensor<int32, [3]> input_161_perm_0 = const()[name = string("input_161_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_65_pad_type_0 = const()[name = string("h_65_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_65_strides_0 = const()[name = string("h_65_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_65_pad_0 = const()[name = string("h_65_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_65_dilations_0 = const()[name = string("h_65_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_65_groups_0 = const()[name = string("h_65_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_1_convnext_1_0_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_1_0_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44756608)))];
            tensor<fp16, [2048]> main_blocks_1_convnext_1_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_1_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46853824)))];
            tensor<fp16, [2, 512, ?]> input_161_cast_fp16 = transpose(perm = input_161_perm_0, x = var_964_cast_fp16)[name = string("transpose_79")];
            tensor<fp16, [2, 2048, ?]> h_65_cast_fp16 = conv(bias = main_blocks_1_convnext_1_0_pwconv1_bias_to_fp16, dilations = h_65_dilations_0, groups = h_65_groups_0, pad = h_65_pad_0, pad_type = h_65_pad_type_0, strides = h_65_strides_0, weight = main_blocks_1_convnext_1_0_pwconv1_weight_to_fp16, x = input_161_cast_fp16)[name = string("h_65_cast_fp16")];
            string input_163_mode_0 = const()[name = string("input_163_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_163_cast_fp16 = gelu(mode = input_163_mode_0, x = h_65_cast_fp16)[name = string("input_163_cast_fp16")];
            string h_67_pad_type_0 = const()[name = string("h_67_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_67_strides_0 = const()[name = string("h_67_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_67_pad_0 = const()[name = string("h_67_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_67_dilations_0 = const()[name = string("h_67_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_67_groups_0 = const()[name = string("h_67_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_981_weight_0_to_fp16 = const()[name = string("op_981_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46857984)))];
            tensor<fp16, [512]> var_981_bias_0_to_fp16 = const()[name = string("op_981_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48955200)))];
            tensor<fp16, [2, 512, ?]> var_981_cast_fp16 = conv(bias = var_981_bias_0_to_fp16, dilations = h_67_dilations_0, groups = h_67_groups_0, pad = h_67_pad_0, pad_type = h_67_pad_type_0, strides = h_67_strides_0, weight = var_981_weight_0_to_fp16, x = input_163_cast_fp16)[name = string("op_981_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_25_cast_fp16 = add(x = x_81_cast_fp16, y = var_981_cast_fp16)[name = string("out_25_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_85_cast_fp16 = mul(x = out_25_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_85_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_m_5_cast_fp16 = mul(x = x_85_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_m_5_cast_fp16")];
            tensor<int32, [3]> var_991_shape_cast_fp16 = shape(x = x_85_cast_fp16)[name = string("op_991_shape_cast_fp16")];
            int32 gather_9_axis_0 = const()[name = string("gather_9_axis_0"), val = int32(0)];
            int32 gather_9_batch_dims_0 = const()[name = string("gather_9_batch_dims_0"), val = int32(0)];
            bool gather_9_validate_indices_0 = const()[name = string("gather_9_validate_indices_0"), val = bool(false)];
            string var_991_shape_cast_fp16_to_uint16_dtype_0 = const()[name = string("op_991_shape_cast_fp16_to_uint16_dtype_0"), val = string("uint16")];
            uint16 gather_9_indices_0_to_uint16 = const()[name = string("gather_9_indices_0_to_uint16"), val = uint16(2)];
            tensor<uint16, [3]> var_991_shape_cast_fp16_to_uint16 = cast(dtype = var_991_shape_cast_fp16_to_uint16_dtype_0, x = var_991_shape_cast_fp16)[name = string("cast_91")];
            uint16 gather_9_cast_uint16 = gather(axis = gather_9_axis_0, batch_dims = gather_9_batch_dims_0, indices = gather_9_indices_0_to_uint16, validate_indices = gather_9_validate_indices_0, x = var_991_shape_cast_fp16_to_uint16)[name = string("gather_9_cast_uint16")];
            string gather_9_cast_uint16_to_int32_dtype_0 = const()[name = string("gather_9_cast_uint16_to_int32_dtype_0"), val = string("int32")];
            tensor<int32, [3]> input_165_perm_0 = const()[name = string("input_165_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> main_blocks_1_text_attn_W_query_linear_weight_to_fp16 = const()[name = string("main_blocks_1_text_attn_W_query_linear_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48956288)))];
            tensor<fp16, [512]> main_blocks_1_text_attn_W_query_linear_bias_to_fp16 = const()[name = string("main_blocks_1_text_attn_W_query_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49480640)))];
            tensor<fp16, [2, ?, 512]> input_165_cast_fp16 = transpose(perm = input_165_perm_0, x = x_85_cast_fp16)[name = string("transpose_78")];
            tensor<fp16, [2, ?, 512]> linear_12_cast_fp16 = linear(bias = main_blocks_1_text_attn_W_query_linear_bias_to_fp16, weight = main_blocks_1_text_attn_W_query_linear_weight_to_fp16, x = input_165_cast_fp16)[name = string("linear_12_cast_fp16")];
            tensor<int32, [4]> concat_14x = const()[name = string("concat_14x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_1000_cast_fp16 = reshape(shape = concat_14x, x = linear_12_cast_fp16)[name = string("op_1000_cast_fp16")];
            tensor<int32, [4]> x_87_perm_0 = const()[name = string("x_87_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [512, 256]> main_blocks_1_text_attn_W_key_linear_weight_to_fp16 = const()[name = string("main_blocks_1_text_attn_W_key_linear_weight_to_fp16"), val = tensor<fp16, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49481728)))];
            tensor<fp16, [512]> main_blocks_1_text_attn_W_key_linear_bias_to_fp16 = const()[name = string("main_blocks_1_text_attn_W_key_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49743936)))];
            tensor<fp16, [?, ?, 512]> linear_13_cast_fp16 = linear(bias = main_blocks_1_text_attn_W_key_linear_bias_to_fp16, weight = main_blocks_1_text_attn_W_key_linear_weight_to_fp16, x = input_71_cast_fp16)[name = string("linear_13_cast_fp16")];
            tensor<int32, [4]> concat_15x = const()[name = string("concat_15x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_1008_cast_fp16 = reshape(shape = concat_15x, x = linear_13_cast_fp16)[name = string("op_1008_cast_fp16")];
            tensor<int32, [4]> x_89_perm_0 = const()[name = string("x_89_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [512, 256]> main_blocks_1_text_attn_W_value_linear_weight_to_fp16 = const()[name = string("main_blocks_1_text_attn_W_value_linear_weight_to_fp16"), val = tensor<fp16, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49745024)))];
            tensor<fp16, [512]> main_blocks_1_text_attn_W_value_linear_bias_to_fp16 = const()[name = string("main_blocks_1_text_attn_W_value_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50007232)))];
            tensor<fp16, [?, ?, 512]> linear_14_cast_fp16 = linear(bias = main_blocks_1_text_attn_W_value_linear_bias_to_fp16, weight = main_blocks_1_text_attn_W_value_linear_weight_to_fp16, x = input_71_cast_fp16)[name = string("linear_14_cast_fp16")];
            tensor<int32, [4]> concat_16x = const()[name = string("concat_16x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_1016_cast_fp16 = reshape(shape = concat_16x, x = linear_14_cast_fp16)[name = string("op_1016_cast_fp16")];
            tensor<int32, [4]> v_5_perm_0 = const()[name = string("v_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            int32 concat_17_values0_0 = const()[name = string("concat_17_values0_0"), val = int32(1)];
            int32 concat_17_values2_0 = const()[name = string("concat_17_values2_0"), val = int32(1)];
            int32 concat_17_axis_0 = const()[name = string("concat_17_axis_0"), val = int32(0)];
            bool concat_17_interleave_0 = const()[name = string("concat_17_interleave_0"), val = bool(false)];
            int32 gather_9_cast_uint16_to_int32 = cast(dtype = gather_9_cast_uint16_to_int32_dtype_0, x = gather_9_cast_uint16)[name = string("cast_90")];
            tensor<int32, [3]> concat_17 = concat(axis = concat_17_axis_0, interleave = concat_17_interleave_0, values = (concat_17_values0_0, gather_9_cast_uint16_to_int32, concat_17_values2_0))[name = string("concat_17")];
            tensor<int32, [3]> var_1019_begin_0 = const()[name = string("op_1019_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<bool, [3]> var_1019_end_mask_0 = const()[name = string("op_1019_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, ?, 1]> var_1019_cast_fp16 = slice_by_index(begin = var_1019_begin_0, end = concat_17, end_mask = var_1019_end_mask_0, x = main_blocks_0_text_attn_increments_to_fp16)[name = string("op_1019_cast_fp16")];
            tensor<int32, [3]> concat_18 = const()[name = string("concat_18"), val = tensor<int32, [3]>([2, -1, -1])];
            tensor<int32, [3]> shape_3_cast_fp16 = shape(x = var_1019_cast_fp16)[name = string("shape_3_cast_fp16")];
            tensor<bool, [3]> equal_3 = const()[name = string("equal_3"), val = tensor<bool, [3]>([false, true, true])];
            tensor<int32, [3]> select_3 = select(a = shape_3_cast_fp16, b = concat_18, cond = equal_3)[name = string("select_3")];
            tensor<int32, [3]> real_div_3 = real_div(x = select_3, y = shape_3_cast_fp16)[name = string("real_div_3")];
            tensor<fp16, [?, ?, ?]> var_1022_cast_fp16 = tile(reps = real_div_3, x = var_1019_cast_fp16)[name = string("op_1022_cast_fp16")];
            tensor<fp16, [2, ?, ?]> scaled_5_cast_fp16 = real_div(x = var_1022_cast_fp16, y = var_488_cast_fp16)[name = string("scaled_5_cast_fp16")];
            tensor<fp16, [1, 1, 32]> main_blocks_1_text_attn_theta_to_fp16 = const()[name = string("main_blocks_1_text_attn_theta_to_fp16"), val = tensor<fp16, [1, 1, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50008320)))];
            tensor<fp16, [2, ?, 32]> angles_5_cast_fp16 = mul(x = scaled_5_cast_fp16, y = main_blocks_1_text_attn_theta_to_fp16)[name = string("angles_5_cast_fp16")];
            tensor<fp16, [2, ?, 32]> var_1038_cast_fp16 = cos(x = angles_5_cast_fp16)[name = string("op_1038_cast_fp16")];
            tensor<int32, [1]> cos_5_axes_0 = const()[name = string("cos_5_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, ?, 32]> cos_5_cast_fp16 = expand_dims(axes = cos_5_axes_0, x = var_1038_cast_fp16)[name = string("cos_5_cast_fp16")];
            tensor<fp16, [2, ?, 32]> var_1040_cast_fp16 = sin(x = angles_5_cast_fp16)[name = string("op_1040_cast_fp16")];
            tensor<int32, [1]> sin_5_axes_0 = const()[name = string("sin_5_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, ?, 32]> sin_5_cast_fp16 = expand_dims(axes = sin_5_axes_0, x = var_1040_cast_fp16)[name = string("sin_5_cast_fp16")];
            tensor<int32, [4]> x_a_5_begin_0 = const()[name = string("x_a_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x_a_5_end_0 = const()[name = string("x_a_5_end_0"), val = tensor<int32, [4]>([2, 8, 0, 32])];
            tensor<bool, [4]> x_a_5_end_mask_0 = const()[name = string("x_a_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 8, ?, 64]> x_87_cast_fp16 = transpose(perm = x_87_perm_0, x = var_1000_cast_fp16)[name = string("transpose_77")];
            tensor<fp16, [2, 8, ?, 32]> x_a_5_cast_fp16 = slice_by_index(begin = x_a_5_begin_0, end = x_a_5_end_0, end_mask = x_a_5_end_mask_0, x = x_87_cast_fp16)[name = string("x_a_5_cast_fp16")];
            tensor<int32, [4]> x_b_5_begin_0 = const()[name = string("x_b_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x_b_5_end_0 = const()[name = string("x_b_5_end_0"), val = tensor<int32, [4]>([2, 8, 0, 64])];
            tensor<bool, [4]> x_b_5_end_mask_0 = const()[name = string("x_b_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [2, 8, ?, 32]> x_b_5_cast_fp16 = slice_by_index(begin = x_b_5_begin_0, end = x_b_5_end_0, end_mask = x_b_5_end_mask_0, x = x_87_cast_fp16)[name = string("x_b_5_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1044_cast_fp16 = mul(x = x_a_5_cast_fp16, y = cos_5_cast_fp16)[name = string("op_1044_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1045_cast_fp16 = mul(x = x_b_5_cast_fp16, y = sin_5_cast_fp16)[name = string("op_1045_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_a_5_cast_fp16 = sub(x = var_1044_cast_fp16, y = var_1045_cast_fp16)[name = string("rot_a_5_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1047_cast_fp16 = mul(x = x_a_5_cast_fp16, y = sin_5_cast_fp16)[name = string("op_1047_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1048_cast_fp16 = mul(x = x_b_5_cast_fp16, y = cos_5_cast_fp16)[name = string("op_1048_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_b_5_cast_fp16 = add(x = var_1047_cast_fp16, y = var_1048_cast_fp16)[name = string("rot_b_5_cast_fp16")];
            bool q_5_interleave_0 = const()[name = string("q_5_interleave_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> q_5_cast_fp16 = concat(axis = var_681, interleave = q_5_interleave_0, values = (rot_a_5_cast_fp16, rot_b_5_cast_fp16))[name = string("q_5_cast_fp16")];
            tensor<int32, [4]> x_a_7_begin_0 = const()[name = string("x_a_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x_a_7_end_0 = const()[name = string("x_a_7_end_0"), val = tensor<int32, [4]>([2, 8, 0, 32])];
            tensor<bool, [4]> x_a_7_end_mask_0 = const()[name = string("x_a_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 8, ?, 64]> x_89_cast_fp16 = transpose(perm = x_89_perm_0, x = var_1008_cast_fp16)[name = string("transpose_76")];
            tensor<fp16, [2, 8, ?, 32]> x_a_7_cast_fp16 = slice_by_index(begin = x_a_7_begin_0, end = x_a_7_end_0, end_mask = x_a_7_end_mask_0, x = x_89_cast_fp16)[name = string("x_a_7_cast_fp16")];
            tensor<int32, [4]> x_b_7_begin_0 = const()[name = string("x_b_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x_b_7_end_0 = const()[name = string("x_b_7_end_0"), val = tensor<int32, [4]>([2, 8, 0, 64])];
            tensor<bool, [4]> x_b_7_end_mask_0 = const()[name = string("x_b_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [2, 8, ?, 32]> x_b_7_cast_fp16 = slice_by_index(begin = x_b_7_begin_0, end = x_b_7_end_0, end_mask = x_b_7_end_mask_0, x = x_89_cast_fp16)[name = string("x_b_7_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1062_cast_fp16 = mul(x = x_a_7_cast_fp16, y = cos_3_cast_fp16)[name = string("op_1062_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1063_cast_fp16 = mul(x = x_b_7_cast_fp16, y = sin_3_cast_fp16)[name = string("op_1063_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_a_7_cast_fp16 = sub(x = var_1062_cast_fp16, y = var_1063_cast_fp16)[name = string("rot_a_7_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1065_cast_fp16 = mul(x = x_a_7_cast_fp16, y = sin_3_cast_fp16)[name = string("op_1065_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1066_cast_fp16 = mul(x = x_b_7_cast_fp16, y = cos_3_cast_fp16)[name = string("op_1066_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_b_7_cast_fp16 = add(x = var_1065_cast_fp16, y = var_1066_cast_fp16)[name = string("rot_b_7_cast_fp16")];
            bool k_5_interleave_0 = const()[name = string("k_5_interleave_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> k_5_cast_fp16 = concat(axis = var_681, interleave = k_5_interleave_0, values = (rot_a_7_cast_fp16, rot_b_7_cast_fp16))[name = string("k_5_cast_fp16")];
            bool var_1071_transpose_x_1 = const()[name = string("op_1071_transpose_x_1"), val = bool(false)];
            bool var_1071_transpose_y_1 = const()[name = string("op_1071_transpose_y_1"), val = bool(true)];
            tensor<fp16, [2, 8, ?, ?]> var_1071_cast_fp16 = matmul(transpose_x = var_1071_transpose_x_1, transpose_y = var_1071_transpose_y_1, x = q_5_cast_fp16, y = k_5_cast_fp16)[name = string("op_1071_cast_fp16")];
            fp16 _inversed_scores_3_y_0_to_fp16 = const()[name = string("_inversed_scores_3_y_0_to_fp16"), val = fp16(0x1p-4)];
            tensor<fp16, [2, 8, ?, ?]> _inversed_scores_3_cast_fp16 = mul(x = var_1071_cast_fp16, y = _inversed_scores_3_y_0_to_fp16)[name = string("_inversed_scores_3_cast_fp16")];
            tensor<fp16, [2, 8, ?, ?]> input_171_cast_fp16 = sub(x = _inversed_scores_3_cast_fp16, y = var_530_cast_fp16)[name = string("input_171_cast_fp16")];
            tensor<fp16, [2, 8, ?, ?]> attn_7_cast_fp16 = softmax(axis = var_681, x = input_171_cast_fp16)[name = string("attn_7_cast_fp16")];
            bool var_1080_transpose_x_0 = const()[name = string("op_1080_transpose_x_0"), val = bool(false)];
            bool var_1080_transpose_y_0 = const()[name = string("op_1080_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> v_5_cast_fp16 = transpose(perm = v_5_perm_0, x = var_1016_cast_fp16)[name = string("transpose_75")];
            tensor<fp16, [2, 8, ?, 64]> var_1080_cast_fp16 = matmul(transpose_x = var_1080_transpose_x_0, transpose_y = var_1080_transpose_y_0, x = attn_7_cast_fp16, y = v_5_cast_fp16)[name = string("op_1080_cast_fp16")];
            tensor<int32, [4]> var_1081_perm_0 = const()[name = string("op_1081_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_21x = const()[name = string("concat_21x"), val = tensor<int32, [3]>([2, -1, 512])];
            tensor<fp16, [2, ?, 8, 64]> var_1081_cast_fp16 = transpose(perm = var_1081_perm_0, x = var_1080_cast_fp16)[name = string("transpose_74")];
            tensor<fp16, [2, ?, 512]> input_173_cast_fp16 = reshape(shape = concat_21x, x = var_1081_cast_fp16)[name = string("input_173_cast_fp16")];
            tensor<fp16, [512, 512]> main_blocks_1_text_attn_out_fc_linear_weight_to_fp16 = const()[name = string("main_blocks_1_text_attn_out_fc_linear_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50008448)))];
            tensor<fp16, [512]> main_blocks_1_text_attn_out_fc_linear_bias_to_fp16 = const()[name = string("main_blocks_1_text_attn_out_fc_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50532800)))];
            tensor<fp16, [2, ?, 512]> linear_15_cast_fp16 = linear(bias = main_blocks_1_text_attn_out_fc_linear_bias_to_fp16, weight = main_blocks_1_text_attn_out_fc_linear_weight_to_fp16, x = input_173_cast_fp16)[name = string("linear_15_cast_fp16")];
            tensor<int32, [3]> out_27_perm_0 = const()[name = string("out_27_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> out_27_cast_fp16 = transpose(perm = out_27_perm_0, x = linear_15_cast_fp16)[name = string("transpose_73")];
            tensor<fp16, [2, 512, ?]> var_1089_cast_fp16 = mul(x = out_27_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_1089_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_91_cast_fp16 = add(x = x_m_5_cast_fp16, y = var_1089_cast_fp16)[name = string("x_91_cast_fp16")];
            tensor<int32, [3]> input_175_perm_0 = const()[name = string("input_175_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1096_axes_0 = const()[name = string("op_1096_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_1_text_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_1_text_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50533888)))];
            tensor<fp16, [512]> main_blocks_1_text_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_1_text_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50534976)))];
            tensor<fp16, [2, ?, 512]> input_175_cast_fp16 = transpose(perm = input_175_perm_0, x = x_91_cast_fp16)[name = string("transpose_72")];
            tensor<fp16, [2, ?, 512]> var_1096_cast_fp16 = layer_norm(axes = var_1096_axes_0, beta = main_blocks_1_text_norm_norm_bias_to_fp16, epsilon = var_686_to_fp16, gamma = main_blocks_1_text_norm_norm_weight_to_fp16, x = input_175_cast_fp16)[name = string("op_1096_cast_fp16")];
            tensor<int32, [3]> var_1097_perm_0 = const()[name = string("op_1097_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> var_1097_cast_fp16 = transpose(perm = var_1097_perm_0, x = var_1096_cast_fp16)[name = string("transpose_71")];
            tensor<fp16, [2, 512, ?]> x_93_cast_fp16 = mul(x = var_1097_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_93_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_95_cast_fp16 = mul(x = x_93_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_95_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1110_cast_fp16 = mul(x = x_95_cast_fp16, y = last_1_cast_fp16)[name = string("op_1110_cast_fp16")];
            tensor<int32, [1]> x_last_23_axes_0 = const()[name = string("x_last_23_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_23_keep_dims_0 = const()[name = string("x_last_23_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_23_cast_fp16 = reduce_sum(axes = x_last_23_axes_0, keep_dims = x_last_23_keep_dims_0, x = var_1110_cast_fp16)[name = string("x_last_23_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1113_cast_fp16 = mul(x = x_95_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_1113_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1115_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_23_cast_fp16)[name = string("op_1115_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_179_cast_fp16 = add(x = var_1113_cast_fp16, y = var_1115_cast_fp16)[name = string("input_179_cast_fp16")];
            tensor<int32, [6]> input_181_pad_0 = const()[name = string("input_181_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_181_mode_0 = const()[name = string("input_181_mode_0"), val = string("replicate")];
            fp16 const_24_to_fp16 = const()[name = string("const_24_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_181_cast_fp16 = pad(constant_val = const_24_to_fp16, mode = input_181_mode_0, pad = input_181_pad_0, x = input_179_cast_fp16)[name = string("input_181_cast_fp16")];
            string h_69_pad_type_0 = const()[name = string("h_69_pad_type_0"), val = string("valid")];
            int32 h_69_groups_0 = const()[name = string("h_69_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_69_strides_0 = const()[name = string("h_69_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_69_pad_0 = const()[name = string("h_69_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_69_dilations_0 = const()[name = string("h_69_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> main_blocks_1_convnext_2_0_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_2_0_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50536064)))];
            tensor<fp16, [512]> main_blocks_1_convnext_2_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_2_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50541248)))];
            tensor<fp16, [2, 512, ?]> h_69_cast_fp16 = conv(bias = main_blocks_1_convnext_2_0_dwconv__conv_bias_to_fp16, dilations = h_69_dilations_0, groups = h_69_groups_0, pad = h_69_pad_0, pad_type = h_69_pad_type_0, strides = h_69_strides_0, weight = main_blocks_1_convnext_2_0_dwconv__conv_weight_to_fp16, x = input_181_cast_fp16)[name = string("h_69_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_97_cast_fp16 = mul(x = h_69_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_97_cast_fp16")];
            tensor<int32, [3]> input_183_perm_0 = const()[name = string("input_183_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1133_axes_0 = const()[name = string("op_1133_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_1_convnext_2_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_2_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50542336)))];
            tensor<fp16, [512]> main_blocks_1_convnext_2_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_2_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50543424)))];
            tensor<fp16, [2, ?, 512]> input_183_cast_fp16 = transpose(perm = input_183_perm_0, x = x_97_cast_fp16)[name = string("transpose_70")];
            tensor<fp16, [2, ?, 512]> var_1133_cast_fp16 = layer_norm(axes = var_1133_axes_0, beta = main_blocks_1_convnext_2_0_norm_norm_bias_to_fp16, epsilon = var_686_to_fp16, gamma = main_blocks_1_convnext_2_0_norm_norm_weight_to_fp16, x = input_183_cast_fp16)[name = string("op_1133_cast_fp16")];
            tensor<int32, [3]> input_185_perm_0 = const()[name = string("input_185_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_71_pad_type_0 = const()[name = string("h_71_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_71_strides_0 = const()[name = string("h_71_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_71_pad_0 = const()[name = string("h_71_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_71_dilations_0 = const()[name = string("h_71_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_71_groups_0 = const()[name = string("h_71_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_1_convnext_2_0_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_2_0_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50544512)))];
            tensor<fp16, [2048]> main_blocks_1_convnext_2_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_2_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52641728)))];
            tensor<fp16, [2, 512, ?]> input_185_cast_fp16 = transpose(perm = input_185_perm_0, x = var_1133_cast_fp16)[name = string("transpose_69")];
            tensor<fp16, [2, 2048, ?]> h_71_cast_fp16 = conv(bias = main_blocks_1_convnext_2_0_pwconv1_bias_to_fp16, dilations = h_71_dilations_0, groups = h_71_groups_0, pad = h_71_pad_0, pad_type = h_71_pad_type_0, strides = h_71_strides_0, weight = main_blocks_1_convnext_2_0_pwconv1_weight_to_fp16, x = input_185_cast_fp16)[name = string("h_71_cast_fp16")];
            string input_187_mode_0 = const()[name = string("input_187_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_187_cast_fp16 = gelu(mode = input_187_mode_0, x = h_71_cast_fp16)[name = string("input_187_cast_fp16")];
            string h_73_pad_type_0 = const()[name = string("h_73_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_73_strides_0 = const()[name = string("h_73_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_73_pad_0 = const()[name = string("h_73_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_73_dilations_0 = const()[name = string("h_73_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_73_groups_0 = const()[name = string("h_73_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_1150_weight_0_to_fp16 = const()[name = string("op_1150_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52645888)))];
            tensor<fp16, [512]> var_1150_bias_0_to_fp16 = const()[name = string("op_1150_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54743104)))];
            tensor<fp16, [2, 512, ?]> var_1150_cast_fp16 = conv(bias = var_1150_bias_0_to_fp16, dilations = h_73_dilations_0, groups = h_73_groups_0, pad = h_73_pad_0, pad_type = h_73_pad_type_0, strides = h_73_strides_0, weight = var_1150_weight_0_to_fp16, x = input_187_cast_fp16)[name = string("op_1150_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_29_cast_fp16 = add(x = x_95_cast_fp16, y = var_1150_cast_fp16)[name = string("out_29_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_99_cast_fp16 = mul(x = out_29_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_99_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_m_7_cast_fp16 = mul(x = x_99_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_m_7_cast_fp16")];
            tensor<int32, [3]> input_189_perm_0 = const()[name = string("input_189_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [256, 512]> main_blocks_1_style_attn_W_query_linear_weight_to_fp16 = const()[name = string("main_blocks_1_style_attn_W_query_linear_weight_to_fp16"), val = tensor<fp16, [256, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54744192)))];
            tensor<fp16, [256]> main_blocks_1_style_attn_W_query_linear_bias_to_fp16 = const()[name = string("main_blocks_1_style_attn_W_query_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55006400)))];
            tensor<fp16, [2, ?, 512]> input_189_cast_fp16 = transpose(perm = input_189_perm_0, x = x_99_cast_fp16)[name = string("transpose_68")];
            tensor<fp16, [2, ?, 256]> linear_16_cast_fp16 = linear(bias = main_blocks_1_style_attn_W_query_linear_bias_to_fp16, weight = main_blocks_1_style_attn_W_query_linear_weight_to_fp16, x = input_189_cast_fp16)[name = string("linear_16_cast_fp16")];
            tensor<int32, [4]> concat_22x = const()[name = string("concat_22x"), val = tensor<int32, [4]>([2, -1, 2, 128])];
            tensor<fp16, [2, ?, 2, 128]> var_1167_cast_fp16 = reshape(shape = concat_22x, x = linear_16_cast_fp16)[name = string("op_1167_cast_fp16")];
            tensor<int32, [4]> q_7_perm_0 = const()[name = string("q_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [256, 256]> main_blocks_1_style_attn_W_value_linear_weight_to_fp16 = const()[name = string("main_blocks_1_style_attn_W_value_linear_weight_to_fp16"), val = tensor<fp16, [256, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55006976)))];
            tensor<fp16, [256]> main_blocks_1_style_attn_W_value_linear_bias_to_fp16 = const()[name = string("main_blocks_1_style_attn_W_value_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55138112)))];
            tensor<fp16, [2, 50, 256]> linear_18_cast_fp16 = linear(bias = main_blocks_1_style_attn_W_value_linear_bias_to_fp16, weight = main_blocks_1_style_attn_W_value_linear_weight_to_fp16, x = input_95_cast_fp16)[name = string("linear_18_cast_fp16")];
            tensor<int32, [4]> var_1180 = const()[name = string("op_1180"), val = tensor<int32, [4]>([2, 50, 2, 128])];
            tensor<fp16, [2, 50, 2, 128]> var_1181_cast_fp16 = reshape(shape = var_1180, x = linear_18_cast_fp16)[name = string("op_1181_cast_fp16")];
            tensor<int32, [4]> v_7_perm_0 = const()[name = string("v_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_1185_transpose_x_0 = const()[name = string("op_1185_transpose_x_0"), val = bool(false)];
            bool var_1185_transpose_y_0 = const()[name = string("op_1185_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 2, 128, 50]> var_1184_to_fp16 = const()[name = string("op_1184_to_fp16"), val = tensor<fp16, [2, 2, 128, 50]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55138688)))];
            tensor<fp16, [2, 2, ?, 128]> q_7_cast_fp16 = transpose(perm = q_7_perm_0, x = var_1167_cast_fp16)[name = string("transpose_67")];
            tensor<fp16, [2, 2, ?, 50]> var_1185_cast_fp16 = matmul(transpose_x = var_1185_transpose_x_0, transpose_y = var_1185_transpose_y_0, x = q_7_cast_fp16, y = var_1184_to_fp16)[name = string("op_1185_cast_fp16")];
            fp16 _inversed_input_191_y_0_to_fp16 = const()[name = string("_inversed_input_191_y_0_to_fp16"), val = fp16(0x1p-4)];
            tensor<fp16, [2, 2, ?, 50]> _inversed_input_191_cast_fp16 = mul(x = var_1185_cast_fp16, y = _inversed_input_191_y_0_to_fp16)[name = string("_inversed_input_191_cast_fp16")];
            tensor<fp16, [2, 2, ?, 50]> attn_9_cast_fp16 = softmax(axis = var_681, x = _inversed_input_191_cast_fp16)[name = string("attn_9_cast_fp16")];
            tensor<fp16, [2, 2, ?, 50]> attn_11_cast_fp16 = mul(x = attn_9_cast_fp16, y = mask_1_cast_fp16)[name = string("attn_11_cast_fp16")];
            bool var_1192_transpose_x_0 = const()[name = string("op_1192_transpose_x_0"), val = bool(false)];
            bool var_1192_transpose_y_0 = const()[name = string("op_1192_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 2, 50, 128]> v_7_cast_fp16 = transpose(perm = v_7_perm_0, x = var_1181_cast_fp16)[name = string("transpose_66")];
            tensor<fp16, [2, 2, ?, 128]> var_1192_cast_fp16 = matmul(transpose_x = var_1192_transpose_x_0, transpose_y = var_1192_transpose_y_0, x = attn_11_cast_fp16, y = v_7_cast_fp16)[name = string("op_1192_cast_fp16")];
            tensor<int32, [4]> var_1193_perm_0 = const()[name = string("op_1193_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_23x = const()[name = string("concat_23x"), val = tensor<int32, [3]>([2, -1, 256])];
            tensor<fp16, [2, ?, 2, 128]> var_1193_cast_fp16 = transpose(perm = var_1193_perm_0, x = var_1192_cast_fp16)[name = string("transpose_65")];
            tensor<fp16, [2, ?, 256]> input_193_cast_fp16 = reshape(shape = concat_23x, x = var_1193_cast_fp16)[name = string("input_193_cast_fp16")];
            tensor<fp16, [512, 256]> main_blocks_1_style_attn_out_fc_linear_weight_to_fp16 = const()[name = string("main_blocks_1_style_attn_out_fc_linear_weight_to_fp16"), val = tensor<fp16, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55189952)))];
            tensor<fp16, [512]> main_blocks_1_style_attn_out_fc_linear_bias_to_fp16 = const()[name = string("main_blocks_1_style_attn_out_fc_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55452160)))];
            tensor<fp16, [2, ?, 512]> linear_19_cast_fp16 = linear(bias = main_blocks_1_style_attn_out_fc_linear_bias_to_fp16, weight = main_blocks_1_style_attn_out_fc_linear_weight_to_fp16, x = input_193_cast_fp16)[name = string("linear_19_cast_fp16")];
            tensor<int32, [3]> out_31_perm_0 = const()[name = string("out_31_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> out_31_cast_fp16 = transpose(perm = out_31_perm_0, x = linear_19_cast_fp16)[name = string("transpose_64")];
            tensor<fp16, [2, 512, ?]> var_1201_cast_fp16 = mul(x = out_31_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_1201_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_101_cast_fp16 = add(x = x_m_7_cast_fp16, y = var_1201_cast_fp16)[name = string("x_101_cast_fp16")];
            tensor<int32, [3]> input_195_perm_0 = const()[name = string("input_195_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1208_axes_0 = const()[name = string("op_1208_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_1_style_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_1_style_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55453248)))];
            tensor<fp16, [512]> main_blocks_1_style_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_1_style_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55454336)))];
            tensor<fp16, [2, ?, 512]> input_195_cast_fp16 = transpose(perm = input_195_perm_0, x = x_101_cast_fp16)[name = string("transpose_63")];
            tensor<fp16, [2, ?, 512]> var_1208_cast_fp16 = layer_norm(axes = var_1208_axes_0, beta = main_blocks_1_style_norm_norm_bias_to_fp16, epsilon = var_686_to_fp16, gamma = main_blocks_1_style_norm_norm_weight_to_fp16, x = input_195_cast_fp16)[name = string("op_1208_cast_fp16")];
            tensor<int32, [3]> var_1209_perm_0 = const()[name = string("op_1209_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> var_1209_cast_fp16 = transpose(perm = var_1209_perm_0, x = var_1208_cast_fp16)[name = string("transpose_62")];
            tensor<fp16, [2, 512, ?]> x_103_cast_fp16 = mul(x = var_1209_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_103_cast_fp16")];
            int32 var_1228 = const()[name = string("op_1228"), val = int32(-1)];
            tensor<fp16, [2, 512, ?]> x_105_cast_fp16 = mul(x = x_103_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_105_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1263_cast_fp16 = mul(x = x_105_cast_fp16, y = last_1_cast_fp16)[name = string("op_1263_cast_fp16")];
            tensor<int32, [1]> x_last_25_axes_0 = const()[name = string("x_last_25_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_25_keep_dims_0 = const()[name = string("x_last_25_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_25_cast_fp16 = reduce_sum(axes = x_last_25_axes_0, keep_dims = x_last_25_keep_dims_0, x = var_1263_cast_fp16)[name = string("x_last_25_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1266_cast_fp16 = mul(x = x_105_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_1266_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1268_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_25_cast_fp16)[name = string("op_1268_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_199_cast_fp16 = add(x = var_1266_cast_fp16, y = var_1268_cast_fp16)[name = string("input_199_cast_fp16")];
            tensor<int32, [6]> input_201_pad_0 = const()[name = string("input_201_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_201_mode_0 = const()[name = string("input_201_mode_0"), val = string("replicate")];
            fp16 const_27_to_fp16 = const()[name = string("const_27_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_201_cast_fp16 = pad(constant_val = const_27_to_fp16, mode = input_201_mode_0, pad = input_201_pad_0, x = input_199_cast_fp16)[name = string("input_201_cast_fp16")];
            string h_75_pad_type_0 = const()[name = string("h_75_pad_type_0"), val = string("valid")];
            int32 h_75_groups_0 = const()[name = string("h_75_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_75_strides_0 = const()[name = string("h_75_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_75_pad_0 = const()[name = string("h_75_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_75_dilations_0 = const()[name = string("h_75_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> main_blocks_2_convnext_0_0_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_0_0_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55455424)))];
            tensor<fp16, [512]> main_blocks_2_convnext_0_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55460608)))];
            tensor<fp16, [2, 512, ?]> h_75_cast_fp16 = conv(bias = main_blocks_2_convnext_0_0_dwconv__conv_bias_to_fp16, dilations = h_75_dilations_0, groups = h_75_groups_0, pad = h_75_pad_0, pad_type = h_75_pad_type_0, strides = h_75_strides_0, weight = main_blocks_2_convnext_0_0_dwconv__conv_weight_to_fp16, x = input_201_cast_fp16)[name = string("h_75_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_107_cast_fp16 = mul(x = h_75_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_107_cast_fp16")];
            tensor<int32, [3]> input_203_perm_0 = const()[name = string("input_203_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1286_axes_0 = const()[name = string("op_1286_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_2_convnext_0_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_0_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55461696)))];
            tensor<fp16, [512]> main_blocks_2_convnext_0_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55462784)))];
            fp16 var_1233_to_fp16 = const()[name = string("op_1233_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [2, ?, 512]> input_203_cast_fp16 = transpose(perm = input_203_perm_0, x = x_107_cast_fp16)[name = string("transpose_61")];
            tensor<fp16, [2, ?, 512]> var_1286_cast_fp16 = layer_norm(axes = var_1286_axes_0, beta = main_blocks_2_convnext_0_0_norm_norm_bias_to_fp16, epsilon = var_1233_to_fp16, gamma = main_blocks_2_convnext_0_0_norm_norm_weight_to_fp16, x = input_203_cast_fp16)[name = string("op_1286_cast_fp16")];
            tensor<int32, [3]> input_205_perm_0 = const()[name = string("input_205_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_77_pad_type_0 = const()[name = string("h_77_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_77_strides_0 = const()[name = string("h_77_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_77_pad_0 = const()[name = string("h_77_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_77_dilations_0 = const()[name = string("h_77_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_77_groups_0 = const()[name = string("h_77_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_2_convnext_0_0_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_0_0_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55463872)))];
            tensor<fp16, [2048]> main_blocks_2_convnext_0_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57561088)))];
            tensor<fp16, [2, 512, ?]> input_205_cast_fp16 = transpose(perm = input_205_perm_0, x = var_1286_cast_fp16)[name = string("transpose_60")];
            tensor<fp16, [2, 2048, ?]> h_77_cast_fp16 = conv(bias = main_blocks_2_convnext_0_0_pwconv1_bias_to_fp16, dilations = h_77_dilations_0, groups = h_77_groups_0, pad = h_77_pad_0, pad_type = h_77_pad_type_0, strides = h_77_strides_0, weight = main_blocks_2_convnext_0_0_pwconv1_weight_to_fp16, x = input_205_cast_fp16)[name = string("h_77_cast_fp16")];
            string input_207_mode_0 = const()[name = string("input_207_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_207_cast_fp16 = gelu(mode = input_207_mode_0, x = h_77_cast_fp16)[name = string("input_207_cast_fp16")];
            string h_79_pad_type_0 = const()[name = string("h_79_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_79_strides_0 = const()[name = string("h_79_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_79_pad_0 = const()[name = string("h_79_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_79_dilations_0 = const()[name = string("h_79_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_79_groups_0 = const()[name = string("h_79_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_1303_weight_0_to_fp16 = const()[name = string("op_1303_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57565248)))];
            tensor<fp16, [512]> var_1303_bias_0_to_fp16 = const()[name = string("op_1303_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59662464)))];
            tensor<fp16, [2, 512, ?]> var_1303_cast_fp16 = conv(bias = var_1303_bias_0_to_fp16, dilations = h_79_dilations_0, groups = h_79_groups_0, pad = h_79_pad_0, pad_type = h_79_pad_type_0, strides = h_79_strides_0, weight = var_1303_weight_0_to_fp16, x = input_207_cast_fp16)[name = string("op_1303_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_33_cast_fp16 = add(x = x_105_cast_fp16, y = var_1303_cast_fp16)[name = string("out_33_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_109_cast_fp16 = mul(x = out_33_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_109_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_111_cast_fp16 = mul(x = x_109_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_111_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1317_cast_fp16 = mul(x = x_111_cast_fp16, y = last_1_cast_fp16)[name = string("op_1317_cast_fp16")];
            tensor<int32, [1]> x_last_27_axes_0 = const()[name = string("x_last_27_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_27_keep_dims_0 = const()[name = string("x_last_27_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_27_cast_fp16 = reduce_sum(axes = x_last_27_axes_0, keep_dims = x_last_27_keep_dims_0, x = var_1317_cast_fp16)[name = string("x_last_27_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1320_cast_fp16 = mul(x = x_111_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_1320_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1322_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_27_cast_fp16)[name = string("op_1322_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_211_cast_fp16 = add(x = var_1320_cast_fp16, y = var_1322_cast_fp16)[name = string("input_211_cast_fp16")];
            tensor<int32, [6]> input_213_pad_0 = const()[name = string("input_213_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_213_mode_0 = const()[name = string("input_213_mode_0"), val = string("replicate")];
            fp16 const_29_to_fp16 = const()[name = string("const_29_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_213_cast_fp16 = pad(constant_val = const_29_to_fp16, mode = input_213_mode_0, pad = input_213_pad_0, x = input_211_cast_fp16)[name = string("input_213_cast_fp16")];
            string h_81_pad_type_0 = const()[name = string("h_81_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_81_dilations_0 = const()[name = string("h_81_dilations_0"), val = tensor<int32, [1]>([2])];
            int32 h_81_groups_0 = const()[name = string("h_81_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_81_strides_0 = const()[name = string("h_81_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_81_pad_0 = const()[name = string("h_81_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [512, 1, 5]> main_blocks_2_convnext_0_1_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_0_1_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59663552)))];
            tensor<fp16, [512]> main_blocks_2_convnext_0_1_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_1_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59668736)))];
            tensor<fp16, [2, 512, ?]> h_81_cast_fp16 = conv(bias = main_blocks_2_convnext_0_1_dwconv__conv_bias_to_fp16, dilations = h_81_dilations_0, groups = h_81_groups_0, pad = h_81_pad_0, pad_type = h_81_pad_type_0, strides = h_81_strides_0, weight = main_blocks_2_convnext_0_1_dwconv__conv_weight_to_fp16, x = input_213_cast_fp16)[name = string("h_81_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_113_cast_fp16 = mul(x = h_81_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_113_cast_fp16")];
            tensor<int32, [3]> input_215_perm_0 = const()[name = string("input_215_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1340_axes_0 = const()[name = string("op_1340_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_2_convnext_0_1_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_0_1_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59669824)))];
            tensor<fp16, [512]> main_blocks_2_convnext_0_1_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_1_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59670912)))];
            tensor<fp16, [2, ?, 512]> input_215_cast_fp16 = transpose(perm = input_215_perm_0, x = x_113_cast_fp16)[name = string("transpose_59")];
            tensor<fp16, [2, ?, 512]> var_1340_cast_fp16 = layer_norm(axes = var_1340_axes_0, beta = main_blocks_2_convnext_0_1_norm_norm_bias_to_fp16, epsilon = var_1233_to_fp16, gamma = main_blocks_2_convnext_0_1_norm_norm_weight_to_fp16, x = input_215_cast_fp16)[name = string("op_1340_cast_fp16")];
            tensor<int32, [3]> input_217_perm_0 = const()[name = string("input_217_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_83_pad_type_0 = const()[name = string("h_83_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_83_strides_0 = const()[name = string("h_83_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_83_pad_0 = const()[name = string("h_83_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_83_dilations_0 = const()[name = string("h_83_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_83_groups_0 = const()[name = string("h_83_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_2_convnext_0_1_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_0_1_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59672000)))];
            tensor<fp16, [2048]> main_blocks_2_convnext_0_1_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_1_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61769216)))];
            tensor<fp16, [2, 512, ?]> input_217_cast_fp16 = transpose(perm = input_217_perm_0, x = var_1340_cast_fp16)[name = string("transpose_58")];
            tensor<fp16, [2, 2048, ?]> h_83_cast_fp16 = conv(bias = main_blocks_2_convnext_0_1_pwconv1_bias_to_fp16, dilations = h_83_dilations_0, groups = h_83_groups_0, pad = h_83_pad_0, pad_type = h_83_pad_type_0, strides = h_83_strides_0, weight = main_blocks_2_convnext_0_1_pwconv1_weight_to_fp16, x = input_217_cast_fp16)[name = string("h_83_cast_fp16")];
            string input_219_mode_0 = const()[name = string("input_219_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_219_cast_fp16 = gelu(mode = input_219_mode_0, x = h_83_cast_fp16)[name = string("input_219_cast_fp16")];
            string h_85_pad_type_0 = const()[name = string("h_85_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_85_strides_0 = const()[name = string("h_85_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_85_pad_0 = const()[name = string("h_85_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_85_dilations_0 = const()[name = string("h_85_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_85_groups_0 = const()[name = string("h_85_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_1357_weight_0_to_fp16 = const()[name = string("op_1357_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61773376)))];
            tensor<fp16, [512]> var_1357_bias_0_to_fp16 = const()[name = string("op_1357_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63870592)))];
            tensor<fp16, [2, 512, ?]> var_1357_cast_fp16 = conv(bias = var_1357_bias_0_to_fp16, dilations = h_85_dilations_0, groups = h_85_groups_0, pad = h_85_pad_0, pad_type = h_85_pad_type_0, strides = h_85_strides_0, weight = var_1357_weight_0_to_fp16, x = input_219_cast_fp16)[name = string("op_1357_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_35_cast_fp16 = add(x = x_111_cast_fp16, y = var_1357_cast_fp16)[name = string("out_35_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_115_cast_fp16 = mul(x = out_35_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_115_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_117_cast_fp16 = mul(x = x_115_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_117_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1371_cast_fp16 = mul(x = x_117_cast_fp16, y = last_1_cast_fp16)[name = string("op_1371_cast_fp16")];
            tensor<int32, [1]> x_last_29_axes_0 = const()[name = string("x_last_29_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_29_keep_dims_0 = const()[name = string("x_last_29_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_29_cast_fp16 = reduce_sum(axes = x_last_29_axes_0, keep_dims = x_last_29_keep_dims_0, x = var_1371_cast_fp16)[name = string("x_last_29_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1374_cast_fp16 = mul(x = x_117_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_1374_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1376_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_29_cast_fp16)[name = string("op_1376_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_223_cast_fp16 = add(x = var_1374_cast_fp16, y = var_1376_cast_fp16)[name = string("input_223_cast_fp16")];
            tensor<int32, [6]> input_225_pad_0 = const()[name = string("input_225_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 8, 8])];
            string input_225_mode_0 = const()[name = string("input_225_mode_0"), val = string("replicate")];
            fp16 const_31_to_fp16 = const()[name = string("const_31_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_225_cast_fp16 = pad(constant_val = const_31_to_fp16, mode = input_225_mode_0, pad = input_225_pad_0, x = input_223_cast_fp16)[name = string("input_225_cast_fp16")];
            string h_87_pad_type_0 = const()[name = string("h_87_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_87_dilations_0 = const()[name = string("h_87_dilations_0"), val = tensor<int32, [1]>([4])];
            int32 h_87_groups_0 = const()[name = string("h_87_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_87_strides_0 = const()[name = string("h_87_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_87_pad_0 = const()[name = string("h_87_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [512, 1, 5]> main_blocks_2_convnext_0_2_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_0_2_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63871680)))];
            tensor<fp16, [512]> main_blocks_2_convnext_0_2_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_2_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63876864)))];
            tensor<fp16, [2, 512, ?]> h_87_cast_fp16 = conv(bias = main_blocks_2_convnext_0_2_dwconv__conv_bias_to_fp16, dilations = h_87_dilations_0, groups = h_87_groups_0, pad = h_87_pad_0, pad_type = h_87_pad_type_0, strides = h_87_strides_0, weight = main_blocks_2_convnext_0_2_dwconv__conv_weight_to_fp16, x = input_225_cast_fp16)[name = string("h_87_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_119_cast_fp16 = mul(x = h_87_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_119_cast_fp16")];
            tensor<int32, [3]> input_227_perm_0 = const()[name = string("input_227_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1394_axes_0 = const()[name = string("op_1394_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_2_convnext_0_2_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_0_2_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63877952)))];
            tensor<fp16, [512]> main_blocks_2_convnext_0_2_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_2_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63879040)))];
            tensor<fp16, [2, ?, 512]> input_227_cast_fp16 = transpose(perm = input_227_perm_0, x = x_119_cast_fp16)[name = string("transpose_57")];
            tensor<fp16, [2, ?, 512]> var_1394_cast_fp16 = layer_norm(axes = var_1394_axes_0, beta = main_blocks_2_convnext_0_2_norm_norm_bias_to_fp16, epsilon = var_1233_to_fp16, gamma = main_blocks_2_convnext_0_2_norm_norm_weight_to_fp16, x = input_227_cast_fp16)[name = string("op_1394_cast_fp16")];
            tensor<int32, [3]> input_229_perm_0 = const()[name = string("input_229_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_89_pad_type_0 = const()[name = string("h_89_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_89_strides_0 = const()[name = string("h_89_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_89_pad_0 = const()[name = string("h_89_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_89_dilations_0 = const()[name = string("h_89_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_89_groups_0 = const()[name = string("h_89_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_2_convnext_0_2_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_0_2_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63880128)))];
            tensor<fp16, [2048]> main_blocks_2_convnext_0_2_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_2_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65977344)))];
            tensor<fp16, [2, 512, ?]> input_229_cast_fp16 = transpose(perm = input_229_perm_0, x = var_1394_cast_fp16)[name = string("transpose_56")];
            tensor<fp16, [2, 2048, ?]> h_89_cast_fp16 = conv(bias = main_blocks_2_convnext_0_2_pwconv1_bias_to_fp16, dilations = h_89_dilations_0, groups = h_89_groups_0, pad = h_89_pad_0, pad_type = h_89_pad_type_0, strides = h_89_strides_0, weight = main_blocks_2_convnext_0_2_pwconv1_weight_to_fp16, x = input_229_cast_fp16)[name = string("h_89_cast_fp16")];
            string input_231_mode_0 = const()[name = string("input_231_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_231_cast_fp16 = gelu(mode = input_231_mode_0, x = h_89_cast_fp16)[name = string("input_231_cast_fp16")];
            string h_91_pad_type_0 = const()[name = string("h_91_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_91_strides_0 = const()[name = string("h_91_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_91_pad_0 = const()[name = string("h_91_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_91_dilations_0 = const()[name = string("h_91_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_91_groups_0 = const()[name = string("h_91_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_1411_weight_0_to_fp16 = const()[name = string("op_1411_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65981504)))];
            tensor<fp16, [512]> var_1411_bias_0_to_fp16 = const()[name = string("op_1411_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68078720)))];
            tensor<fp16, [2, 512, ?]> var_1411_cast_fp16 = conv(bias = var_1411_bias_0_to_fp16, dilations = h_91_dilations_0, groups = h_91_groups_0, pad = h_91_pad_0, pad_type = h_91_pad_type_0, strides = h_91_strides_0, weight = var_1411_weight_0_to_fp16, x = input_231_cast_fp16)[name = string("op_1411_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_37_cast_fp16 = add(x = x_117_cast_fp16, y = var_1411_cast_fp16)[name = string("out_37_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_121_cast_fp16 = mul(x = out_37_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_121_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_123_cast_fp16 = mul(x = x_121_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_123_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1425_cast_fp16 = mul(x = x_123_cast_fp16, y = last_1_cast_fp16)[name = string("op_1425_cast_fp16")];
            tensor<int32, [1]> x_last_31_axes_0 = const()[name = string("x_last_31_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_31_keep_dims_0 = const()[name = string("x_last_31_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_31_cast_fp16 = reduce_sum(axes = x_last_31_axes_0, keep_dims = x_last_31_keep_dims_0, x = var_1425_cast_fp16)[name = string("x_last_31_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1428_cast_fp16 = mul(x = x_123_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_1428_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1430_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_31_cast_fp16)[name = string("op_1430_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_235_cast_fp16 = add(x = var_1428_cast_fp16, y = var_1430_cast_fp16)[name = string("input_235_cast_fp16")];
            tensor<int32, [6]> input_237_pad_0 = const()[name = string("input_237_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 16, 16])];
            string input_237_mode_0 = const()[name = string("input_237_mode_0"), val = string("replicate")];
            fp16 const_33_to_fp16 = const()[name = string("const_33_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_237_cast_fp16 = pad(constant_val = const_33_to_fp16, mode = input_237_mode_0, pad = input_237_pad_0, x = input_235_cast_fp16)[name = string("input_237_cast_fp16")];
            string h_93_pad_type_0 = const()[name = string("h_93_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_93_dilations_0 = const()[name = string("h_93_dilations_0"), val = tensor<int32, [1]>([8])];
            int32 h_93_groups_0 = const()[name = string("h_93_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_93_strides_0 = const()[name = string("h_93_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_93_pad_0 = const()[name = string("h_93_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [512, 1, 5]> main_blocks_2_convnext_0_3_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_0_3_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68079808)))];
            tensor<fp16, [512]> main_blocks_2_convnext_0_3_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_3_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68084992)))];
            tensor<fp16, [2, 512, ?]> h_93_cast_fp16 = conv(bias = main_blocks_2_convnext_0_3_dwconv__conv_bias_to_fp16, dilations = h_93_dilations_0, groups = h_93_groups_0, pad = h_93_pad_0, pad_type = h_93_pad_type_0, strides = h_93_strides_0, weight = main_blocks_2_convnext_0_3_dwconv__conv_weight_to_fp16, x = input_237_cast_fp16)[name = string("h_93_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_125_cast_fp16 = mul(x = h_93_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_125_cast_fp16")];
            tensor<int32, [3]> input_239_perm_0 = const()[name = string("input_239_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1448_axes_0 = const()[name = string("op_1448_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_2_convnext_0_3_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_0_3_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68086080)))];
            tensor<fp16, [512]> main_blocks_2_convnext_0_3_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_3_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68087168)))];
            tensor<fp16, [2, ?, 512]> input_239_cast_fp16 = transpose(perm = input_239_perm_0, x = x_125_cast_fp16)[name = string("transpose_55")];
            tensor<fp16, [2, ?, 512]> var_1448_cast_fp16 = layer_norm(axes = var_1448_axes_0, beta = main_blocks_2_convnext_0_3_norm_norm_bias_to_fp16, epsilon = var_1233_to_fp16, gamma = main_blocks_2_convnext_0_3_norm_norm_weight_to_fp16, x = input_239_cast_fp16)[name = string("op_1448_cast_fp16")];
            tensor<int32, [3]> input_241_perm_0 = const()[name = string("input_241_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_95_pad_type_0 = const()[name = string("h_95_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_95_strides_0 = const()[name = string("h_95_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_95_pad_0 = const()[name = string("h_95_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_95_dilations_0 = const()[name = string("h_95_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_95_groups_0 = const()[name = string("h_95_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_2_convnext_0_3_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_0_3_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68088256)))];
            tensor<fp16, [2048]> main_blocks_2_convnext_0_3_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_3_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70185472)))];
            tensor<fp16, [2, 512, ?]> input_241_cast_fp16 = transpose(perm = input_241_perm_0, x = var_1448_cast_fp16)[name = string("transpose_54")];
            tensor<fp16, [2, 2048, ?]> h_95_cast_fp16 = conv(bias = main_blocks_2_convnext_0_3_pwconv1_bias_to_fp16, dilations = h_95_dilations_0, groups = h_95_groups_0, pad = h_95_pad_0, pad_type = h_95_pad_type_0, strides = h_95_strides_0, weight = main_blocks_2_convnext_0_3_pwconv1_weight_to_fp16, x = input_241_cast_fp16)[name = string("h_95_cast_fp16")];
            string input_243_mode_0 = const()[name = string("input_243_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_243_cast_fp16 = gelu(mode = input_243_mode_0, x = h_95_cast_fp16)[name = string("input_243_cast_fp16")];
            string h_97_pad_type_0 = const()[name = string("h_97_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_97_strides_0 = const()[name = string("h_97_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_97_pad_0 = const()[name = string("h_97_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_97_dilations_0 = const()[name = string("h_97_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_97_groups_0 = const()[name = string("h_97_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_1465_weight_0_to_fp16 = const()[name = string("op_1465_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70189632)))];
            tensor<fp16, [512]> var_1465_bias_0_to_fp16 = const()[name = string("op_1465_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72286848)))];
            tensor<fp16, [2, 512, ?]> var_1465_cast_fp16 = conv(bias = var_1465_bias_0_to_fp16, dilations = h_97_dilations_0, groups = h_97_groups_0, pad = h_97_pad_0, pad_type = h_97_pad_type_0, strides = h_97_strides_0, weight = var_1465_weight_0_to_fp16, x = input_243_cast_fp16)[name = string("op_1465_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_39_cast_fp16 = add(x = x_123_cast_fp16, y = var_1465_cast_fp16)[name = string("out_39_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_127_cast_fp16 = mul(x = out_39_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_127_cast_fp16")];
            tensor<fp16, [512, 64]> main_blocks_2_time_cond_linear_linear_weight_to_fp16 = const()[name = string("main_blocks_2_time_cond_linear_linear_weight_to_fp16"), val = tensor<fp16, [512, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72287936)))];
            tensor<fp16, [512]> main_blocks_2_time_cond_linear_linear_bias_to_fp16 = const()[name = string("main_blocks_2_time_cond_linear_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72353536)))];
            tensor<fp16, [2, 512]> linear_20_cast_fp16 = linear(bias = main_blocks_2_time_cond_linear_linear_bias_to_fp16, weight = main_blocks_2_time_cond_linear_linear_weight_to_fp16, x = input_55_cast_fp16)[name = string("linear_20_cast_fp16")];
            tensor<int32, [1]> t_9_axes_0 = const()[name = string("t_9_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2, 512, 1]> t_9_cast_fp16 = expand_dims(axes = t_9_axes_0, x = linear_20_cast_fp16)[name = string("t_9_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1475_cast_fp16 = add(x = x_127_cast_fp16, y = t_9_cast_fp16)[name = string("op_1475_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_129_cast_fp16 = mul(x = var_1475_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_129_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_131_cast_fp16 = mul(x = x_129_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_131_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1488_cast_fp16 = mul(x = x_131_cast_fp16, y = last_1_cast_fp16)[name = string("op_1488_cast_fp16")];
            tensor<int32, [1]> x_last_33_axes_0 = const()[name = string("x_last_33_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_33_keep_dims_0 = const()[name = string("x_last_33_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_33_cast_fp16 = reduce_sum(axes = x_last_33_axes_0, keep_dims = x_last_33_keep_dims_0, x = var_1488_cast_fp16)[name = string("x_last_33_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1491_cast_fp16 = mul(x = x_131_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_1491_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1493_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_33_cast_fp16)[name = string("op_1493_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_249_cast_fp16 = add(x = var_1491_cast_fp16, y = var_1493_cast_fp16)[name = string("input_249_cast_fp16")];
            tensor<int32, [6]> input_251_pad_0 = const()[name = string("input_251_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_251_mode_0 = const()[name = string("input_251_mode_0"), val = string("replicate")];
            fp16 const_35_to_fp16 = const()[name = string("const_35_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_251_cast_fp16 = pad(constant_val = const_35_to_fp16, mode = input_251_mode_0, pad = input_251_pad_0, x = input_249_cast_fp16)[name = string("input_251_cast_fp16")];
            string h_99_pad_type_0 = const()[name = string("h_99_pad_type_0"), val = string("valid")];
            int32 h_99_groups_0 = const()[name = string("h_99_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_99_strides_0 = const()[name = string("h_99_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_99_pad_0 = const()[name = string("h_99_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_99_dilations_0 = const()[name = string("h_99_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> main_blocks_2_convnext_1_0_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_1_0_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72354624)))];
            tensor<fp16, [512]> main_blocks_2_convnext_1_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_1_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72359808)))];
            tensor<fp16, [2, 512, ?]> h_99_cast_fp16 = conv(bias = main_blocks_2_convnext_1_0_dwconv__conv_bias_to_fp16, dilations = h_99_dilations_0, groups = h_99_groups_0, pad = h_99_pad_0, pad_type = h_99_pad_type_0, strides = h_99_strides_0, weight = main_blocks_2_convnext_1_0_dwconv__conv_weight_to_fp16, x = input_251_cast_fp16)[name = string("h_99_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_133_cast_fp16 = mul(x = h_99_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_133_cast_fp16")];
            tensor<int32, [3]> input_253_perm_0 = const()[name = string("input_253_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1511_axes_0 = const()[name = string("op_1511_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_2_convnext_1_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_1_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72360896)))];
            tensor<fp16, [512]> main_blocks_2_convnext_1_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_1_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72361984)))];
            tensor<fp16, [2, ?, 512]> input_253_cast_fp16 = transpose(perm = input_253_perm_0, x = x_133_cast_fp16)[name = string("transpose_53")];
            tensor<fp16, [2, ?, 512]> var_1511_cast_fp16 = layer_norm(axes = var_1511_axes_0, beta = main_blocks_2_convnext_1_0_norm_norm_bias_to_fp16, epsilon = var_1233_to_fp16, gamma = main_blocks_2_convnext_1_0_norm_norm_weight_to_fp16, x = input_253_cast_fp16)[name = string("op_1511_cast_fp16")];
            tensor<int32, [3]> input_255_perm_0 = const()[name = string("input_255_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_101_pad_type_0 = const()[name = string("h_101_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_101_strides_0 = const()[name = string("h_101_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_101_pad_0 = const()[name = string("h_101_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_101_dilations_0 = const()[name = string("h_101_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_101_groups_0 = const()[name = string("h_101_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_2_convnext_1_0_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_1_0_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72363072)))];
            tensor<fp16, [2048]> main_blocks_2_convnext_1_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_1_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74460288)))];
            tensor<fp16, [2, 512, ?]> input_255_cast_fp16 = transpose(perm = input_255_perm_0, x = var_1511_cast_fp16)[name = string("transpose_52")];
            tensor<fp16, [2, 2048, ?]> h_101_cast_fp16 = conv(bias = main_blocks_2_convnext_1_0_pwconv1_bias_to_fp16, dilations = h_101_dilations_0, groups = h_101_groups_0, pad = h_101_pad_0, pad_type = h_101_pad_type_0, strides = h_101_strides_0, weight = main_blocks_2_convnext_1_0_pwconv1_weight_to_fp16, x = input_255_cast_fp16)[name = string("h_101_cast_fp16")];
            string input_257_mode_0 = const()[name = string("input_257_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_257_cast_fp16 = gelu(mode = input_257_mode_0, x = h_101_cast_fp16)[name = string("input_257_cast_fp16")];
            string h_103_pad_type_0 = const()[name = string("h_103_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_103_strides_0 = const()[name = string("h_103_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_103_pad_0 = const()[name = string("h_103_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_103_dilations_0 = const()[name = string("h_103_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_103_groups_0 = const()[name = string("h_103_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_1528_weight_0_to_fp16 = const()[name = string("op_1528_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74464448)))];
            tensor<fp16, [512]> var_1528_bias_0_to_fp16 = const()[name = string("op_1528_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76561664)))];
            tensor<fp16, [2, 512, ?]> var_1528_cast_fp16 = conv(bias = var_1528_bias_0_to_fp16, dilations = h_103_dilations_0, groups = h_103_groups_0, pad = h_103_pad_0, pad_type = h_103_pad_type_0, strides = h_103_strides_0, weight = var_1528_weight_0_to_fp16, x = input_257_cast_fp16)[name = string("op_1528_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_41_cast_fp16 = add(x = x_131_cast_fp16, y = var_1528_cast_fp16)[name = string("out_41_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_135_cast_fp16 = mul(x = out_41_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_135_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_m_9_cast_fp16 = mul(x = x_135_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_m_9_cast_fp16")];
            tensor<int32, [3]> var_1538_shape_cast_fp16 = shape(x = x_135_cast_fp16)[name = string("op_1538_shape_cast_fp16")];
            int32 gather_14_axis_0 = const()[name = string("gather_14_axis_0"), val = int32(0)];
            int32 gather_14_batch_dims_0 = const()[name = string("gather_14_batch_dims_0"), val = int32(0)];
            bool gather_14_validate_indices_0 = const()[name = string("gather_14_validate_indices_0"), val = bool(false)];
            string var_1538_shape_cast_fp16_to_uint16_dtype_0 = const()[name = string("op_1538_shape_cast_fp16_to_uint16_dtype_0"), val = string("uint16")];
            uint16 gather_14_indices_0_to_uint16 = const()[name = string("gather_14_indices_0_to_uint16"), val = uint16(2)];
            tensor<uint16, [3]> var_1538_shape_cast_fp16_to_uint16 = cast(dtype = var_1538_shape_cast_fp16_to_uint16_dtype_0, x = var_1538_shape_cast_fp16)[name = string("cast_89")];
            uint16 gather_14_cast_uint16 = gather(axis = gather_14_axis_0, batch_dims = gather_14_batch_dims_0, indices = gather_14_indices_0_to_uint16, validate_indices = gather_14_validate_indices_0, x = var_1538_shape_cast_fp16_to_uint16)[name = string("gather_14_cast_uint16")];
            string gather_14_cast_uint16_to_int32_dtype_0 = const()[name = string("gather_14_cast_uint16_to_int32_dtype_0"), val = string("int32")];
            tensor<int32, [3]> input_259_perm_0 = const()[name = string("input_259_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> main_blocks_2_text_attn_W_query_linear_weight_to_fp16 = const()[name = string("main_blocks_2_text_attn_W_query_linear_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76562752)))];
            tensor<fp16, [512]> main_blocks_2_text_attn_W_query_linear_bias_to_fp16 = const()[name = string("main_blocks_2_text_attn_W_query_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77087104)))];
            tensor<fp16, [2, ?, 512]> input_259_cast_fp16 = transpose(perm = input_259_perm_0, x = x_135_cast_fp16)[name = string("transpose_51")];
            tensor<fp16, [2, ?, 512]> linear_21_cast_fp16 = linear(bias = main_blocks_2_text_attn_W_query_linear_bias_to_fp16, weight = main_blocks_2_text_attn_W_query_linear_weight_to_fp16, x = input_259_cast_fp16)[name = string("linear_21_cast_fp16")];
            tensor<int32, [4]> concat_24x = const()[name = string("concat_24x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_1547_cast_fp16 = reshape(shape = concat_24x, x = linear_21_cast_fp16)[name = string("op_1547_cast_fp16")];
            tensor<int32, [4]> x_137_perm_0 = const()[name = string("x_137_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [512, 256]> main_blocks_2_text_attn_W_key_linear_weight_to_fp16 = const()[name = string("main_blocks_2_text_attn_W_key_linear_weight_to_fp16"), val = tensor<fp16, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77088192)))];
            tensor<fp16, [512]> main_blocks_2_text_attn_W_key_linear_bias_to_fp16 = const()[name = string("main_blocks_2_text_attn_W_key_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77350400)))];
            tensor<fp16, [?, ?, 512]> linear_22_cast_fp16 = linear(bias = main_blocks_2_text_attn_W_key_linear_bias_to_fp16, weight = main_blocks_2_text_attn_W_key_linear_weight_to_fp16, x = input_71_cast_fp16)[name = string("linear_22_cast_fp16")];
            tensor<int32, [4]> concat_25x = const()[name = string("concat_25x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_1555_cast_fp16 = reshape(shape = concat_25x, x = linear_22_cast_fp16)[name = string("op_1555_cast_fp16")];
            tensor<int32, [4]> x_139_perm_0 = const()[name = string("x_139_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [512, 256]> main_blocks_2_text_attn_W_value_linear_weight_to_fp16 = const()[name = string("main_blocks_2_text_attn_W_value_linear_weight_to_fp16"), val = tensor<fp16, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77351488)))];
            tensor<fp16, [512]> main_blocks_2_text_attn_W_value_linear_bias_to_fp16 = const()[name = string("main_blocks_2_text_attn_W_value_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77613696)))];
            tensor<fp16, [?, ?, 512]> linear_23_cast_fp16 = linear(bias = main_blocks_2_text_attn_W_value_linear_bias_to_fp16, weight = main_blocks_2_text_attn_W_value_linear_weight_to_fp16, x = input_71_cast_fp16)[name = string("linear_23_cast_fp16")];
            tensor<int32, [4]> concat_26x = const()[name = string("concat_26x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_1563_cast_fp16 = reshape(shape = concat_26x, x = linear_23_cast_fp16)[name = string("op_1563_cast_fp16")];
            tensor<int32, [4]> v_9_perm_0 = const()[name = string("v_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            int32 concat_27_values0_0 = const()[name = string("concat_27_values0_0"), val = int32(1)];
            int32 concat_27_values2_0 = const()[name = string("concat_27_values2_0"), val = int32(1)];
            int32 concat_27_axis_0 = const()[name = string("concat_27_axis_0"), val = int32(0)];
            bool concat_27_interleave_0 = const()[name = string("concat_27_interleave_0"), val = bool(false)];
            int32 gather_14_cast_uint16_to_int32 = cast(dtype = gather_14_cast_uint16_to_int32_dtype_0, x = gather_14_cast_uint16)[name = string("cast_88")];
            tensor<int32, [3]> concat_27 = concat(axis = concat_27_axis_0, interleave = concat_27_interleave_0, values = (concat_27_values0_0, gather_14_cast_uint16_to_int32, concat_27_values2_0))[name = string("concat_27")];
            tensor<int32, [3]> var_1566_begin_0 = const()[name = string("op_1566_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<bool, [3]> var_1566_end_mask_0 = const()[name = string("op_1566_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, ?, 1]> var_1566_cast_fp16 = slice_by_index(begin = var_1566_begin_0, end = concat_27, end_mask = var_1566_end_mask_0, x = main_blocks_0_text_attn_increments_to_fp16)[name = string("op_1566_cast_fp16")];
            tensor<int32, [3]> concat_28 = const()[name = string("concat_28"), val = tensor<int32, [3]>([2, -1, -1])];
            tensor<int32, [3]> shape_5_cast_fp16 = shape(x = var_1566_cast_fp16)[name = string("shape_5_cast_fp16")];
            tensor<bool, [3]> equal_5 = const()[name = string("equal_5"), val = tensor<bool, [3]>([false, true, true])];
            tensor<int32, [3]> select_5 = select(a = shape_5_cast_fp16, b = concat_28, cond = equal_5)[name = string("select_5")];
            tensor<int32, [3]> real_div_5 = real_div(x = select_5, y = shape_5_cast_fp16)[name = string("real_div_5")];
            tensor<fp16, [?, ?, ?]> var_1569_cast_fp16 = tile(reps = real_div_5, x = var_1566_cast_fp16)[name = string("op_1569_cast_fp16")];
            tensor<fp16, [2, ?, ?]> scaled_9_cast_fp16 = real_div(x = var_1569_cast_fp16, y = var_488_cast_fp16)[name = string("scaled_9_cast_fp16")];
            tensor<fp16, [1, 1, 32]> main_blocks_2_text_attn_theta_to_fp16 = const()[name = string("main_blocks_2_text_attn_theta_to_fp16"), val = tensor<fp16, [1, 1, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77614784)))];
            tensor<fp16, [2, ?, 32]> angles_9_cast_fp16 = mul(x = scaled_9_cast_fp16, y = main_blocks_2_text_attn_theta_to_fp16)[name = string("angles_9_cast_fp16")];
            tensor<fp16, [2, ?, 32]> var_1585_cast_fp16 = cos(x = angles_9_cast_fp16)[name = string("op_1585_cast_fp16")];
            tensor<int32, [1]> cos_9_axes_0 = const()[name = string("cos_9_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, ?, 32]> cos_9_cast_fp16 = expand_dims(axes = cos_9_axes_0, x = var_1585_cast_fp16)[name = string("cos_9_cast_fp16")];
            tensor<fp16, [2, ?, 32]> var_1587_cast_fp16 = sin(x = angles_9_cast_fp16)[name = string("op_1587_cast_fp16")];
            tensor<int32, [1]> sin_9_axes_0 = const()[name = string("sin_9_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, ?, 32]> sin_9_cast_fp16 = expand_dims(axes = sin_9_axes_0, x = var_1587_cast_fp16)[name = string("sin_9_cast_fp16")];
            tensor<int32, [4]> x_a_9_begin_0 = const()[name = string("x_a_9_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x_a_9_end_0 = const()[name = string("x_a_9_end_0"), val = tensor<int32, [4]>([2, 8, 0, 32])];
            tensor<bool, [4]> x_a_9_end_mask_0 = const()[name = string("x_a_9_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 8, ?, 64]> x_137_cast_fp16 = transpose(perm = x_137_perm_0, x = var_1547_cast_fp16)[name = string("transpose_50")];
            tensor<fp16, [2, 8, ?, 32]> x_a_9_cast_fp16 = slice_by_index(begin = x_a_9_begin_0, end = x_a_9_end_0, end_mask = x_a_9_end_mask_0, x = x_137_cast_fp16)[name = string("x_a_9_cast_fp16")];
            tensor<int32, [4]> x_b_9_begin_0 = const()[name = string("x_b_9_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x_b_9_end_0 = const()[name = string("x_b_9_end_0"), val = tensor<int32, [4]>([2, 8, 0, 64])];
            tensor<bool, [4]> x_b_9_end_mask_0 = const()[name = string("x_b_9_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [2, 8, ?, 32]> x_b_9_cast_fp16 = slice_by_index(begin = x_b_9_begin_0, end = x_b_9_end_0, end_mask = x_b_9_end_mask_0, x = x_137_cast_fp16)[name = string("x_b_9_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1591_cast_fp16 = mul(x = x_a_9_cast_fp16, y = cos_9_cast_fp16)[name = string("op_1591_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1592_cast_fp16 = mul(x = x_b_9_cast_fp16, y = sin_9_cast_fp16)[name = string("op_1592_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_a_9_cast_fp16 = sub(x = var_1591_cast_fp16, y = var_1592_cast_fp16)[name = string("rot_a_9_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1594_cast_fp16 = mul(x = x_a_9_cast_fp16, y = sin_9_cast_fp16)[name = string("op_1594_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1595_cast_fp16 = mul(x = x_b_9_cast_fp16, y = cos_9_cast_fp16)[name = string("op_1595_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_b_9_cast_fp16 = add(x = var_1594_cast_fp16, y = var_1595_cast_fp16)[name = string("rot_b_9_cast_fp16")];
            bool q_9_interleave_0 = const()[name = string("q_9_interleave_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> q_9_cast_fp16 = concat(axis = var_1228, interleave = q_9_interleave_0, values = (rot_a_9_cast_fp16, rot_b_9_cast_fp16))[name = string("q_9_cast_fp16")];
            tensor<int32, [4]> x_a_11_begin_0 = const()[name = string("x_a_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x_a_11_end_0 = const()[name = string("x_a_11_end_0"), val = tensor<int32, [4]>([2, 8, 0, 32])];
            tensor<bool, [4]> x_a_11_end_mask_0 = const()[name = string("x_a_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 8, ?, 64]> x_139_cast_fp16 = transpose(perm = x_139_perm_0, x = var_1555_cast_fp16)[name = string("transpose_49")];
            tensor<fp16, [2, 8, ?, 32]> x_a_11_cast_fp16 = slice_by_index(begin = x_a_11_begin_0, end = x_a_11_end_0, end_mask = x_a_11_end_mask_0, x = x_139_cast_fp16)[name = string("x_a_11_cast_fp16")];
            tensor<int32, [4]> x_b_11_begin_0 = const()[name = string("x_b_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x_b_11_end_0 = const()[name = string("x_b_11_end_0"), val = tensor<int32, [4]>([2, 8, 0, 64])];
            tensor<bool, [4]> x_b_11_end_mask_0 = const()[name = string("x_b_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [2, 8, ?, 32]> x_b_11_cast_fp16 = slice_by_index(begin = x_b_11_begin_0, end = x_b_11_end_0, end_mask = x_b_11_end_mask_0, x = x_139_cast_fp16)[name = string("x_b_11_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1609_cast_fp16 = mul(x = x_a_11_cast_fp16, y = cos_3_cast_fp16)[name = string("op_1609_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1610_cast_fp16 = mul(x = x_b_11_cast_fp16, y = sin_3_cast_fp16)[name = string("op_1610_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_a_11_cast_fp16 = sub(x = var_1609_cast_fp16, y = var_1610_cast_fp16)[name = string("rot_a_11_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1612_cast_fp16 = mul(x = x_a_11_cast_fp16, y = sin_3_cast_fp16)[name = string("op_1612_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1613_cast_fp16 = mul(x = x_b_11_cast_fp16, y = cos_3_cast_fp16)[name = string("op_1613_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_b_11_cast_fp16 = add(x = var_1612_cast_fp16, y = var_1613_cast_fp16)[name = string("rot_b_11_cast_fp16")];
            bool k_9_interleave_0 = const()[name = string("k_9_interleave_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> k_9_cast_fp16 = concat(axis = var_1228, interleave = k_9_interleave_0, values = (rot_a_11_cast_fp16, rot_b_11_cast_fp16))[name = string("k_9_cast_fp16")];
            bool var_1618_transpose_x_1 = const()[name = string("op_1618_transpose_x_1"), val = bool(false)];
            bool var_1618_transpose_y_1 = const()[name = string("op_1618_transpose_y_1"), val = bool(true)];
            tensor<fp16, [2, 8, ?, ?]> var_1618_cast_fp16 = matmul(transpose_x = var_1618_transpose_x_1, transpose_y = var_1618_transpose_y_1, x = q_9_cast_fp16, y = k_9_cast_fp16)[name = string("op_1618_cast_fp16")];
            fp16 _inversed_scores_5_y_0_to_fp16 = const()[name = string("_inversed_scores_5_y_0_to_fp16"), val = fp16(0x1p-4)];
            tensor<fp16, [2, 8, ?, ?]> _inversed_scores_5_cast_fp16 = mul(x = var_1618_cast_fp16, y = _inversed_scores_5_y_0_to_fp16)[name = string("_inversed_scores_5_cast_fp16")];
            tensor<fp16, [2, 8, ?, ?]> input_265_cast_fp16 = sub(x = _inversed_scores_5_cast_fp16, y = var_530_cast_fp16)[name = string("input_265_cast_fp16")];
            tensor<fp16, [2, 8, ?, ?]> attn_13_cast_fp16 = softmax(axis = var_1228, x = input_265_cast_fp16)[name = string("attn_13_cast_fp16")];
            bool var_1627_transpose_x_0 = const()[name = string("op_1627_transpose_x_0"), val = bool(false)];
            bool var_1627_transpose_y_0 = const()[name = string("op_1627_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> v_9_cast_fp16 = transpose(perm = v_9_perm_0, x = var_1563_cast_fp16)[name = string("transpose_48")];
            tensor<fp16, [2, 8, ?, 64]> var_1627_cast_fp16 = matmul(transpose_x = var_1627_transpose_x_0, transpose_y = var_1627_transpose_y_0, x = attn_13_cast_fp16, y = v_9_cast_fp16)[name = string("op_1627_cast_fp16")];
            tensor<int32, [4]> var_1628_perm_0 = const()[name = string("op_1628_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_31x = const()[name = string("concat_31x"), val = tensor<int32, [3]>([2, -1, 512])];
            tensor<fp16, [2, ?, 8, 64]> var_1628_cast_fp16 = transpose(perm = var_1628_perm_0, x = var_1627_cast_fp16)[name = string("transpose_47")];
            tensor<fp16, [2, ?, 512]> input_267_cast_fp16 = reshape(shape = concat_31x, x = var_1628_cast_fp16)[name = string("input_267_cast_fp16")];
            tensor<fp16, [512, 512]> main_blocks_2_text_attn_out_fc_linear_weight_to_fp16 = const()[name = string("main_blocks_2_text_attn_out_fc_linear_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77614912)))];
            tensor<fp16, [512]> main_blocks_2_text_attn_out_fc_linear_bias_to_fp16 = const()[name = string("main_blocks_2_text_attn_out_fc_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78139264)))];
            tensor<fp16, [2, ?, 512]> linear_24_cast_fp16 = linear(bias = main_blocks_2_text_attn_out_fc_linear_bias_to_fp16, weight = main_blocks_2_text_attn_out_fc_linear_weight_to_fp16, x = input_267_cast_fp16)[name = string("linear_24_cast_fp16")];
            tensor<int32, [3]> out_43_perm_0 = const()[name = string("out_43_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> out_43_cast_fp16 = transpose(perm = out_43_perm_0, x = linear_24_cast_fp16)[name = string("transpose_46")];
            tensor<fp16, [2, 512, ?]> var_1636_cast_fp16 = mul(x = out_43_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_1636_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_141_cast_fp16 = add(x = x_m_9_cast_fp16, y = var_1636_cast_fp16)[name = string("x_141_cast_fp16")];
            tensor<int32, [3]> input_269_perm_0 = const()[name = string("input_269_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1643_axes_0 = const()[name = string("op_1643_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_2_text_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_2_text_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78140352)))];
            tensor<fp16, [512]> main_blocks_2_text_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_2_text_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78141440)))];
            tensor<fp16, [2, ?, 512]> input_269_cast_fp16 = transpose(perm = input_269_perm_0, x = x_141_cast_fp16)[name = string("transpose_45")];
            tensor<fp16, [2, ?, 512]> var_1643_cast_fp16 = layer_norm(axes = var_1643_axes_0, beta = main_blocks_2_text_norm_norm_bias_to_fp16, epsilon = var_1233_to_fp16, gamma = main_blocks_2_text_norm_norm_weight_to_fp16, x = input_269_cast_fp16)[name = string("op_1643_cast_fp16")];
            tensor<int32, [3]> var_1644_perm_0 = const()[name = string("op_1644_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> var_1644_cast_fp16 = transpose(perm = var_1644_perm_0, x = var_1643_cast_fp16)[name = string("transpose_44")];
            tensor<fp16, [2, 512, ?]> x_143_cast_fp16 = mul(x = var_1644_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_143_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_145_cast_fp16 = mul(x = x_143_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_145_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1657_cast_fp16 = mul(x = x_145_cast_fp16, y = last_1_cast_fp16)[name = string("op_1657_cast_fp16")];
            tensor<int32, [1]> x_last_35_axes_0 = const()[name = string("x_last_35_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_35_keep_dims_0 = const()[name = string("x_last_35_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_35_cast_fp16 = reduce_sum(axes = x_last_35_axes_0, keep_dims = x_last_35_keep_dims_0, x = var_1657_cast_fp16)[name = string("x_last_35_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1660_cast_fp16 = mul(x = x_145_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_1660_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1662_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_35_cast_fp16)[name = string("op_1662_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_273_cast_fp16 = add(x = var_1660_cast_fp16, y = var_1662_cast_fp16)[name = string("input_273_cast_fp16")];
            tensor<int32, [6]> input_275_pad_0 = const()[name = string("input_275_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_275_mode_0 = const()[name = string("input_275_mode_0"), val = string("replicate")];
            fp16 const_37_to_fp16 = const()[name = string("const_37_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_275_cast_fp16 = pad(constant_val = const_37_to_fp16, mode = input_275_mode_0, pad = input_275_pad_0, x = input_273_cast_fp16)[name = string("input_275_cast_fp16")];
            string h_105_pad_type_0 = const()[name = string("h_105_pad_type_0"), val = string("valid")];
            int32 h_105_groups_0 = const()[name = string("h_105_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_105_strides_0 = const()[name = string("h_105_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_105_pad_0 = const()[name = string("h_105_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_105_dilations_0 = const()[name = string("h_105_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> main_blocks_2_convnext_2_0_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_2_0_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78142528)))];
            tensor<fp16, [512]> main_blocks_2_convnext_2_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_2_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78147712)))];
            tensor<fp16, [2, 512, ?]> h_105_cast_fp16 = conv(bias = main_blocks_2_convnext_2_0_dwconv__conv_bias_to_fp16, dilations = h_105_dilations_0, groups = h_105_groups_0, pad = h_105_pad_0, pad_type = h_105_pad_type_0, strides = h_105_strides_0, weight = main_blocks_2_convnext_2_0_dwconv__conv_weight_to_fp16, x = input_275_cast_fp16)[name = string("h_105_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_147_cast_fp16 = mul(x = h_105_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_147_cast_fp16")];
            tensor<int32, [3]> input_277_perm_0 = const()[name = string("input_277_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1680_axes_0 = const()[name = string("op_1680_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_2_convnext_2_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_2_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78148800)))];
            tensor<fp16, [512]> main_blocks_2_convnext_2_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_2_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78149888)))];
            tensor<fp16, [2, ?, 512]> input_277_cast_fp16 = transpose(perm = input_277_perm_0, x = x_147_cast_fp16)[name = string("transpose_43")];
            tensor<fp16, [2, ?, 512]> var_1680_cast_fp16 = layer_norm(axes = var_1680_axes_0, beta = main_blocks_2_convnext_2_0_norm_norm_bias_to_fp16, epsilon = var_1233_to_fp16, gamma = main_blocks_2_convnext_2_0_norm_norm_weight_to_fp16, x = input_277_cast_fp16)[name = string("op_1680_cast_fp16")];
            tensor<int32, [3]> input_279_perm_0 = const()[name = string("input_279_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_107_pad_type_0 = const()[name = string("h_107_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_107_strides_0 = const()[name = string("h_107_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_107_pad_0 = const()[name = string("h_107_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_107_dilations_0 = const()[name = string("h_107_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_107_groups_0 = const()[name = string("h_107_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_2_convnext_2_0_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_2_0_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78150976)))];
            tensor<fp16, [2048]> main_blocks_2_convnext_2_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_2_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80248192)))];
            tensor<fp16, [2, 512, ?]> input_279_cast_fp16 = transpose(perm = input_279_perm_0, x = var_1680_cast_fp16)[name = string("transpose_42")];
            tensor<fp16, [2, 2048, ?]> h_107_cast_fp16 = conv(bias = main_blocks_2_convnext_2_0_pwconv1_bias_to_fp16, dilations = h_107_dilations_0, groups = h_107_groups_0, pad = h_107_pad_0, pad_type = h_107_pad_type_0, strides = h_107_strides_0, weight = main_blocks_2_convnext_2_0_pwconv1_weight_to_fp16, x = input_279_cast_fp16)[name = string("h_107_cast_fp16")];
            string input_281_mode_0 = const()[name = string("input_281_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_281_cast_fp16 = gelu(mode = input_281_mode_0, x = h_107_cast_fp16)[name = string("input_281_cast_fp16")];
            string h_109_pad_type_0 = const()[name = string("h_109_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_109_strides_0 = const()[name = string("h_109_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_109_pad_0 = const()[name = string("h_109_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_109_dilations_0 = const()[name = string("h_109_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_109_groups_0 = const()[name = string("h_109_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_1697_weight_0_to_fp16 = const()[name = string("op_1697_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80252352)))];
            tensor<fp16, [512]> var_1697_bias_0_to_fp16 = const()[name = string("op_1697_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82349568)))];
            tensor<fp16, [2, 512, ?]> var_1697_cast_fp16 = conv(bias = var_1697_bias_0_to_fp16, dilations = h_109_dilations_0, groups = h_109_groups_0, pad = h_109_pad_0, pad_type = h_109_pad_type_0, strides = h_109_strides_0, weight = var_1697_weight_0_to_fp16, x = input_281_cast_fp16)[name = string("op_1697_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_45_cast_fp16 = add(x = x_145_cast_fp16, y = var_1697_cast_fp16)[name = string("out_45_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_149_cast_fp16 = mul(x = out_45_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_149_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_m_11_cast_fp16 = mul(x = x_149_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_m_11_cast_fp16")];
            tensor<int32, [3]> input_283_perm_0 = const()[name = string("input_283_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [256, 512]> main_blocks_2_style_attn_W_query_linear_weight_to_fp16 = const()[name = string("main_blocks_2_style_attn_W_query_linear_weight_to_fp16"), val = tensor<fp16, [256, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82350656)))];
            tensor<fp16, [256]> main_blocks_2_style_attn_W_query_linear_bias_to_fp16 = const()[name = string("main_blocks_2_style_attn_W_query_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82612864)))];
            tensor<fp16, [2, ?, 512]> input_283_cast_fp16 = transpose(perm = input_283_perm_0, x = x_149_cast_fp16)[name = string("transpose_41")];
            tensor<fp16, [2, ?, 256]> linear_25_cast_fp16 = linear(bias = main_blocks_2_style_attn_W_query_linear_bias_to_fp16, weight = main_blocks_2_style_attn_W_query_linear_weight_to_fp16, x = input_283_cast_fp16)[name = string("linear_25_cast_fp16")];
            tensor<int32, [4]> concat_32x = const()[name = string("concat_32x"), val = tensor<int32, [4]>([2, -1, 2, 128])];
            tensor<fp16, [2, ?, 2, 128]> var_1714_cast_fp16 = reshape(shape = concat_32x, x = linear_25_cast_fp16)[name = string("op_1714_cast_fp16")];
            tensor<int32, [4]> q_11_perm_0 = const()[name = string("q_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [256, 256]> main_blocks_2_style_attn_W_value_linear_weight_to_fp16 = const()[name = string("main_blocks_2_style_attn_W_value_linear_weight_to_fp16"), val = tensor<fp16, [256, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82613440)))];
            tensor<fp16, [256]> main_blocks_2_style_attn_W_value_linear_bias_to_fp16 = const()[name = string("main_blocks_2_style_attn_W_value_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82744576)))];
            tensor<fp16, [2, 50, 256]> linear_27_cast_fp16 = linear(bias = main_blocks_2_style_attn_W_value_linear_bias_to_fp16, weight = main_blocks_2_style_attn_W_value_linear_weight_to_fp16, x = input_95_cast_fp16)[name = string("linear_27_cast_fp16")];
            tensor<int32, [4]> var_1727 = const()[name = string("op_1727"), val = tensor<int32, [4]>([2, 50, 2, 128])];
            tensor<fp16, [2, 50, 2, 128]> var_1728_cast_fp16 = reshape(shape = var_1727, x = linear_27_cast_fp16)[name = string("op_1728_cast_fp16")];
            tensor<int32, [4]> v_11_perm_0 = const()[name = string("v_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_1732_transpose_x_0 = const()[name = string("op_1732_transpose_x_0"), val = bool(false)];
            bool var_1732_transpose_y_0 = const()[name = string("op_1732_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 2, 128, 50]> var_1731_to_fp16 = const()[name = string("op_1731_to_fp16"), val = tensor<fp16, [2, 2, 128, 50]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82745152)))];
            tensor<fp16, [2, 2, ?, 128]> q_11_cast_fp16 = transpose(perm = q_11_perm_0, x = var_1714_cast_fp16)[name = string("transpose_40")];
            tensor<fp16, [2, 2, ?, 50]> var_1732_cast_fp16 = matmul(transpose_x = var_1732_transpose_x_0, transpose_y = var_1732_transpose_y_0, x = q_11_cast_fp16, y = var_1731_to_fp16)[name = string("op_1732_cast_fp16")];
            fp16 _inversed_input_285_y_0_to_fp16 = const()[name = string("_inversed_input_285_y_0_to_fp16"), val = fp16(0x1p-4)];
            tensor<fp16, [2, 2, ?, 50]> _inversed_input_285_cast_fp16 = mul(x = var_1732_cast_fp16, y = _inversed_input_285_y_0_to_fp16)[name = string("_inversed_input_285_cast_fp16")];
            tensor<fp16, [2, 2, ?, 50]> attn_15_cast_fp16 = softmax(axis = var_1228, x = _inversed_input_285_cast_fp16)[name = string("attn_15_cast_fp16")];
            tensor<fp16, [2, 2, ?, 50]> attn_17_cast_fp16 = mul(x = attn_15_cast_fp16, y = mask_1_cast_fp16)[name = string("attn_17_cast_fp16")];
            bool var_1739_transpose_x_0 = const()[name = string("op_1739_transpose_x_0"), val = bool(false)];
            bool var_1739_transpose_y_0 = const()[name = string("op_1739_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 2, 50, 128]> v_11_cast_fp16 = transpose(perm = v_11_perm_0, x = var_1728_cast_fp16)[name = string("transpose_39")];
            tensor<fp16, [2, 2, ?, 128]> var_1739_cast_fp16 = matmul(transpose_x = var_1739_transpose_x_0, transpose_y = var_1739_transpose_y_0, x = attn_17_cast_fp16, y = v_11_cast_fp16)[name = string("op_1739_cast_fp16")];
            tensor<int32, [4]> var_1740_perm_0 = const()[name = string("op_1740_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_33x = const()[name = string("concat_33x"), val = tensor<int32, [3]>([2, -1, 256])];
            tensor<fp16, [2, ?, 2, 128]> var_1740_cast_fp16 = transpose(perm = var_1740_perm_0, x = var_1739_cast_fp16)[name = string("transpose_38")];
            tensor<fp16, [2, ?, 256]> input_287_cast_fp16 = reshape(shape = concat_33x, x = var_1740_cast_fp16)[name = string("input_287_cast_fp16")];
            tensor<fp16, [512, 256]> main_blocks_2_style_attn_out_fc_linear_weight_to_fp16 = const()[name = string("main_blocks_2_style_attn_out_fc_linear_weight_to_fp16"), val = tensor<fp16, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82796416)))];
            tensor<fp16, [512]> main_blocks_2_style_attn_out_fc_linear_bias_to_fp16 = const()[name = string("main_blocks_2_style_attn_out_fc_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83058624)))];
            tensor<fp16, [2, ?, 512]> linear_28_cast_fp16 = linear(bias = main_blocks_2_style_attn_out_fc_linear_bias_to_fp16, weight = main_blocks_2_style_attn_out_fc_linear_weight_to_fp16, x = input_287_cast_fp16)[name = string("linear_28_cast_fp16")];
            tensor<int32, [3]> out_47_perm_0 = const()[name = string("out_47_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> out_47_cast_fp16 = transpose(perm = out_47_perm_0, x = linear_28_cast_fp16)[name = string("transpose_37")];
            tensor<fp16, [2, 512, ?]> var_1748_cast_fp16 = mul(x = out_47_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_1748_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_151_cast_fp16 = add(x = x_m_11_cast_fp16, y = var_1748_cast_fp16)[name = string("x_151_cast_fp16")];
            tensor<int32, [3]> input_289_perm_0 = const()[name = string("input_289_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1755_axes_0 = const()[name = string("op_1755_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_2_style_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_2_style_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83059712)))];
            tensor<fp16, [512]> main_blocks_2_style_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_2_style_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83060800)))];
            tensor<fp16, [2, ?, 512]> input_289_cast_fp16 = transpose(perm = input_289_perm_0, x = x_151_cast_fp16)[name = string("transpose_36")];
            tensor<fp16, [2, ?, 512]> var_1755_cast_fp16 = layer_norm(axes = var_1755_axes_0, beta = main_blocks_2_style_norm_norm_bias_to_fp16, epsilon = var_1233_to_fp16, gamma = main_blocks_2_style_norm_norm_weight_to_fp16, x = input_289_cast_fp16)[name = string("op_1755_cast_fp16")];
            tensor<int32, [3]> var_1756_perm_0 = const()[name = string("op_1756_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> var_1756_cast_fp16 = transpose(perm = var_1756_perm_0, x = var_1755_cast_fp16)[name = string("transpose_35")];
            tensor<fp16, [2, 512, ?]> x_153_cast_fp16 = mul(x = var_1756_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_153_cast_fp16")];
            int32 var_1775 = const()[name = string("op_1775"), val = int32(-1)];
            tensor<fp16, [2, 512, ?]> x_155_cast_fp16 = mul(x = x_153_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_155_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1810_cast_fp16 = mul(x = x_155_cast_fp16, y = last_1_cast_fp16)[name = string("op_1810_cast_fp16")];
            tensor<int32, [1]> x_last_37_axes_0 = const()[name = string("x_last_37_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_37_keep_dims_0 = const()[name = string("x_last_37_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_37_cast_fp16 = reduce_sum(axes = x_last_37_axes_0, keep_dims = x_last_37_keep_dims_0, x = var_1810_cast_fp16)[name = string("x_last_37_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1813_cast_fp16 = mul(x = x_155_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_1813_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1815_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_37_cast_fp16)[name = string("op_1815_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_293_cast_fp16 = add(x = var_1813_cast_fp16, y = var_1815_cast_fp16)[name = string("input_293_cast_fp16")];
            tensor<int32, [6]> input_295_pad_0 = const()[name = string("input_295_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_295_mode_0 = const()[name = string("input_295_mode_0"), val = string("replicate")];
            fp16 const_40_to_fp16 = const()[name = string("const_40_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_295_cast_fp16 = pad(constant_val = const_40_to_fp16, mode = input_295_mode_0, pad = input_295_pad_0, x = input_293_cast_fp16)[name = string("input_295_cast_fp16")];
            string h_111_pad_type_0 = const()[name = string("h_111_pad_type_0"), val = string("valid")];
            int32 h_111_groups_0 = const()[name = string("h_111_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_111_strides_0 = const()[name = string("h_111_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_111_pad_0 = const()[name = string("h_111_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_111_dilations_0 = const()[name = string("h_111_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> main_blocks_3_convnext_0_0_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_0_0_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83061888)))];
            tensor<fp16, [512]> main_blocks_3_convnext_0_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83067072)))];
            tensor<fp16, [2, 512, ?]> h_111_cast_fp16 = conv(bias = main_blocks_3_convnext_0_0_dwconv__conv_bias_to_fp16, dilations = h_111_dilations_0, groups = h_111_groups_0, pad = h_111_pad_0, pad_type = h_111_pad_type_0, strides = h_111_strides_0, weight = main_blocks_3_convnext_0_0_dwconv__conv_weight_to_fp16, x = input_295_cast_fp16)[name = string("h_111_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_157_cast_fp16 = mul(x = h_111_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_157_cast_fp16")];
            tensor<int32, [3]> input_297_perm_0 = const()[name = string("input_297_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1833_axes_0 = const()[name = string("op_1833_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_3_convnext_0_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_0_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83068160)))];
            tensor<fp16, [512]> main_blocks_3_convnext_0_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83069248)))];
            fp16 var_1780_to_fp16 = const()[name = string("op_1780_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [2, ?, 512]> input_297_cast_fp16 = transpose(perm = input_297_perm_0, x = x_157_cast_fp16)[name = string("transpose_34")];
            tensor<fp16, [2, ?, 512]> var_1833_cast_fp16 = layer_norm(axes = var_1833_axes_0, beta = main_blocks_3_convnext_0_0_norm_norm_bias_to_fp16, epsilon = var_1780_to_fp16, gamma = main_blocks_3_convnext_0_0_norm_norm_weight_to_fp16, x = input_297_cast_fp16)[name = string("op_1833_cast_fp16")];
            tensor<int32, [3]> input_299_perm_0 = const()[name = string("input_299_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_113_pad_type_0 = const()[name = string("h_113_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_113_strides_0 = const()[name = string("h_113_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_113_pad_0 = const()[name = string("h_113_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_113_dilations_0 = const()[name = string("h_113_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_113_groups_0 = const()[name = string("h_113_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_3_convnext_0_0_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_0_0_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83070336)))];
            tensor<fp16, [2048]> main_blocks_3_convnext_0_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85167552)))];
            tensor<fp16, [2, 512, ?]> input_299_cast_fp16 = transpose(perm = input_299_perm_0, x = var_1833_cast_fp16)[name = string("transpose_33")];
            tensor<fp16, [2, 2048, ?]> h_113_cast_fp16 = conv(bias = main_blocks_3_convnext_0_0_pwconv1_bias_to_fp16, dilations = h_113_dilations_0, groups = h_113_groups_0, pad = h_113_pad_0, pad_type = h_113_pad_type_0, strides = h_113_strides_0, weight = main_blocks_3_convnext_0_0_pwconv1_weight_to_fp16, x = input_299_cast_fp16)[name = string("h_113_cast_fp16")];
            string input_301_mode_0 = const()[name = string("input_301_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_301_cast_fp16 = gelu(mode = input_301_mode_0, x = h_113_cast_fp16)[name = string("input_301_cast_fp16")];
            string h_115_pad_type_0 = const()[name = string("h_115_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_115_strides_0 = const()[name = string("h_115_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_115_pad_0 = const()[name = string("h_115_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_115_dilations_0 = const()[name = string("h_115_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_115_groups_0 = const()[name = string("h_115_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_1850_weight_0_to_fp16 = const()[name = string("op_1850_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85171712)))];
            tensor<fp16, [512]> var_1850_bias_0_to_fp16 = const()[name = string("op_1850_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87268928)))];
            tensor<fp16, [2, 512, ?]> var_1850_cast_fp16 = conv(bias = var_1850_bias_0_to_fp16, dilations = h_115_dilations_0, groups = h_115_groups_0, pad = h_115_pad_0, pad_type = h_115_pad_type_0, strides = h_115_strides_0, weight = var_1850_weight_0_to_fp16, x = input_301_cast_fp16)[name = string("op_1850_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_49_cast_fp16 = add(x = x_155_cast_fp16, y = var_1850_cast_fp16)[name = string("out_49_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_159_cast_fp16 = mul(x = out_49_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_159_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_161_cast_fp16 = mul(x = x_159_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_161_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1864_cast_fp16 = mul(x = x_161_cast_fp16, y = last_1_cast_fp16)[name = string("op_1864_cast_fp16")];
            tensor<int32, [1]> x_last_39_axes_0 = const()[name = string("x_last_39_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_39_keep_dims_0 = const()[name = string("x_last_39_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_39_cast_fp16 = reduce_sum(axes = x_last_39_axes_0, keep_dims = x_last_39_keep_dims_0, x = var_1864_cast_fp16)[name = string("x_last_39_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1867_cast_fp16 = mul(x = x_161_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_1867_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1869_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_39_cast_fp16)[name = string("op_1869_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_305_cast_fp16 = add(x = var_1867_cast_fp16, y = var_1869_cast_fp16)[name = string("input_305_cast_fp16")];
            tensor<int32, [6]> input_307_pad_0 = const()[name = string("input_307_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_307_mode_0 = const()[name = string("input_307_mode_0"), val = string("replicate")];
            fp16 const_42_to_fp16 = const()[name = string("const_42_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_307_cast_fp16 = pad(constant_val = const_42_to_fp16, mode = input_307_mode_0, pad = input_307_pad_0, x = input_305_cast_fp16)[name = string("input_307_cast_fp16")];
            string h_117_pad_type_0 = const()[name = string("h_117_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_117_dilations_0 = const()[name = string("h_117_dilations_0"), val = tensor<int32, [1]>([2])];
            int32 h_117_groups_0 = const()[name = string("h_117_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_117_strides_0 = const()[name = string("h_117_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_117_pad_0 = const()[name = string("h_117_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [512, 1, 5]> main_blocks_3_convnext_0_1_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_0_1_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87270016)))];
            tensor<fp16, [512]> main_blocks_3_convnext_0_1_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_1_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87275200)))];
            tensor<fp16, [2, 512, ?]> h_117_cast_fp16 = conv(bias = main_blocks_3_convnext_0_1_dwconv__conv_bias_to_fp16, dilations = h_117_dilations_0, groups = h_117_groups_0, pad = h_117_pad_0, pad_type = h_117_pad_type_0, strides = h_117_strides_0, weight = main_blocks_3_convnext_0_1_dwconv__conv_weight_to_fp16, x = input_307_cast_fp16)[name = string("h_117_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_163_cast_fp16 = mul(x = h_117_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_163_cast_fp16")];
            tensor<int32, [3]> input_309_perm_0 = const()[name = string("input_309_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1887_axes_0 = const()[name = string("op_1887_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_3_convnext_0_1_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_0_1_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87276288)))];
            tensor<fp16, [512]> main_blocks_3_convnext_0_1_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_1_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87277376)))];
            tensor<fp16, [2, ?, 512]> input_309_cast_fp16 = transpose(perm = input_309_perm_0, x = x_163_cast_fp16)[name = string("transpose_32")];
            tensor<fp16, [2, ?, 512]> var_1887_cast_fp16 = layer_norm(axes = var_1887_axes_0, beta = main_blocks_3_convnext_0_1_norm_norm_bias_to_fp16, epsilon = var_1780_to_fp16, gamma = main_blocks_3_convnext_0_1_norm_norm_weight_to_fp16, x = input_309_cast_fp16)[name = string("op_1887_cast_fp16")];
            tensor<int32, [3]> input_311_perm_0 = const()[name = string("input_311_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_119_pad_type_0 = const()[name = string("h_119_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_119_strides_0 = const()[name = string("h_119_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_119_pad_0 = const()[name = string("h_119_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_119_dilations_0 = const()[name = string("h_119_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_119_groups_0 = const()[name = string("h_119_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_3_convnext_0_1_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_0_1_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87278464)))];
            tensor<fp16, [2048]> main_blocks_3_convnext_0_1_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_1_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89375680)))];
            tensor<fp16, [2, 512, ?]> input_311_cast_fp16 = transpose(perm = input_311_perm_0, x = var_1887_cast_fp16)[name = string("transpose_31")];
            tensor<fp16, [2, 2048, ?]> h_119_cast_fp16 = conv(bias = main_blocks_3_convnext_0_1_pwconv1_bias_to_fp16, dilations = h_119_dilations_0, groups = h_119_groups_0, pad = h_119_pad_0, pad_type = h_119_pad_type_0, strides = h_119_strides_0, weight = main_blocks_3_convnext_0_1_pwconv1_weight_to_fp16, x = input_311_cast_fp16)[name = string("h_119_cast_fp16")];
            string input_313_mode_0 = const()[name = string("input_313_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_313_cast_fp16 = gelu(mode = input_313_mode_0, x = h_119_cast_fp16)[name = string("input_313_cast_fp16")];
            string h_121_pad_type_0 = const()[name = string("h_121_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_121_strides_0 = const()[name = string("h_121_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_121_pad_0 = const()[name = string("h_121_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_121_dilations_0 = const()[name = string("h_121_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_121_groups_0 = const()[name = string("h_121_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_1904_weight_0_to_fp16 = const()[name = string("op_1904_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89379840)))];
            tensor<fp16, [512]> var_1904_bias_0_to_fp16 = const()[name = string("op_1904_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91477056)))];
            tensor<fp16, [2, 512, ?]> var_1904_cast_fp16 = conv(bias = var_1904_bias_0_to_fp16, dilations = h_121_dilations_0, groups = h_121_groups_0, pad = h_121_pad_0, pad_type = h_121_pad_type_0, strides = h_121_strides_0, weight = var_1904_weight_0_to_fp16, x = input_313_cast_fp16)[name = string("op_1904_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_51_cast_fp16 = add(x = x_161_cast_fp16, y = var_1904_cast_fp16)[name = string("out_51_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_165_cast_fp16 = mul(x = out_51_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_165_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_167_cast_fp16 = mul(x = x_165_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_167_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1918_cast_fp16 = mul(x = x_167_cast_fp16, y = last_1_cast_fp16)[name = string("op_1918_cast_fp16")];
            tensor<int32, [1]> x_last_41_axes_0 = const()[name = string("x_last_41_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_41_keep_dims_0 = const()[name = string("x_last_41_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_41_cast_fp16 = reduce_sum(axes = x_last_41_axes_0, keep_dims = x_last_41_keep_dims_0, x = var_1918_cast_fp16)[name = string("x_last_41_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1921_cast_fp16 = mul(x = x_167_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_1921_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1923_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_41_cast_fp16)[name = string("op_1923_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_317_cast_fp16 = add(x = var_1921_cast_fp16, y = var_1923_cast_fp16)[name = string("input_317_cast_fp16")];
            tensor<int32, [6]> input_319_pad_0 = const()[name = string("input_319_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 8, 8])];
            string input_319_mode_0 = const()[name = string("input_319_mode_0"), val = string("replicate")];
            fp16 const_44_to_fp16 = const()[name = string("const_44_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_319_cast_fp16 = pad(constant_val = const_44_to_fp16, mode = input_319_mode_0, pad = input_319_pad_0, x = input_317_cast_fp16)[name = string("input_319_cast_fp16")];
            string h_123_pad_type_0 = const()[name = string("h_123_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_123_dilations_0 = const()[name = string("h_123_dilations_0"), val = tensor<int32, [1]>([4])];
            int32 h_123_groups_0 = const()[name = string("h_123_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_123_strides_0 = const()[name = string("h_123_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_123_pad_0 = const()[name = string("h_123_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [512, 1, 5]> main_blocks_3_convnext_0_2_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_0_2_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91478144)))];
            tensor<fp16, [512]> main_blocks_3_convnext_0_2_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_2_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91483328)))];
            tensor<fp16, [2, 512, ?]> h_123_cast_fp16 = conv(bias = main_blocks_3_convnext_0_2_dwconv__conv_bias_to_fp16, dilations = h_123_dilations_0, groups = h_123_groups_0, pad = h_123_pad_0, pad_type = h_123_pad_type_0, strides = h_123_strides_0, weight = main_blocks_3_convnext_0_2_dwconv__conv_weight_to_fp16, x = input_319_cast_fp16)[name = string("h_123_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_169_cast_fp16 = mul(x = h_123_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_169_cast_fp16")];
            tensor<int32, [3]> input_321_perm_0 = const()[name = string("input_321_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1941_axes_0 = const()[name = string("op_1941_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_3_convnext_0_2_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_0_2_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91484416)))];
            tensor<fp16, [512]> main_blocks_3_convnext_0_2_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_2_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91485504)))];
            tensor<fp16, [2, ?, 512]> input_321_cast_fp16 = transpose(perm = input_321_perm_0, x = x_169_cast_fp16)[name = string("transpose_30")];
            tensor<fp16, [2, ?, 512]> var_1941_cast_fp16 = layer_norm(axes = var_1941_axes_0, beta = main_blocks_3_convnext_0_2_norm_norm_bias_to_fp16, epsilon = var_1780_to_fp16, gamma = main_blocks_3_convnext_0_2_norm_norm_weight_to_fp16, x = input_321_cast_fp16)[name = string("op_1941_cast_fp16")];
            tensor<int32, [3]> input_323_perm_0 = const()[name = string("input_323_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_125_pad_type_0 = const()[name = string("h_125_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_125_strides_0 = const()[name = string("h_125_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_125_pad_0 = const()[name = string("h_125_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_125_dilations_0 = const()[name = string("h_125_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_125_groups_0 = const()[name = string("h_125_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_3_convnext_0_2_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_0_2_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91486592)))];
            tensor<fp16, [2048]> main_blocks_3_convnext_0_2_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_2_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93583808)))];
            tensor<fp16, [2, 512, ?]> input_323_cast_fp16 = transpose(perm = input_323_perm_0, x = var_1941_cast_fp16)[name = string("transpose_29")];
            tensor<fp16, [2, 2048, ?]> h_125_cast_fp16 = conv(bias = main_blocks_3_convnext_0_2_pwconv1_bias_to_fp16, dilations = h_125_dilations_0, groups = h_125_groups_0, pad = h_125_pad_0, pad_type = h_125_pad_type_0, strides = h_125_strides_0, weight = main_blocks_3_convnext_0_2_pwconv1_weight_to_fp16, x = input_323_cast_fp16)[name = string("h_125_cast_fp16")];
            string input_325_mode_0 = const()[name = string("input_325_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_325_cast_fp16 = gelu(mode = input_325_mode_0, x = h_125_cast_fp16)[name = string("input_325_cast_fp16")];
            string h_127_pad_type_0 = const()[name = string("h_127_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_127_strides_0 = const()[name = string("h_127_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_127_pad_0 = const()[name = string("h_127_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_127_dilations_0 = const()[name = string("h_127_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_127_groups_0 = const()[name = string("h_127_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_1958_weight_0_to_fp16 = const()[name = string("op_1958_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93587968)))];
            tensor<fp16, [512]> var_1958_bias_0_to_fp16 = const()[name = string("op_1958_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95685184)))];
            tensor<fp16, [2, 512, ?]> var_1958_cast_fp16 = conv(bias = var_1958_bias_0_to_fp16, dilations = h_127_dilations_0, groups = h_127_groups_0, pad = h_127_pad_0, pad_type = h_127_pad_type_0, strides = h_127_strides_0, weight = var_1958_weight_0_to_fp16, x = input_325_cast_fp16)[name = string("op_1958_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_53_cast_fp16 = add(x = x_167_cast_fp16, y = var_1958_cast_fp16)[name = string("out_53_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_171_cast_fp16 = mul(x = out_53_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_171_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_173_cast_fp16 = mul(x = x_171_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_173_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1972_cast_fp16 = mul(x = x_173_cast_fp16, y = last_1_cast_fp16)[name = string("op_1972_cast_fp16")];
            tensor<int32, [1]> x_last_43_axes_0 = const()[name = string("x_last_43_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_43_keep_dims_0 = const()[name = string("x_last_43_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_43_cast_fp16 = reduce_sum(axes = x_last_43_axes_0, keep_dims = x_last_43_keep_dims_0, x = var_1972_cast_fp16)[name = string("x_last_43_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1975_cast_fp16 = mul(x = x_173_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_1975_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1977_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_43_cast_fp16)[name = string("op_1977_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_329_cast_fp16 = add(x = var_1975_cast_fp16, y = var_1977_cast_fp16)[name = string("input_329_cast_fp16")];
            tensor<int32, [6]> input_331_pad_0 = const()[name = string("input_331_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 16, 16])];
            string input_331_mode_0 = const()[name = string("input_331_mode_0"), val = string("replicate")];
            fp16 const_46_to_fp16 = const()[name = string("const_46_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_331_cast_fp16 = pad(constant_val = const_46_to_fp16, mode = input_331_mode_0, pad = input_331_pad_0, x = input_329_cast_fp16)[name = string("input_331_cast_fp16")];
            string h_129_pad_type_0 = const()[name = string("h_129_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_129_dilations_0 = const()[name = string("h_129_dilations_0"), val = tensor<int32, [1]>([8])];
            int32 h_129_groups_0 = const()[name = string("h_129_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_129_strides_0 = const()[name = string("h_129_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_129_pad_0 = const()[name = string("h_129_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [512, 1, 5]> main_blocks_3_convnext_0_3_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_0_3_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95686272)))];
            tensor<fp16, [512]> main_blocks_3_convnext_0_3_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_3_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95691456)))];
            tensor<fp16, [2, 512, ?]> h_129_cast_fp16 = conv(bias = main_blocks_3_convnext_0_3_dwconv__conv_bias_to_fp16, dilations = h_129_dilations_0, groups = h_129_groups_0, pad = h_129_pad_0, pad_type = h_129_pad_type_0, strides = h_129_strides_0, weight = main_blocks_3_convnext_0_3_dwconv__conv_weight_to_fp16, x = input_331_cast_fp16)[name = string("h_129_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_175_cast_fp16 = mul(x = h_129_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_175_cast_fp16")];
            tensor<int32, [3]> input_333_perm_0 = const()[name = string("input_333_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1995_axes_0 = const()[name = string("op_1995_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_3_convnext_0_3_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_0_3_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95692544)))];
            tensor<fp16, [512]> main_blocks_3_convnext_0_3_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_3_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95693632)))];
            tensor<fp16, [2, ?, 512]> input_333_cast_fp16 = transpose(perm = input_333_perm_0, x = x_175_cast_fp16)[name = string("transpose_28")];
            tensor<fp16, [2, ?, 512]> var_1995_cast_fp16 = layer_norm(axes = var_1995_axes_0, beta = main_blocks_3_convnext_0_3_norm_norm_bias_to_fp16, epsilon = var_1780_to_fp16, gamma = main_blocks_3_convnext_0_3_norm_norm_weight_to_fp16, x = input_333_cast_fp16)[name = string("op_1995_cast_fp16")];
            tensor<int32, [3]> input_335_perm_0 = const()[name = string("input_335_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_131_pad_type_0 = const()[name = string("h_131_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_131_strides_0 = const()[name = string("h_131_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_131_pad_0 = const()[name = string("h_131_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_131_dilations_0 = const()[name = string("h_131_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_131_groups_0 = const()[name = string("h_131_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_3_convnext_0_3_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_0_3_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95694720)))];
            tensor<fp16, [2048]> main_blocks_3_convnext_0_3_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_3_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97791936)))];
            tensor<fp16, [2, 512, ?]> input_335_cast_fp16 = transpose(perm = input_335_perm_0, x = var_1995_cast_fp16)[name = string("transpose_27")];
            tensor<fp16, [2, 2048, ?]> h_131_cast_fp16 = conv(bias = main_blocks_3_convnext_0_3_pwconv1_bias_to_fp16, dilations = h_131_dilations_0, groups = h_131_groups_0, pad = h_131_pad_0, pad_type = h_131_pad_type_0, strides = h_131_strides_0, weight = main_blocks_3_convnext_0_3_pwconv1_weight_to_fp16, x = input_335_cast_fp16)[name = string("h_131_cast_fp16")];
            string input_337_mode_0 = const()[name = string("input_337_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_337_cast_fp16 = gelu(mode = input_337_mode_0, x = h_131_cast_fp16)[name = string("input_337_cast_fp16")];
            string h_133_pad_type_0 = const()[name = string("h_133_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_133_strides_0 = const()[name = string("h_133_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_133_pad_0 = const()[name = string("h_133_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_133_dilations_0 = const()[name = string("h_133_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_133_groups_0 = const()[name = string("h_133_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_2012_weight_0_to_fp16 = const()[name = string("op_2012_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97796096)))];
            tensor<fp16, [512]> var_2012_bias_0_to_fp16 = const()[name = string("op_2012_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99893312)))];
            tensor<fp16, [2, 512, ?]> var_2012_cast_fp16 = conv(bias = var_2012_bias_0_to_fp16, dilations = h_133_dilations_0, groups = h_133_groups_0, pad = h_133_pad_0, pad_type = h_133_pad_type_0, strides = h_133_strides_0, weight = var_2012_weight_0_to_fp16, x = input_337_cast_fp16)[name = string("op_2012_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_55_cast_fp16 = add(x = x_173_cast_fp16, y = var_2012_cast_fp16)[name = string("out_55_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_177_cast_fp16 = mul(x = out_55_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_177_cast_fp16")];
            tensor<fp16, [512, 64]> main_blocks_3_time_cond_linear_linear_weight_to_fp16 = const()[name = string("main_blocks_3_time_cond_linear_linear_weight_to_fp16"), val = tensor<fp16, [512, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99894400)))];
            tensor<fp16, [512]> main_blocks_3_time_cond_linear_linear_bias_to_fp16 = const()[name = string("main_blocks_3_time_cond_linear_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99960000)))];
            tensor<fp16, [2, 512]> linear_29_cast_fp16 = linear(bias = main_blocks_3_time_cond_linear_linear_bias_to_fp16, weight = main_blocks_3_time_cond_linear_linear_weight_to_fp16, x = input_55_cast_fp16)[name = string("linear_29_cast_fp16")];
            tensor<int32, [1]> t_axes_0 = const()[name = string("t_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2, 512, 1]> t_cast_fp16 = expand_dims(axes = t_axes_0, x = linear_29_cast_fp16)[name = string("t_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_2022_cast_fp16 = add(x = x_177_cast_fp16, y = t_cast_fp16)[name = string("op_2022_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_179_cast_fp16 = mul(x = var_2022_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_179_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_181_cast_fp16 = mul(x = x_179_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_181_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_2035_cast_fp16 = mul(x = x_181_cast_fp16, y = last_1_cast_fp16)[name = string("op_2035_cast_fp16")];
            tensor<int32, [1]> x_last_45_axes_0 = const()[name = string("x_last_45_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_45_keep_dims_0 = const()[name = string("x_last_45_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_45_cast_fp16 = reduce_sum(axes = x_last_45_axes_0, keep_dims = x_last_45_keep_dims_0, x = var_2035_cast_fp16)[name = string("x_last_45_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_2038_cast_fp16 = mul(x = x_181_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_2038_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_2040_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_45_cast_fp16)[name = string("op_2040_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_343_cast_fp16 = add(x = var_2038_cast_fp16, y = var_2040_cast_fp16)[name = string("input_343_cast_fp16")];
            tensor<int32, [6]> input_345_pad_0 = const()[name = string("input_345_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_345_mode_0 = const()[name = string("input_345_mode_0"), val = string("replicate")];
            fp16 const_48_to_fp16 = const()[name = string("const_48_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_345_cast_fp16 = pad(constant_val = const_48_to_fp16, mode = input_345_mode_0, pad = input_345_pad_0, x = input_343_cast_fp16)[name = string("input_345_cast_fp16")];
            string h_135_pad_type_0 = const()[name = string("h_135_pad_type_0"), val = string("valid")];
            int32 h_135_groups_0 = const()[name = string("h_135_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_135_strides_0 = const()[name = string("h_135_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_135_pad_0 = const()[name = string("h_135_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_135_dilations_0 = const()[name = string("h_135_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> main_blocks_3_convnext_1_0_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_1_0_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99961088)))];
            tensor<fp16, [512]> main_blocks_3_convnext_1_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_1_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99966272)))];
            tensor<fp16, [2, 512, ?]> h_135_cast_fp16 = conv(bias = main_blocks_3_convnext_1_0_dwconv__conv_bias_to_fp16, dilations = h_135_dilations_0, groups = h_135_groups_0, pad = h_135_pad_0, pad_type = h_135_pad_type_0, strides = h_135_strides_0, weight = main_blocks_3_convnext_1_0_dwconv__conv_weight_to_fp16, x = input_345_cast_fp16)[name = string("h_135_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_183_cast_fp16 = mul(x = h_135_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_183_cast_fp16")];
            tensor<int32, [3]> input_347_perm_0 = const()[name = string("input_347_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_2058_axes_0 = const()[name = string("op_2058_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_3_convnext_1_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_1_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99967360)))];
            tensor<fp16, [512]> main_blocks_3_convnext_1_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_1_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99968448)))];
            tensor<fp16, [2, ?, 512]> input_347_cast_fp16 = transpose(perm = input_347_perm_0, x = x_183_cast_fp16)[name = string("transpose_26")];
            tensor<fp16, [2, ?, 512]> var_2058_cast_fp16 = layer_norm(axes = var_2058_axes_0, beta = main_blocks_3_convnext_1_0_norm_norm_bias_to_fp16, epsilon = var_1780_to_fp16, gamma = main_blocks_3_convnext_1_0_norm_norm_weight_to_fp16, x = input_347_cast_fp16)[name = string("op_2058_cast_fp16")];
            tensor<int32, [3]> input_349_perm_0 = const()[name = string("input_349_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_137_pad_type_0 = const()[name = string("h_137_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_137_strides_0 = const()[name = string("h_137_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_137_pad_0 = const()[name = string("h_137_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_137_dilations_0 = const()[name = string("h_137_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_137_groups_0 = const()[name = string("h_137_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_3_convnext_1_0_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_1_0_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99969536)))];
            tensor<fp16, [2048]> main_blocks_3_convnext_1_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_1_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102066752)))];
            tensor<fp16, [2, 512, ?]> input_349_cast_fp16 = transpose(perm = input_349_perm_0, x = var_2058_cast_fp16)[name = string("transpose_25")];
            tensor<fp16, [2, 2048, ?]> h_137_cast_fp16 = conv(bias = main_blocks_3_convnext_1_0_pwconv1_bias_to_fp16, dilations = h_137_dilations_0, groups = h_137_groups_0, pad = h_137_pad_0, pad_type = h_137_pad_type_0, strides = h_137_strides_0, weight = main_blocks_3_convnext_1_0_pwconv1_weight_to_fp16, x = input_349_cast_fp16)[name = string("h_137_cast_fp16")];
            string input_351_mode_0 = const()[name = string("input_351_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_351_cast_fp16 = gelu(mode = input_351_mode_0, x = h_137_cast_fp16)[name = string("input_351_cast_fp16")];
            string h_139_pad_type_0 = const()[name = string("h_139_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_139_strides_0 = const()[name = string("h_139_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_139_pad_0 = const()[name = string("h_139_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_139_dilations_0 = const()[name = string("h_139_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_139_groups_0 = const()[name = string("h_139_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_2075_weight_0_to_fp16 = const()[name = string("op_2075_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102070912)))];
            tensor<fp16, [512]> var_2075_bias_0_to_fp16 = const()[name = string("op_2075_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104168128)))];
            tensor<fp16, [2, 512, ?]> var_2075_cast_fp16 = conv(bias = var_2075_bias_0_to_fp16, dilations = h_139_dilations_0, groups = h_139_groups_0, pad = h_139_pad_0, pad_type = h_139_pad_type_0, strides = h_139_strides_0, weight = var_2075_weight_0_to_fp16, x = input_351_cast_fp16)[name = string("op_2075_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_57_cast_fp16 = add(x = x_181_cast_fp16, y = var_2075_cast_fp16)[name = string("out_57_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_185_cast_fp16 = mul(x = out_57_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_185_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_m_13_cast_fp16 = mul(x = x_185_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_m_13_cast_fp16")];
            tensor<int32, [3]> var_2085_shape_cast_fp16 = shape(x = x_185_cast_fp16)[name = string("op_2085_shape_cast_fp16")];
            int32 gather_19_axis_0 = const()[name = string("gather_19_axis_0"), val = int32(0)];
            int32 gather_19_batch_dims_0 = const()[name = string("gather_19_batch_dims_0"), val = int32(0)];
            bool gather_19_validate_indices_0 = const()[name = string("gather_19_validate_indices_0"), val = bool(false)];
            string var_2085_shape_cast_fp16_to_uint16_dtype_0 = const()[name = string("op_2085_shape_cast_fp16_to_uint16_dtype_0"), val = string("uint16")];
            uint16 gather_19_indices_0_to_uint16 = const()[name = string("gather_19_indices_0_to_uint16"), val = uint16(2)];
            tensor<uint16, [3]> var_2085_shape_cast_fp16_to_uint16 = cast(dtype = var_2085_shape_cast_fp16_to_uint16_dtype_0, x = var_2085_shape_cast_fp16)[name = string("cast_87")];
            uint16 gather_19_cast_uint16 = gather(axis = gather_19_axis_0, batch_dims = gather_19_batch_dims_0, indices = gather_19_indices_0_to_uint16, validate_indices = gather_19_validate_indices_0, x = var_2085_shape_cast_fp16_to_uint16)[name = string("gather_19_cast_uint16")];
            string gather_19_cast_uint16_to_int32_dtype_0 = const()[name = string("gather_19_cast_uint16_to_int32_dtype_0"), val = string("int32")];
            tensor<int32, [3]> input_353_perm_0 = const()[name = string("input_353_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> main_blocks_3_text_attn_W_query_linear_weight_to_fp16 = const()[name = string("main_blocks_3_text_attn_W_query_linear_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104169216)))];
            tensor<fp16, [512]> main_blocks_3_text_attn_W_query_linear_bias_to_fp16 = const()[name = string("main_blocks_3_text_attn_W_query_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104693568)))];
            tensor<fp16, [2, ?, 512]> input_353_cast_fp16 = transpose(perm = input_353_perm_0, x = x_185_cast_fp16)[name = string("transpose_24")];
            tensor<fp16, [2, ?, 512]> linear_30_cast_fp16 = linear(bias = main_blocks_3_text_attn_W_query_linear_bias_to_fp16, weight = main_blocks_3_text_attn_W_query_linear_weight_to_fp16, x = input_353_cast_fp16)[name = string("linear_30_cast_fp16")];
            tensor<int32, [4]> concat_34x = const()[name = string("concat_34x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_2094_cast_fp16 = reshape(shape = concat_34x, x = linear_30_cast_fp16)[name = string("op_2094_cast_fp16")];
            tensor<int32, [4]> x_187_perm_0 = const()[name = string("x_187_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [512, 256]> main_blocks_3_text_attn_W_key_linear_weight_to_fp16 = const()[name = string("main_blocks_3_text_attn_W_key_linear_weight_to_fp16"), val = tensor<fp16, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104694656)))];
            tensor<fp16, [512]> main_blocks_3_text_attn_W_key_linear_bias_to_fp16 = const()[name = string("main_blocks_3_text_attn_W_key_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104956864)))];
            tensor<fp16, [?, ?, 512]> linear_31_cast_fp16 = linear(bias = main_blocks_3_text_attn_W_key_linear_bias_to_fp16, weight = main_blocks_3_text_attn_W_key_linear_weight_to_fp16, x = input_71_cast_fp16)[name = string("linear_31_cast_fp16")];
            tensor<int32, [4]> concat_35x = const()[name = string("concat_35x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_2102_cast_fp16 = reshape(shape = concat_35x, x = linear_31_cast_fp16)[name = string("op_2102_cast_fp16")];
            tensor<int32, [4]> x_189_perm_0 = const()[name = string("x_189_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [512, 256]> main_blocks_3_text_attn_W_value_linear_weight_to_fp16 = const()[name = string("main_blocks_3_text_attn_W_value_linear_weight_to_fp16"), val = tensor<fp16, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104957952)))];
            tensor<fp16, [512]> main_blocks_3_text_attn_W_value_linear_bias_to_fp16 = const()[name = string("main_blocks_3_text_attn_W_value_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105220160)))];
            tensor<fp16, [?, ?, 512]> linear_32_cast_fp16 = linear(bias = main_blocks_3_text_attn_W_value_linear_bias_to_fp16, weight = main_blocks_3_text_attn_W_value_linear_weight_to_fp16, x = input_71_cast_fp16)[name = string("linear_32_cast_fp16")];
            tensor<int32, [4]> concat_36x = const()[name = string("concat_36x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_2110_cast_fp16 = reshape(shape = concat_36x, x = linear_32_cast_fp16)[name = string("op_2110_cast_fp16")];
            tensor<int32, [4]> v_13_perm_0 = const()[name = string("v_13_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            int32 concat_37_values0_0 = const()[name = string("concat_37_values0_0"), val = int32(1)];
            int32 concat_37_values2_0 = const()[name = string("concat_37_values2_0"), val = int32(1)];
            int32 concat_37_axis_0 = const()[name = string("concat_37_axis_0"), val = int32(0)];
            bool concat_37_interleave_0 = const()[name = string("concat_37_interleave_0"), val = bool(false)];
            int32 gather_19_cast_uint16_to_int32 = cast(dtype = gather_19_cast_uint16_to_int32_dtype_0, x = gather_19_cast_uint16)[name = string("cast_86")];
            tensor<int32, [3]> concat_37 = concat(axis = concat_37_axis_0, interleave = concat_37_interleave_0, values = (concat_37_values0_0, gather_19_cast_uint16_to_int32, concat_37_values2_0))[name = string("concat_37")];
            tensor<int32, [3]> var_2113_begin_0 = const()[name = string("op_2113_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<bool, [3]> var_2113_end_mask_0 = const()[name = string("op_2113_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, ?, 1]> var_2113_cast_fp16 = slice_by_index(begin = var_2113_begin_0, end = concat_37, end_mask = var_2113_end_mask_0, x = main_blocks_0_text_attn_increments_to_fp16)[name = string("op_2113_cast_fp16")];
            tensor<int32, [3]> concat_38 = const()[name = string("concat_38"), val = tensor<int32, [3]>([2, -1, -1])];
            tensor<int32, [3]> shape_7_cast_fp16 = shape(x = var_2113_cast_fp16)[name = string("shape_7_cast_fp16")];
            tensor<bool, [3]> equal_7 = const()[name = string("equal_7"), val = tensor<bool, [3]>([false, true, true])];
            tensor<int32, [3]> select_7 = select(a = shape_7_cast_fp16, b = concat_38, cond = equal_7)[name = string("select_7")];
            tensor<int32, [3]> real_div_7 = real_div(x = select_7, y = shape_7_cast_fp16)[name = string("real_div_7")];
            tensor<fp16, [?, ?, ?]> var_2116_cast_fp16 = tile(reps = real_div_7, x = var_2113_cast_fp16)[name = string("op_2116_cast_fp16")];
            tensor<fp16, [2, ?, ?]> scaled_13_cast_fp16 = real_div(x = var_2116_cast_fp16, y = var_488_cast_fp16)[name = string("scaled_13_cast_fp16")];
            tensor<fp16, [1, 1, 32]> main_blocks_3_text_attn_theta_to_fp16 = const()[name = string("main_blocks_3_text_attn_theta_to_fp16"), val = tensor<fp16, [1, 1, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105221248)))];
            tensor<fp16, [2, ?, 32]> angles_13_cast_fp16 = mul(x = scaled_13_cast_fp16, y = main_blocks_3_text_attn_theta_to_fp16)[name = string("angles_13_cast_fp16")];
            tensor<fp16, [2, ?, 32]> var_2132_cast_fp16 = cos(x = angles_13_cast_fp16)[name = string("op_2132_cast_fp16")];
            tensor<int32, [1]> cos_13_axes_0 = const()[name = string("cos_13_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, ?, 32]> cos_13_cast_fp16 = expand_dims(axes = cos_13_axes_0, x = var_2132_cast_fp16)[name = string("cos_13_cast_fp16")];
            tensor<fp16, [2, ?, 32]> var_2134_cast_fp16 = sin(x = angles_13_cast_fp16)[name = string("op_2134_cast_fp16")];
            tensor<int32, [1]> sin_13_axes_0 = const()[name = string("sin_13_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, ?, 32]> sin_13_cast_fp16 = expand_dims(axes = sin_13_axes_0, x = var_2134_cast_fp16)[name = string("sin_13_cast_fp16")];
            tensor<int32, [4]> x_a_13_begin_0 = const()[name = string("x_a_13_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x_a_13_end_0 = const()[name = string("x_a_13_end_0"), val = tensor<int32, [4]>([2, 8, 0, 32])];
            tensor<bool, [4]> x_a_13_end_mask_0 = const()[name = string("x_a_13_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 8, ?, 64]> x_187_cast_fp16 = transpose(perm = x_187_perm_0, x = var_2094_cast_fp16)[name = string("transpose_23")];
            tensor<fp16, [2, 8, ?, 32]> x_a_13_cast_fp16 = slice_by_index(begin = x_a_13_begin_0, end = x_a_13_end_0, end_mask = x_a_13_end_mask_0, x = x_187_cast_fp16)[name = string("x_a_13_cast_fp16")];
            tensor<int32, [4]> x_b_13_begin_0 = const()[name = string("x_b_13_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x_b_13_end_0 = const()[name = string("x_b_13_end_0"), val = tensor<int32, [4]>([2, 8, 0, 64])];
            tensor<bool, [4]> x_b_13_end_mask_0 = const()[name = string("x_b_13_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [2, 8, ?, 32]> x_b_13_cast_fp16 = slice_by_index(begin = x_b_13_begin_0, end = x_b_13_end_0, end_mask = x_b_13_end_mask_0, x = x_187_cast_fp16)[name = string("x_b_13_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_2138_cast_fp16 = mul(x = x_a_13_cast_fp16, y = cos_13_cast_fp16)[name = string("op_2138_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_2139_cast_fp16 = mul(x = x_b_13_cast_fp16, y = sin_13_cast_fp16)[name = string("op_2139_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_a_13_cast_fp16 = sub(x = var_2138_cast_fp16, y = var_2139_cast_fp16)[name = string("rot_a_13_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_2141_cast_fp16 = mul(x = x_a_13_cast_fp16, y = sin_13_cast_fp16)[name = string("op_2141_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_2142_cast_fp16 = mul(x = x_b_13_cast_fp16, y = cos_13_cast_fp16)[name = string("op_2142_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_b_13_cast_fp16 = add(x = var_2141_cast_fp16, y = var_2142_cast_fp16)[name = string("rot_b_13_cast_fp16")];
            bool q_13_interleave_0 = const()[name = string("q_13_interleave_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> q_13_cast_fp16 = concat(axis = var_1775, interleave = q_13_interleave_0, values = (rot_a_13_cast_fp16, rot_b_13_cast_fp16))[name = string("q_13_cast_fp16")];
            tensor<int32, [4]> x_a_begin_0 = const()[name = string("x_a_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x_a_end_0 = const()[name = string("x_a_end_0"), val = tensor<int32, [4]>([2, 8, 0, 32])];
            tensor<bool, [4]> x_a_end_mask_0 = const()[name = string("x_a_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 8, ?, 64]> x_189_cast_fp16 = transpose(perm = x_189_perm_0, x = var_2102_cast_fp16)[name = string("transpose_22")];
            tensor<fp16, [2, 8, ?, 32]> x_a_cast_fp16 = slice_by_index(begin = x_a_begin_0, end = x_a_end_0, end_mask = x_a_end_mask_0, x = x_189_cast_fp16)[name = string("x_a_cast_fp16")];
            tensor<int32, [4]> x_b_begin_0 = const()[name = string("x_b_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x_b_end_0 = const()[name = string("x_b_end_0"), val = tensor<int32, [4]>([2, 8, 0, 64])];
            tensor<bool, [4]> x_b_end_mask_0 = const()[name = string("x_b_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [2, 8, ?, 32]> x_b_cast_fp16 = slice_by_index(begin = x_b_begin_0, end = x_b_end_0, end_mask = x_b_end_mask_0, x = x_189_cast_fp16)[name = string("x_b_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_2156_cast_fp16 = mul(x = x_a_cast_fp16, y = cos_3_cast_fp16)[name = string("op_2156_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_2157_cast_fp16 = mul(x = x_b_cast_fp16, y = sin_3_cast_fp16)[name = string("op_2157_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_a_cast_fp16 = sub(x = var_2156_cast_fp16, y = var_2157_cast_fp16)[name = string("rot_a_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_2159_cast_fp16 = mul(x = x_a_cast_fp16, y = sin_3_cast_fp16)[name = string("op_2159_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_2160_cast_fp16 = mul(x = x_b_cast_fp16, y = cos_3_cast_fp16)[name = string("op_2160_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_b_cast_fp16 = add(x = var_2159_cast_fp16, y = var_2160_cast_fp16)[name = string("rot_b_cast_fp16")];
            bool k_13_interleave_0 = const()[name = string("k_13_interleave_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> k_13_cast_fp16 = concat(axis = var_1775, interleave = k_13_interleave_0, values = (rot_a_cast_fp16, rot_b_cast_fp16))[name = string("k_13_cast_fp16")];
            bool var_2165_transpose_x_1 = const()[name = string("op_2165_transpose_x_1"), val = bool(false)];
            bool var_2165_transpose_y_1 = const()[name = string("op_2165_transpose_y_1"), val = bool(true)];
            tensor<fp16, [2, 8, ?, ?]> var_2165_cast_fp16 = matmul(transpose_x = var_2165_transpose_x_1, transpose_y = var_2165_transpose_y_1, x = q_13_cast_fp16, y = k_13_cast_fp16)[name = string("op_2165_cast_fp16")];
            fp16 _inversed_scores_y_0_to_fp16 = const()[name = string("_inversed_scores_y_0_to_fp16"), val = fp16(0x1p-4)];
            tensor<fp16, [2, 8, ?, ?]> _inversed_scores_cast_fp16 = mul(x = var_2165_cast_fp16, y = _inversed_scores_y_0_to_fp16)[name = string("_inversed_scores_cast_fp16")];
            tensor<fp16, [2, 8, ?, ?]> input_359_cast_fp16 = sub(x = _inversed_scores_cast_fp16, y = var_530_cast_fp16)[name = string("input_359_cast_fp16")];
            tensor<fp16, [2, 8, ?, ?]> attn_19_cast_fp16 = softmax(axis = var_1775, x = input_359_cast_fp16)[name = string("attn_19_cast_fp16")];
            bool var_2174_transpose_x_0 = const()[name = string("op_2174_transpose_x_0"), val = bool(false)];
            bool var_2174_transpose_y_0 = const()[name = string("op_2174_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> v_13_cast_fp16 = transpose(perm = v_13_perm_0, x = var_2110_cast_fp16)[name = string("transpose_21")];
            tensor<fp16, [2, 8, ?, 64]> var_2174_cast_fp16 = matmul(transpose_x = var_2174_transpose_x_0, transpose_y = var_2174_transpose_y_0, x = attn_19_cast_fp16, y = v_13_cast_fp16)[name = string("op_2174_cast_fp16")];
            tensor<int32, [4]> var_2175_perm_0 = const()[name = string("op_2175_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_41x = const()[name = string("concat_41x"), val = tensor<int32, [3]>([2, -1, 512])];
            tensor<fp16, [2, ?, 8, 64]> var_2175_cast_fp16 = transpose(perm = var_2175_perm_0, x = var_2174_cast_fp16)[name = string("transpose_20")];
            tensor<fp16, [2, ?, 512]> input_361_cast_fp16 = reshape(shape = concat_41x, x = var_2175_cast_fp16)[name = string("input_361_cast_fp16")];
            tensor<fp16, [512, 512]> main_blocks_3_text_attn_out_fc_linear_weight_to_fp16 = const()[name = string("main_blocks_3_text_attn_out_fc_linear_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105221376)))];
            tensor<fp16, [512]> main_blocks_3_text_attn_out_fc_linear_bias_to_fp16 = const()[name = string("main_blocks_3_text_attn_out_fc_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105745728)))];
            tensor<fp16, [2, ?, 512]> linear_33_cast_fp16 = linear(bias = main_blocks_3_text_attn_out_fc_linear_bias_to_fp16, weight = main_blocks_3_text_attn_out_fc_linear_weight_to_fp16, x = input_361_cast_fp16)[name = string("linear_33_cast_fp16")];
            tensor<int32, [3]> out_59_perm_0 = const()[name = string("out_59_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> out_59_cast_fp16 = transpose(perm = out_59_perm_0, x = linear_33_cast_fp16)[name = string("transpose_19")];
            tensor<fp16, [2, 512, ?]> var_2183_cast_fp16 = mul(x = out_59_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_2183_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_191_cast_fp16 = add(x = x_m_13_cast_fp16, y = var_2183_cast_fp16)[name = string("x_191_cast_fp16")];
            tensor<int32, [3]> input_363_perm_0 = const()[name = string("input_363_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_2190_axes_0 = const()[name = string("op_2190_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_3_text_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_3_text_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105746816)))];
            tensor<fp16, [512]> main_blocks_3_text_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_3_text_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105747904)))];
            tensor<fp16, [2, ?, 512]> input_363_cast_fp16 = transpose(perm = input_363_perm_0, x = x_191_cast_fp16)[name = string("transpose_18")];
            tensor<fp16, [2, ?, 512]> var_2190_cast_fp16 = layer_norm(axes = var_2190_axes_0, beta = main_blocks_3_text_norm_norm_bias_to_fp16, epsilon = var_1780_to_fp16, gamma = main_blocks_3_text_norm_norm_weight_to_fp16, x = input_363_cast_fp16)[name = string("op_2190_cast_fp16")];
            tensor<int32, [3]> var_2191_perm_0 = const()[name = string("op_2191_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> var_2191_cast_fp16 = transpose(perm = var_2191_perm_0, x = var_2190_cast_fp16)[name = string("transpose_17")];
            tensor<fp16, [2, 512, ?]> x_193_cast_fp16 = mul(x = var_2191_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_193_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_195_cast_fp16 = mul(x = x_193_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_195_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_2204_cast_fp16 = mul(x = x_195_cast_fp16, y = last_1_cast_fp16)[name = string("op_2204_cast_fp16")];
            tensor<int32, [1]> x_last_47_axes_0 = const()[name = string("x_last_47_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_47_keep_dims_0 = const()[name = string("x_last_47_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_47_cast_fp16 = reduce_sum(axes = x_last_47_axes_0, keep_dims = x_last_47_keep_dims_0, x = var_2204_cast_fp16)[name = string("x_last_47_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_2207_cast_fp16 = mul(x = x_195_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_2207_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_2209_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_47_cast_fp16)[name = string("op_2209_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_367_cast_fp16 = add(x = var_2207_cast_fp16, y = var_2209_cast_fp16)[name = string("input_367_cast_fp16")];
            tensor<int32, [6]> input_369_pad_0 = const()[name = string("input_369_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_369_mode_0 = const()[name = string("input_369_mode_0"), val = string("replicate")];
            fp16 const_50_to_fp16 = const()[name = string("const_50_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_369_cast_fp16 = pad(constant_val = const_50_to_fp16, mode = input_369_mode_0, pad = input_369_pad_0, x = input_367_cast_fp16)[name = string("input_369_cast_fp16")];
            string h_141_pad_type_0 = const()[name = string("h_141_pad_type_0"), val = string("valid")];
            int32 h_141_groups_0 = const()[name = string("h_141_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_141_strides_0 = const()[name = string("h_141_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_141_pad_0 = const()[name = string("h_141_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_141_dilations_0 = const()[name = string("h_141_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> main_blocks_3_convnext_2_0_dwconv__conv_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_2_0_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105748992)))];
            tensor<fp16, [512]> main_blocks_3_convnext_2_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_2_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105754176)))];
            tensor<fp16, [2, 512, ?]> h_141_cast_fp16 = conv(bias = main_blocks_3_convnext_2_0_dwconv__conv_bias_to_fp16, dilations = h_141_dilations_0, groups = h_141_groups_0, pad = h_141_pad_0, pad_type = h_141_pad_type_0, strides = h_141_strides_0, weight = main_blocks_3_convnext_2_0_dwconv__conv_weight_to_fp16, x = input_369_cast_fp16)[name = string("h_141_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_197_cast_fp16 = mul(x = h_141_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_197_cast_fp16")];
            tensor<int32, [3]> input_371_perm_0 = const()[name = string("input_371_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_2227_axes_0 = const()[name = string("op_2227_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_3_convnext_2_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_2_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105755264)))];
            tensor<fp16, [512]> main_blocks_3_convnext_2_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_2_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105756352)))];
            tensor<fp16, [2, ?, 512]> input_371_cast_fp16 = transpose(perm = input_371_perm_0, x = x_197_cast_fp16)[name = string("transpose_16")];
            tensor<fp16, [2, ?, 512]> var_2227_cast_fp16 = layer_norm(axes = var_2227_axes_0, beta = main_blocks_3_convnext_2_0_norm_norm_bias_to_fp16, epsilon = var_1780_to_fp16, gamma = main_blocks_3_convnext_2_0_norm_norm_weight_to_fp16, x = input_371_cast_fp16)[name = string("op_2227_cast_fp16")];
            tensor<int32, [3]> input_373_perm_0 = const()[name = string("input_373_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_143_pad_type_0 = const()[name = string("h_143_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_143_strides_0 = const()[name = string("h_143_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_143_pad_0 = const()[name = string("h_143_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_143_dilations_0 = const()[name = string("h_143_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_143_groups_0 = const()[name = string("h_143_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_3_convnext_2_0_pwconv1_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_2_0_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105757440)))];
            tensor<fp16, [2048]> main_blocks_3_convnext_2_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_2_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107854656)))];
            tensor<fp16, [2, 512, ?]> input_373_cast_fp16 = transpose(perm = input_373_perm_0, x = var_2227_cast_fp16)[name = string("transpose_15")];
            tensor<fp16, [2, 2048, ?]> h_143_cast_fp16 = conv(bias = main_blocks_3_convnext_2_0_pwconv1_bias_to_fp16, dilations = h_143_dilations_0, groups = h_143_groups_0, pad = h_143_pad_0, pad_type = h_143_pad_type_0, strides = h_143_strides_0, weight = main_blocks_3_convnext_2_0_pwconv1_weight_to_fp16, x = input_373_cast_fp16)[name = string("h_143_cast_fp16")];
            string input_375_mode_0 = const()[name = string("input_375_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_375_cast_fp16 = gelu(mode = input_375_mode_0, x = h_143_cast_fp16)[name = string("input_375_cast_fp16")];
            string h_145_pad_type_0 = const()[name = string("h_145_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_145_strides_0 = const()[name = string("h_145_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_145_pad_0 = const()[name = string("h_145_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_145_dilations_0 = const()[name = string("h_145_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_145_groups_0 = const()[name = string("h_145_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_2244_weight_0_to_fp16 = const()[name = string("op_2244_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107858816)))];
            tensor<fp16, [512]> var_2244_bias_0_to_fp16 = const()[name = string("op_2244_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109956032)))];
            tensor<fp16, [2, 512, ?]> var_2244_cast_fp16 = conv(bias = var_2244_bias_0_to_fp16, dilations = h_145_dilations_0, groups = h_145_groups_0, pad = h_145_pad_0, pad_type = h_145_pad_type_0, strides = h_145_strides_0, weight = var_2244_weight_0_to_fp16, x = input_375_cast_fp16)[name = string("op_2244_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_61_cast_fp16 = add(x = x_195_cast_fp16, y = var_2244_cast_fp16)[name = string("out_61_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_199_cast_fp16 = mul(x = out_61_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_199_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_m_cast_fp16 = mul(x = x_199_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_m_cast_fp16")];
            tensor<int32, [3]> input_377_perm_0 = const()[name = string("input_377_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [256, 512]> main_blocks_3_style_attn_W_query_linear_weight_to_fp16 = const()[name = string("main_blocks_3_style_attn_W_query_linear_weight_to_fp16"), val = tensor<fp16, [256, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109957120)))];
            tensor<fp16, [256]> main_blocks_3_style_attn_W_query_linear_bias_to_fp16 = const()[name = string("main_blocks_3_style_attn_W_query_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110219328)))];
            tensor<fp16, [2, ?, 512]> input_377_cast_fp16 = transpose(perm = input_377_perm_0, x = x_199_cast_fp16)[name = string("transpose_14")];
            tensor<fp16, [2, ?, 256]> linear_34_cast_fp16 = linear(bias = main_blocks_3_style_attn_W_query_linear_bias_to_fp16, weight = main_blocks_3_style_attn_W_query_linear_weight_to_fp16, x = input_377_cast_fp16)[name = string("linear_34_cast_fp16")];
            tensor<int32, [4]> concat_42x = const()[name = string("concat_42x"), val = tensor<int32, [4]>([2, -1, 2, 128])];
            tensor<fp16, [2, ?, 2, 128]> var_2261_cast_fp16 = reshape(shape = concat_42x, x = linear_34_cast_fp16)[name = string("op_2261_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]> main_blocks_3_style_attn_W_value_linear_weight_to_fp16 = const()[name = string("main_blocks_3_style_attn_W_value_linear_weight_to_fp16"), val = tensor<fp16, [256, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110219904)))];
            tensor<fp16, [256]> main_blocks_3_style_attn_W_value_linear_bias_to_fp16 = const()[name = string("main_blocks_3_style_attn_W_value_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110351040)))];
            tensor<fp16, [2, 50, 256]> linear_36_cast_fp16 = linear(bias = main_blocks_3_style_attn_W_value_linear_bias_to_fp16, weight = main_blocks_3_style_attn_W_value_linear_weight_to_fp16, x = input_95_cast_fp16)[name = string("linear_36_cast_fp16")];
            tensor<int32, [4]> var_2274 = const()[name = string("op_2274"), val = tensor<int32, [4]>([2, 50, 2, 128])];
            tensor<fp16, [2, 50, 2, 128]> var_2275_cast_fp16 = reshape(shape = var_2274, x = linear_36_cast_fp16)[name = string("op_2275_cast_fp16")];
            tensor<int32, [4]> v_15_perm_0 = const()[name = string("v_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_2279_transpose_x_0 = const()[name = string("op_2279_transpose_x_0"), val = bool(false)];
            bool var_2279_transpose_y_0 = const()[name = string("op_2279_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 2, 128, 50]> var_2278_to_fp16 = const()[name = string("op_2278_to_fp16"), val = tensor<fp16, [2, 2, 128, 50]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110351616)))];
            tensor<fp16, [2, 2, ?, 128]> q_cast_fp16 = transpose(perm = q_perm_0, x = var_2261_cast_fp16)[name = string("transpose_13")];
            tensor<fp16, [2, 2, ?, 50]> var_2279_cast_fp16 = matmul(transpose_x = var_2279_transpose_x_0, transpose_y = var_2279_transpose_y_0, x = q_cast_fp16, y = var_2278_to_fp16)[name = string("op_2279_cast_fp16")];
            fp16 _inversed_input_379_y_0_to_fp16 = const()[name = string("_inversed_input_379_y_0_to_fp16"), val = fp16(0x1p-4)];
            tensor<fp16, [2, 2, ?, 50]> _inversed_input_379_cast_fp16 = mul(x = var_2279_cast_fp16, y = _inversed_input_379_y_0_to_fp16)[name = string("_inversed_input_379_cast_fp16")];
            tensor<fp16, [2, 2, ?, 50]> attn_21_cast_fp16 = softmax(axis = var_1775, x = _inversed_input_379_cast_fp16)[name = string("attn_21_cast_fp16")];
            tensor<fp16, [2, 2, ?, 50]> attn_cast_fp16 = mul(x = attn_21_cast_fp16, y = mask_1_cast_fp16)[name = string("attn_cast_fp16")];
            bool var_2286_transpose_x_0 = const()[name = string("op_2286_transpose_x_0"), val = bool(false)];
            bool var_2286_transpose_y_0 = const()[name = string("op_2286_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 2, 50, 128]> v_15_cast_fp16 = transpose(perm = v_15_perm_0, x = var_2275_cast_fp16)[name = string("transpose_12")];
            tensor<fp16, [2, 2, ?, 128]> var_2286_cast_fp16 = matmul(transpose_x = var_2286_transpose_x_0, transpose_y = var_2286_transpose_y_0, x = attn_cast_fp16, y = v_15_cast_fp16)[name = string("op_2286_cast_fp16")];
            tensor<int32, [4]> var_2287_perm_0 = const()[name = string("op_2287_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_43x = const()[name = string("concat_43x"), val = tensor<int32, [3]>([2, -1, 256])];
            tensor<fp16, [2, ?, 2, 128]> var_2287_cast_fp16 = transpose(perm = var_2287_perm_0, x = var_2286_cast_fp16)[name = string("transpose_11")];
            tensor<fp16, [2, ?, 256]> input_381_cast_fp16 = reshape(shape = concat_43x, x = var_2287_cast_fp16)[name = string("input_381_cast_fp16")];
            tensor<fp16, [512, 256]> main_blocks_3_style_attn_out_fc_linear_weight_to_fp16 = const()[name = string("main_blocks_3_style_attn_out_fc_linear_weight_to_fp16"), val = tensor<fp16, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110402880)))];
            tensor<fp16, [512]> main_blocks_3_style_attn_out_fc_linear_bias_to_fp16 = const()[name = string("main_blocks_3_style_attn_out_fc_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110665088)))];
            tensor<fp16, [2, ?, 512]> linear_37_cast_fp16 = linear(bias = main_blocks_3_style_attn_out_fc_linear_bias_to_fp16, weight = main_blocks_3_style_attn_out_fc_linear_weight_to_fp16, x = input_381_cast_fp16)[name = string("linear_37_cast_fp16")];
            tensor<int32, [3]> out_63_perm_0 = const()[name = string("out_63_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> out_63_cast_fp16 = transpose(perm = out_63_perm_0, x = linear_37_cast_fp16)[name = string("transpose_10")];
            tensor<fp16, [2, 512, ?]> var_2295_cast_fp16 = mul(x = out_63_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_2295_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_201_cast_fp16 = add(x = x_m_cast_fp16, y = var_2295_cast_fp16)[name = string("x_201_cast_fp16")];
            tensor<int32, [3]> input_383_perm_0 = const()[name = string("input_383_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_2302_axes_0 = const()[name = string("op_2302_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_3_style_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_3_style_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110666176)))];
            tensor<fp16, [512]> main_blocks_3_style_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_3_style_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110667264)))];
            tensor<fp16, [2, ?, 512]> input_383_cast_fp16 = transpose(perm = input_383_perm_0, x = x_201_cast_fp16)[name = string("transpose_9")];
            tensor<fp16, [2, ?, 512]> var_2302_cast_fp16 = layer_norm(axes = var_2302_axes_0, beta = main_blocks_3_style_norm_norm_bias_to_fp16, epsilon = var_1780_to_fp16, gamma = main_blocks_3_style_norm_norm_weight_to_fp16, x = input_383_cast_fp16)[name = string("op_2302_cast_fp16")];
            tensor<int32, [3]> var_2303_perm_0 = const()[name = string("op_2303_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> var_2303_cast_fp16 = transpose(perm = var_2303_perm_0, x = var_2302_cast_fp16)[name = string("transpose_8")];
            tensor<fp16, [2, 512, ?]> x_203_cast_fp16 = mul(x = var_2303_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_203_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_205_cast_fp16 = mul(x = x_203_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_205_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_2330_cast_fp16 = mul(x = x_205_cast_fp16, y = last_1_cast_fp16)[name = string("op_2330_cast_fp16")];
            tensor<int32, [1]> x_last_49_axes_0 = const()[name = string("x_last_49_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_49_keep_dims_0 = const()[name = string("x_last_49_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_49_cast_fp16 = reduce_sum(axes = x_last_49_axes_0, keep_dims = x_last_49_keep_dims_0, x = var_2330_cast_fp16)[name = string("x_last_49_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_2333_cast_fp16 = mul(x = x_205_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_2333_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_2335_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_49_cast_fp16)[name = string("op_2335_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_387_cast_fp16 = add(x = var_2333_cast_fp16, y = var_2335_cast_fp16)[name = string("input_387_cast_fp16")];
            tensor<int32, [6]> input_389_pad_0 = const()[name = string("input_389_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_389_mode_0 = const()[name = string("input_389_mode_0"), val = string("replicate")];
            fp16 const_53_to_fp16 = const()[name = string("const_53_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_389_cast_fp16 = pad(constant_val = const_53_to_fp16, mode = input_389_mode_0, pad = input_389_pad_0, x = input_387_cast_fp16)[name = string("input_389_cast_fp16")];
            string h_147_pad_type_0 = const()[name = string("h_147_pad_type_0"), val = string("valid")];
            int32 h_147_groups_0 = const()[name = string("h_147_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_147_strides_0 = const()[name = string("h_147_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_147_pad_0 = const()[name = string("h_147_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_147_dilations_0 = const()[name = string("h_147_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> last_convnext_0_dwconv__conv_weight_to_fp16 = const()[name = string("last_convnext_0_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110668352)))];
            tensor<fp16, [512]> last_convnext_0_dwconv__conv_bias_to_fp16 = const()[name = string("last_convnext_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110673536)))];
            tensor<fp16, [2, 512, ?]> h_147_cast_fp16 = conv(bias = last_convnext_0_dwconv__conv_bias_to_fp16, dilations = h_147_dilations_0, groups = h_147_groups_0, pad = h_147_pad_0, pad_type = h_147_pad_type_0, strides = h_147_strides_0, weight = last_convnext_0_dwconv__conv_weight_to_fp16, x = input_389_cast_fp16)[name = string("h_147_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_207_cast_fp16 = mul(x = h_147_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_207_cast_fp16")];
            tensor<int32, [3]> input_391_perm_0 = const()[name = string("input_391_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_2353_axes_0 = const()[name = string("op_2353_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> last_convnext_0_norm_norm_weight_to_fp16 = const()[name = string("last_convnext_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110674624)))];
            tensor<fp16, [512]> last_convnext_0_norm_norm_bias_to_fp16 = const()[name = string("last_convnext_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110675712)))];
            fp16 var_2307_to_fp16 = const()[name = string("op_2307_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [2, ?, 512]> input_391_cast_fp16 = transpose(perm = input_391_perm_0, x = x_207_cast_fp16)[name = string("transpose_7")];
            tensor<fp16, [2, ?, 512]> var_2353_cast_fp16 = layer_norm(axes = var_2353_axes_0, beta = last_convnext_0_norm_norm_bias_to_fp16, epsilon = var_2307_to_fp16, gamma = last_convnext_0_norm_norm_weight_to_fp16, x = input_391_cast_fp16)[name = string("op_2353_cast_fp16")];
            tensor<int32, [3]> input_393_perm_0 = const()[name = string("input_393_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_149_pad_type_0 = const()[name = string("h_149_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_149_strides_0 = const()[name = string("h_149_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_149_pad_0 = const()[name = string("h_149_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_149_dilations_0 = const()[name = string("h_149_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_149_groups_0 = const()[name = string("h_149_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> last_convnext_0_pwconv1_weight_to_fp16 = const()[name = string("last_convnext_0_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110676800)))];
            tensor<fp16, [2048]> last_convnext_0_pwconv1_bias_to_fp16 = const()[name = string("last_convnext_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112774016)))];
            tensor<fp16, [2, 512, ?]> input_393_cast_fp16 = transpose(perm = input_393_perm_0, x = var_2353_cast_fp16)[name = string("transpose_6")];
            tensor<fp16, [2, 2048, ?]> h_149_cast_fp16 = conv(bias = last_convnext_0_pwconv1_bias_to_fp16, dilations = h_149_dilations_0, groups = h_149_groups_0, pad = h_149_pad_0, pad_type = h_149_pad_type_0, strides = h_149_strides_0, weight = last_convnext_0_pwconv1_weight_to_fp16, x = input_393_cast_fp16)[name = string("h_149_cast_fp16")];
            string input_395_mode_0 = const()[name = string("input_395_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_395_cast_fp16 = gelu(mode = input_395_mode_0, x = h_149_cast_fp16)[name = string("input_395_cast_fp16")];
            string h_151_pad_type_0 = const()[name = string("h_151_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_151_strides_0 = const()[name = string("h_151_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_151_pad_0 = const()[name = string("h_151_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_151_dilations_0 = const()[name = string("h_151_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_151_groups_0 = const()[name = string("h_151_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_2370_weight_0_to_fp16 = const()[name = string("op_2370_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112778176)))];
            tensor<fp16, [512]> var_2370_bias_0_to_fp16 = const()[name = string("op_2370_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114875392)))];
            tensor<fp16, [2, 512, ?]> var_2370_cast_fp16 = conv(bias = var_2370_bias_0_to_fp16, dilations = h_151_dilations_0, groups = h_151_groups_0, pad = h_151_pad_0, pad_type = h_151_pad_type_0, strides = h_151_strides_0, weight = var_2370_weight_0_to_fp16, x = input_395_cast_fp16)[name = string("op_2370_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_65_cast_fp16 = add(x = x_205_cast_fp16, y = var_2370_cast_fp16)[name = string("out_65_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_209_cast_fp16 = mul(x = out_65_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_209_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_211_cast_fp16 = mul(x = x_209_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_211_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_2398_cast_fp16 = mul(x = x_211_cast_fp16, y = last_1_cast_fp16)[name = string("op_2398_cast_fp16")];
            tensor<int32, [1]> x_last_51_axes_0 = const()[name = string("x_last_51_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_51_keep_dims_0 = const()[name = string("x_last_51_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_51_cast_fp16 = reduce_sum(axes = x_last_51_axes_0, keep_dims = x_last_51_keep_dims_0, x = var_2398_cast_fp16)[name = string("x_last_51_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_2401_cast_fp16 = mul(x = x_211_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_2401_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_2403_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_51_cast_fp16)[name = string("op_2403_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_399_cast_fp16 = add(x = var_2401_cast_fp16, y = var_2403_cast_fp16)[name = string("input_399_cast_fp16")];
            tensor<int32, [6]> input_401_pad_0 = const()[name = string("input_401_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_401_mode_0 = const()[name = string("input_401_mode_0"), val = string("replicate")];
            fp16 const_55_to_fp16 = const()[name = string("const_55_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_401_cast_fp16 = pad(constant_val = const_55_to_fp16, mode = input_401_mode_0, pad = input_401_pad_0, x = input_399_cast_fp16)[name = string("input_401_cast_fp16")];
            string h_153_pad_type_0 = const()[name = string("h_153_pad_type_0"), val = string("valid")];
            int32 h_153_groups_0 = const()[name = string("h_153_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_153_strides_0 = const()[name = string("h_153_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_153_pad_0 = const()[name = string("h_153_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_153_dilations_0 = const()[name = string("h_153_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> last_convnext_1_dwconv__conv_weight_to_fp16 = const()[name = string("last_convnext_1_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114876480)))];
            tensor<fp16, [512]> last_convnext_1_dwconv__conv_bias_to_fp16 = const()[name = string("last_convnext_1_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114881664)))];
            tensor<fp16, [2, 512, ?]> h_153_cast_fp16 = conv(bias = last_convnext_1_dwconv__conv_bias_to_fp16, dilations = h_153_dilations_0, groups = h_153_groups_0, pad = h_153_pad_0, pad_type = h_153_pad_type_0, strides = h_153_strides_0, weight = last_convnext_1_dwconv__conv_weight_to_fp16, x = input_401_cast_fp16)[name = string("h_153_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_213_cast_fp16 = mul(x = h_153_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_213_cast_fp16")];
            tensor<int32, [3]> input_403_perm_0 = const()[name = string("input_403_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_2421_axes_0 = const()[name = string("op_2421_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> last_convnext_1_norm_norm_weight_to_fp16 = const()[name = string("last_convnext_1_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114882752)))];
            tensor<fp16, [512]> last_convnext_1_norm_norm_bias_to_fp16 = const()[name = string("last_convnext_1_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114883840)))];
            fp16 var_2375_to_fp16 = const()[name = string("op_2375_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [2, ?, 512]> input_403_cast_fp16 = transpose(perm = input_403_perm_0, x = x_213_cast_fp16)[name = string("transpose_5")];
            tensor<fp16, [2, ?, 512]> var_2421_cast_fp16 = layer_norm(axes = var_2421_axes_0, beta = last_convnext_1_norm_norm_bias_to_fp16, epsilon = var_2375_to_fp16, gamma = last_convnext_1_norm_norm_weight_to_fp16, x = input_403_cast_fp16)[name = string("op_2421_cast_fp16")];
            tensor<int32, [3]> input_405_perm_0 = const()[name = string("input_405_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_155_pad_type_0 = const()[name = string("h_155_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_155_strides_0 = const()[name = string("h_155_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_155_pad_0 = const()[name = string("h_155_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_155_dilations_0 = const()[name = string("h_155_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_155_groups_0 = const()[name = string("h_155_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> last_convnext_1_pwconv1_weight_to_fp16 = const()[name = string("last_convnext_1_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114884928)))];
            tensor<fp16, [2048]> last_convnext_1_pwconv1_bias_to_fp16 = const()[name = string("last_convnext_1_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116982144)))];
            tensor<fp16, [2, 512, ?]> input_405_cast_fp16 = transpose(perm = input_405_perm_0, x = var_2421_cast_fp16)[name = string("transpose_4")];
            tensor<fp16, [2, 2048, ?]> h_155_cast_fp16 = conv(bias = last_convnext_1_pwconv1_bias_to_fp16, dilations = h_155_dilations_0, groups = h_155_groups_0, pad = h_155_pad_0, pad_type = h_155_pad_type_0, strides = h_155_strides_0, weight = last_convnext_1_pwconv1_weight_to_fp16, x = input_405_cast_fp16)[name = string("h_155_cast_fp16")];
            string input_407_mode_0 = const()[name = string("input_407_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_407_cast_fp16 = gelu(mode = input_407_mode_0, x = h_155_cast_fp16)[name = string("input_407_cast_fp16")];
            string h_157_pad_type_0 = const()[name = string("h_157_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_157_strides_0 = const()[name = string("h_157_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_157_pad_0 = const()[name = string("h_157_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_157_dilations_0 = const()[name = string("h_157_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_157_groups_0 = const()[name = string("h_157_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_2438_weight_0_to_fp16 = const()[name = string("op_2438_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116986304)))];
            tensor<fp16, [512]> var_2438_bias_0_to_fp16 = const()[name = string("op_2438_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119083520)))];
            tensor<fp16, [2, 512, ?]> var_2438_cast_fp16 = conv(bias = var_2438_bias_0_to_fp16, dilations = h_157_dilations_0, groups = h_157_groups_0, pad = h_157_pad_0, pad_type = h_157_pad_type_0, strides = h_157_strides_0, weight = var_2438_weight_0_to_fp16, x = input_407_cast_fp16)[name = string("op_2438_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_67_cast_fp16 = add(x = x_211_cast_fp16, y = var_2438_cast_fp16)[name = string("out_67_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_215_cast_fp16 = mul(x = out_67_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_215_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_217_cast_fp16 = mul(x = x_215_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_217_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_2466_cast_fp16 = mul(x = x_217_cast_fp16, y = last_1_cast_fp16)[name = string("op_2466_cast_fp16")];
            tensor<int32, [1]> x_last_53_axes_0 = const()[name = string("x_last_53_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_53_keep_dims_0 = const()[name = string("x_last_53_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_53_cast_fp16 = reduce_sum(axes = x_last_53_axes_0, keep_dims = x_last_53_keep_dims_0, x = var_2466_cast_fp16)[name = string("x_last_53_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_2469_cast_fp16 = mul(x = x_217_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_2469_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_2471_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_53_cast_fp16)[name = string("op_2471_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_411_cast_fp16 = add(x = var_2469_cast_fp16, y = var_2471_cast_fp16)[name = string("input_411_cast_fp16")];
            tensor<int32, [6]> input_413_pad_0 = const()[name = string("input_413_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_413_mode_0 = const()[name = string("input_413_mode_0"), val = string("replicate")];
            fp16 const_57_to_fp16 = const()[name = string("const_57_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_413_cast_fp16 = pad(constant_val = const_57_to_fp16, mode = input_413_mode_0, pad = input_413_pad_0, x = input_411_cast_fp16)[name = string("input_413_cast_fp16")];
            string h_159_pad_type_0 = const()[name = string("h_159_pad_type_0"), val = string("valid")];
            int32 h_159_groups_0 = const()[name = string("h_159_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_159_strides_0 = const()[name = string("h_159_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_159_pad_0 = const()[name = string("h_159_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_159_dilations_0 = const()[name = string("h_159_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> last_convnext_2_dwconv__conv_weight_to_fp16 = const()[name = string("last_convnext_2_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119084608)))];
            tensor<fp16, [512]> last_convnext_2_dwconv__conv_bias_to_fp16 = const()[name = string("last_convnext_2_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119089792)))];
            tensor<fp16, [2, 512, ?]> h_159_cast_fp16 = conv(bias = last_convnext_2_dwconv__conv_bias_to_fp16, dilations = h_159_dilations_0, groups = h_159_groups_0, pad = h_159_pad_0, pad_type = h_159_pad_type_0, strides = h_159_strides_0, weight = last_convnext_2_dwconv__conv_weight_to_fp16, x = input_413_cast_fp16)[name = string("h_159_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_219_cast_fp16 = mul(x = h_159_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_219_cast_fp16")];
            tensor<int32, [3]> input_415_perm_0 = const()[name = string("input_415_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_2489_axes_0 = const()[name = string("op_2489_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> last_convnext_2_norm_norm_weight_to_fp16 = const()[name = string("last_convnext_2_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119090880)))];
            tensor<fp16, [512]> last_convnext_2_norm_norm_bias_to_fp16 = const()[name = string("last_convnext_2_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119091968)))];
            fp16 var_2443_to_fp16 = const()[name = string("op_2443_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [2, ?, 512]> input_415_cast_fp16 = transpose(perm = input_415_perm_0, x = x_219_cast_fp16)[name = string("transpose_3")];
            tensor<fp16, [2, ?, 512]> var_2489_cast_fp16 = layer_norm(axes = var_2489_axes_0, beta = last_convnext_2_norm_norm_bias_to_fp16, epsilon = var_2443_to_fp16, gamma = last_convnext_2_norm_norm_weight_to_fp16, x = input_415_cast_fp16)[name = string("op_2489_cast_fp16")];
            tensor<int32, [3]> input_417_perm_0 = const()[name = string("input_417_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_161_pad_type_0 = const()[name = string("h_161_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_161_strides_0 = const()[name = string("h_161_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_161_pad_0 = const()[name = string("h_161_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_161_dilations_0 = const()[name = string("h_161_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_161_groups_0 = const()[name = string("h_161_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> last_convnext_2_pwconv1_weight_to_fp16 = const()[name = string("last_convnext_2_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119093056)))];
            tensor<fp16, [2048]> last_convnext_2_pwconv1_bias_to_fp16 = const()[name = string("last_convnext_2_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121190272)))];
            tensor<fp16, [2, 512, ?]> input_417_cast_fp16 = transpose(perm = input_417_perm_0, x = var_2489_cast_fp16)[name = string("transpose_2")];
            tensor<fp16, [2, 2048, ?]> h_161_cast_fp16 = conv(bias = last_convnext_2_pwconv1_bias_to_fp16, dilations = h_161_dilations_0, groups = h_161_groups_0, pad = h_161_pad_0, pad_type = h_161_pad_type_0, strides = h_161_strides_0, weight = last_convnext_2_pwconv1_weight_to_fp16, x = input_417_cast_fp16)[name = string("h_161_cast_fp16")];
            string input_419_mode_0 = const()[name = string("input_419_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_419_cast_fp16 = gelu(mode = input_419_mode_0, x = h_161_cast_fp16)[name = string("input_419_cast_fp16")];
            string h_163_pad_type_0 = const()[name = string("h_163_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_163_strides_0 = const()[name = string("h_163_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_163_pad_0 = const()[name = string("h_163_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_163_dilations_0 = const()[name = string("h_163_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_163_groups_0 = const()[name = string("h_163_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> var_2506_weight_0_to_fp16 = const()[name = string("op_2506_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121194432)))];
            tensor<fp16, [512]> var_2506_bias_0_to_fp16 = const()[name = string("op_2506_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123291648)))];
            tensor<fp16, [2, 512, ?]> var_2506_cast_fp16 = conv(bias = var_2506_bias_0_to_fp16, dilations = h_163_dilations_0, groups = h_163_groups_0, pad = h_163_pad_0, pad_type = h_163_pad_type_0, strides = h_163_strides_0, weight = var_2506_weight_0_to_fp16, x = input_419_cast_fp16)[name = string("op_2506_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_69_cast_fp16 = add(x = x_217_cast_fp16, y = var_2506_cast_fp16)[name = string("out_69_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_221_cast_fp16 = mul(x = out_69_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_221_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_223_cast_fp16 = mul(x = x_221_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_223_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_2534_cast_fp16 = mul(x = x_223_cast_fp16, y = last_1_cast_fp16)[name = string("op_2534_cast_fp16")];
            tensor<int32, [1]> x_last_axes_0 = const()[name = string("x_last_axes_0"), val = tensor<int32, [1]>([-1])];
            bool x_last_keep_dims_0 = const()[name = string("x_last_keep_dims_0"), val = bool(true)];
            tensor<fp16, [2, 512, 1]> x_last_cast_fp16 = reduce_sum(axes = x_last_axes_0, keep_dims = x_last_keep_dims_0, x = var_2534_cast_fp16)[name = string("x_last_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_2537_cast_fp16 = mul(x = x_223_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_2537_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_2539_cast_fp16 = mul(x = var_173_cast_fp16, y = x_last_cast_fp16)[name = string("op_2539_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_423_cast_fp16 = add(x = var_2537_cast_fp16, y = var_2539_cast_fp16)[name = string("input_423_cast_fp16")];
            tensor<int32, [6]> input_425_pad_0 = const()[name = string("input_425_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_425_mode_0 = const()[name = string("input_425_mode_0"), val = string("replicate")];
            fp16 const_59_to_fp16 = const()[name = string("const_59_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_425_cast_fp16 = pad(constant_val = const_59_to_fp16, mode = input_425_mode_0, pad = input_425_pad_0, x = input_423_cast_fp16)[name = string("input_425_cast_fp16")];
            string h_165_pad_type_0 = const()[name = string("h_165_pad_type_0"), val = string("valid")];
            int32 h_165_groups_0 = const()[name = string("h_165_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_165_strides_0 = const()[name = string("h_165_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_165_pad_0 = const()[name = string("h_165_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_165_dilations_0 = const()[name = string("h_165_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> last_convnext_3_dwconv__conv_weight_to_fp16 = const()[name = string("last_convnext_3_dwconv__conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123292736)))];
            tensor<fp16, [512]> last_convnext_3_dwconv__conv_bias_to_fp16 = const()[name = string("last_convnext_3_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123297920)))];
            tensor<fp16, [2, 512, ?]> h_165_cast_fp16 = conv(bias = last_convnext_3_dwconv__conv_bias_to_fp16, dilations = h_165_dilations_0, groups = h_165_groups_0, pad = h_165_pad_0, pad_type = h_165_pad_type_0, strides = h_165_strides_0, weight = last_convnext_3_dwconv__conv_weight_to_fp16, x = input_425_cast_fp16)[name = string("h_165_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_cast_fp16 = mul(x = h_165_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_cast_fp16")];
            tensor<int32, [3]> input_427_perm_0 = const()[name = string("input_427_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_2557_axes_0 = const()[name = string("op_2557_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> last_convnext_3_norm_norm_weight_to_fp16 = const()[name = string("last_convnext_3_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123299008)))];
            tensor<fp16, [512]> last_convnext_3_norm_norm_bias_to_fp16 = const()[name = string("last_convnext_3_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123300096)))];
            fp16 var_2511_to_fp16 = const()[name = string("op_2511_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [2, ?, 512]> input_427_cast_fp16 = transpose(perm = input_427_perm_0, x = x_cast_fp16)[name = string("transpose_1")];
            tensor<fp16, [2, ?, 512]> var_2557_cast_fp16 = layer_norm(axes = var_2557_axes_0, beta = last_convnext_3_norm_norm_bias_to_fp16, epsilon = var_2511_to_fp16, gamma = last_convnext_3_norm_norm_weight_to_fp16, x = input_427_cast_fp16)[name = string("op_2557_cast_fp16")];
            tensor<int32, [3]> input_429_perm_0 = const()[name = string("input_429_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_167_pad_type_0 = const()[name = string("h_167_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_167_strides_0 = const()[name = string("h_167_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_167_pad_0 = const()[name = string("h_167_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_167_dilations_0 = const()[name = string("h_167_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_167_groups_0 = const()[name = string("h_167_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> last_convnext_3_pwconv1_weight_to_fp16 = const()[name = string("last_convnext_3_pwconv1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123301184)))];
            tensor<fp16, [2048]> last_convnext_3_pwconv1_bias_to_fp16 = const()[name = string("last_convnext_3_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125398400)))];
            tensor<fp16, [2, 512, ?]> input_429_cast_fp16 = transpose(perm = input_429_perm_0, x = var_2557_cast_fp16)[name = string("transpose_0")];
            tensor<fp16, [2, 2048, ?]> h_167_cast_fp16 = conv(bias = last_convnext_3_pwconv1_bias_to_fp16, dilations = h_167_dilations_0, groups = h_167_groups_0, pad = h_167_pad_0, pad_type = h_167_pad_type_0, strides = h_167_strides_0, weight = last_convnext_3_pwconv1_weight_to_fp16, x = input_429_cast_fp16)[name = string("h_167_cast_fp16")];
            string input_431_mode_0 = const()[name = string("input_431_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_431_cast_fp16 = gelu(mode = input_431_mode_0, x = h_167_cast_fp16)[name = string("input_431_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, [512, 2048, 1]> var_2574_weight_0_to_fp16 = const()[name = string("op_2574_weight_0_to_fp16"), val = tensor<fp16, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125402560)))];
            tensor<fp16, [512]> var_2574_bias_0_to_fp16 = const()[name = string("op_2574_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127499776)))];
            tensor<fp16, [2, 512, ?]> var_2574_cast_fp16 = conv(bias = var_2574_bias_0_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 = var_2574_weight_0_to_fp16, x = input_431_cast_fp16)[name = string("op_2574_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_cast_fp16 = add(x = x_223_cast_fp16, y = var_2574_cast_fp16)[name = string("out_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_cast_fp16 = mul(x = out_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_cast_fp16")];
            string var_2587_pad_type_0 = const()[name = string("op_2587_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_2587_strides_0 = const()[name = string("op_2587_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_2587_pad_0 = const()[name = string("op_2587_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_2587_dilations_0 = const()[name = string("op_2587_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_2587_groups_0 = const()[name = string("op_2587_groups_0"), val = int32(1)];
            tensor<fp16, [144, 512, 1]> proj_out_weight_to_fp16 = const()[name = string("proj_out_weight_to_fp16"), val = tensor<fp16, [144, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127500864)))];
            tensor<fp16, [2, 144, ?]> var_2587_cast_fp16 = conv(dilations = var_2587_dilations_0, groups = var_2587_groups_0, pad = var_2587_pad_0, pad_type = var_2587_pad_type_0, strides = var_2587_strides_0, weight = proj_out_weight_to_fp16, x = input_cast_fp16)[name = string("op_2587_cast_fp16")];
            tensor<fp16, [2, 144, ?]> v_cast_fp16 = mul(x = var_2587_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("v_cast_fp16")];
            tensor<int32, [3]> cond_begin_0 = const()[name = string("cond_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> cond_end_0 = const()[name = string("cond_end_0"), val = tensor<int32, [3]>([1, 144, 0])];
            tensor<bool, [3]> cond_end_mask_0 = const()[name = string("cond_end_mask_0"), val = tensor<bool, [3]>([false, true, true])];
            tensor<fp16, [1, 144, ?]> cond_cast_fp16 = slice_by_index(begin = cond_begin_0, end = cond_end_0, end_mask = cond_end_mask_0, x = v_cast_fp16)[name = string("cond_cast_fp16")];
            tensor<int32, [3]> uncond_begin_0 = const()[name = string("uncond_begin_0"), val = tensor<int32, [3]>([1, 0, 0])];
            tensor<int32, [3]> uncond_end_0 = const()[name = string("uncond_end_0"), val = tensor<int32, [3]>([2, 144, 0])];
            tensor<bool, [3]> uncond_end_mask_0 = const()[name = string("uncond_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 144, ?]> uncond_cast_fp16 = slice_by_index(begin = uncond_begin_0, end = uncond_end_0, end_mask = uncond_end_mask_0, x = v_cast_fp16)[name = string("uncond_cast_fp16")];
            fp32 var_2597_epsilon_0 = const()[name = string("op_2597_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1]> var_2597_cast_fp16 = inverse(epsilon = var_2597_epsilon_0, x = total_step_to_fp16)[name = string("op_2597_cast_fp16")];
            tensor<int32, [3]> var_2603 = const()[name = string("op_2603"), val = tensor<int32, [3]>([-1, 1, 1])];
            tensor<fp16, [1, 1, 1]> step_cast_fp16 = reshape(shape = var_2603, x = var_2597_cast_fp16)[name = string("step_cast_fp16")];
            fp16 var_2605_to_fp16 = const()[name = string("op_2605_to_fp16"), val = fp16(0x1p+2)];
            tensor<fp16, [1, 144, ?]> var_2606_cast_fp16 = mul(x = cond_cast_fp16, y = var_2605_to_fp16)[name = string("op_2606_cast_fp16")];
            fp16 var_2607_to_fp16 = const()[name = string("op_2607_to_fp16"), val = fp16(0x1.8p+1)];
            tensor<fp16, [1, 144, ?]> var_2608_cast_fp16 = mul(x = uncond_cast_fp16, y = var_2607_to_fp16)[name = string("op_2608_cast_fp16")];
            tensor<fp16, [1, 144, ?]> var_2610_cast_fp16 = sub(x = var_2606_cast_fp16, y = var_2608_cast_fp16)[name = string("op_2610_cast_fp16")];
            tensor<fp16, [1, 144, ?]> var_2611_cast_fp16 = mul(x = step_cast_fp16, y = var_2610_cast_fp16)[name = string("op_2611_cast_fp16")];
            tensor<fp16, [1, 144, ?]> var_2613_cast_fp16 = add(x = noisy_latent_to_fp16, y = var_2611_cast_fp16)[name = string("op_2613_cast_fp16")];
            tensor<fp16, [1, 144, ?]> var_2614_cast_fp16 = mul(x = var_2613_cast_fp16, y = latent_mask_to_fp16)[name = string("op_2614_cast_fp16")];
            string var_2614_cast_fp16_to_fp32_dtype_0 = const()[name = string("op_2614_cast_fp16_to_fp32_dtype_0"), val = string("fp32")];
            tensor<fp32, [1, 144, ?]> denoised_latent = cast(dtype = var_2614_cast_fp16_to_fp32_dtype_0, x = var_2614_cast_fp16)[name = string("cast_85")];
        } -> (denoised_latent);
}