program(1.3)
[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3400.43.1"}, {"coremlc-version", "3400.58.3.14.1"}}), mldb_token = string("mldb-lvy3n6qgdo")]
{
    func main<ios18>(tensor<fp32, [1, 4, 1024]> audio, state<tensor<fp16, [1, 4, 1056]>> cast_1_state, tensor<fp32, [1, 1]> embedding, state<tensor<fp16, [1, 1536, 33]>> input3_1_cast_fp16_state, state<tensor<fp16, [1, 448, 48]>> input_107_cast_fp16_state, state<tensor<fp16, [1, 448, 64]>> input_115_cast_fp16_state, state<tensor<fp16, [1, 448, 34]>> input_11_cast_fp16_state, state<tensor<fp16, [1, 448, 96]>> input_123_cast_fp16_state, state<tensor<fp16, [1, 448, 160]>> input_131_cast_fp16_state, state<tensor<fp16, [1, 448, 288]>> input_139_cast_fp16_state, state<tensor<fp16, [1, 448, 544]>> input_147_cast_fp16_state, state<tensor<fp16, [1, 448, 34]>> input_155_cast_fp16_state, state<tensor<fp16, [1, 448, 36]>> input_163_cast_fp16_state, state<tensor<fp16, [1, 448, 40]>> input_171_cast_fp16_state, state<tensor<fp16, [1, 448, 48]>> input_179_cast_fp16_state, state<tensor<fp16, [1, 448, 64]>> input_187_cast_fp16_state, state<tensor<fp16, [1, 448, 96]>> input_195_cast_fp16_state, state<tensor<fp16, [1, 448, 36]>> input_19_cast_fp16_state, state<tensor<fp16, [1, 448, 160]>> input_203_cast_fp16_state, state<tensor<fp16, [1, 448, 288]>> input_211_cast_fp16_state, state<tensor<fp16, [1, 448, 544]>> input_219_cast_fp16_state, state<tensor<fp16, [1, 448, 34]>> input_227_cast_fp16_state, state<tensor<fp16, [1, 448, 36]>> input_235_cast_fp16_state, state<tensor<fp16, [1, 448, 40]>> input_243_cast_fp16_state, state<tensor<fp16, [1, 448, 48]>> input_251_cast_fp16_state, state<tensor<fp16, [1, 448, 64]>> input_259_cast_fp16_state, state<tensor<fp16, [1, 448, 96]>> input_267_cast_fp16_state, state<tensor<fp16, [1, 448, 160]>> input_275_cast_fp16_state, state<tensor<fp16, [1, 448, 40]>> input_27_cast_fp16_state, state<tensor<fp16, [1, 448, 288]>> input_283_cast_fp16_state, state<tensor<fp16, [1, 448, 48]>> input_35_cast_fp16_state, state<tensor<fp16, [1, 448, 64]>> input_43_cast_fp16_state, state<tensor<fp16, [1, 448, 544]>> input_4_cast_fp16_state, state<tensor<fp16, [1, 448, 96]>> input_51_cast_fp16_state, state<tensor<fp16, [1, 448, 160]>> input_59_cast_fp16_state, state<tensor<fp16, [1, 448, 288]>> input_67_cast_fp16_state, state<tensor<fp16, [1, 448, 544]>> input_75_cast_fp16_state, state<tensor<fp16, [1, 448, 34]>> input_83_cast_fp16_state, state<tensor<fp16, [1, 448, 36]>> input_91_cast_fp16_state, state<tensor<fp16, [1, 448, 40]>> input_99_cast_fp16_state, state<tensor<fp16, [1, 1, 384, 287]>> var_2668_cast_fp16_state, state<tensor<fp16, [1, 448, 36]>> x_11_cast_fp16_state, state<tensor<fp16, [1, 448, 40]>> x_13_cast_fp16_state, state<tensor<fp16, [1, 448, 48]>> x_15_cast_fp16_state, state<tensor<fp16, [1, 448, 64]>> x_17_cast_fp16_state, state<tensor<fp16, [1, 448, 96]>> x_19_cast_fp16_state, state<tensor<fp16, [1, 448, 160]>> x_21_cast_fp16_state, state<tensor<fp16, [1, 448, 33]>> x_7_cast_fp16_state, state<tensor<fp16, [1, 448, 34]>> x_9_cast_fp16_state) [BNNSOptions = dict<string, string>({{"StateMode", "Streaming"}}), UserMetadata = dict<string, string>({{"iteration", "272205"}, {"taskid", "avvg7m7u75"}})] {
            tensor<int32, [1]> t_1_axes_1 = const()[name = string("t_1_axes_1"), val = tensor<int32, [1]>([1])];
            string embedding_to_fp16_dtype_0 = const()[name = string("embedding_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 1]> cast_7 = cast(dtype = embedding_to_fp16_dtype_0, x = embedding)[name = string("cast_7")];
            tensor<fp16, [1]> t_1_cast_fp16 = squeeze(axes = t_1_axes_1, x = cast_7)[name = string("t_1_cast_fp16")];
            tensor<fp16, [1]> var_102_cast_fp16 = floor(x = t_1_cast_fp16)[name = string("op_102_cast_fp16")];
            string low_idx_1_dtype_1 = const()[name = string("low_idx_1_dtype_1"), val = string("int32")];
            tensor<fp16, [1]> var_104_cast_fp16 = ceil(x = t_1_cast_fp16)[name = string("op_104_cast_fp16")];
            string high_idx_1_dtype_1 = const()[name = string("high_idx_1_dtype_1"), val = string("int32")];
            int32 greater_equal_2_y_0 = const()[name = string("greater_equal_2_y_0"), val = int32(0)];
            tensor<int32, [1]> cast_6 = cast(dtype = low_idx_1_dtype_1, x = var_102_cast_fp16)[name = string("cast_6")];
            tensor<bool, [1]> greater_equal_2 = greater_equal(x = cast_6, y = greater_equal_2_y_0)[name = string("greater_equal_2")];
            int32 slice_by_index_2 = const()[name = string("slice_by_index_2"), val = int32(6)];
            tensor<int32, [1]> add_2 = add(x = cast_6, y = slice_by_index_2)[name = string("add_2")];
            tensor<int32, [1]> select_2 = select(a = cast_6, b = add_2, cond = greater_equal_2)[name = string("select_2")];
            int32 low_1_axis_1 = const()[name = string("low_1_axis_1"), val = int32(0)];
            int32 low_1_batch_dims_1 = const()[name = string("low_1_batch_dims_1"), val = int32(0)];
            bool low_1_validate_indices_1 = const()[name = string("low_1_validate_indices_1"), val = bool(false)];
            tensor<fp16, [6, 448]> t_embedding_positional_embedding_to_fp16 = const()[name = string("t_embedding_positional_embedding_to_fp16"), val = tensor<fp16, [6, 448]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(64)))];
            string select_2_to_int16_dtype_0 = const()[name = string("select_2_to_int16_dtype_0"), val = string("int16")];
            tensor<int16, [1]> cast_5 = cast(dtype = select_2_to_int16_dtype_0, x = select_2)[name = string("cast_5")];
            tensor<fp16, [1, 448]> low_1_cast_fp16_cast_uint16 = gather(axis = low_1_axis_1, batch_dims = low_1_batch_dims_1, indices = cast_5, validate_indices = low_1_validate_indices_1, x = t_embedding_positional_embedding_to_fp16)[name = string("low_1_cast_fp16_cast_uint16")];
            int32 greater_equal_3_y_0 = const()[name = string("greater_equal_3_y_0"), val = int32(0)];
            tensor<int32, [1]> cast_4 = cast(dtype = high_idx_1_dtype_1, x = var_104_cast_fp16)[name = string("cast_4")];
            tensor<bool, [1]> greater_equal_3 = greater_equal(x = cast_4, y = greater_equal_3_y_0)[name = string("greater_equal_3")];
            int32 slice_by_index_3 = const()[name = string("slice_by_index_3"), val = int32(6)];
            tensor<int32, [1]> add_3 = add(x = cast_4, y = slice_by_index_3)[name = string("add_3")];
            tensor<int32, [1]> select_3 = select(a = cast_4, b = add_3, cond = greater_equal_3)[name = string("select_3")];
            int32 high_1_axis_1 = const()[name = string("high_1_axis_1"), val = int32(0)];
            int32 high_1_batch_dims_1 = const()[name = string("high_1_batch_dims_1"), val = int32(0)];
            bool high_1_validate_indices_1 = const()[name = string("high_1_validate_indices_1"), val = bool(false)];
            string select_3_to_uint16_dtype_0 = const()[name = string("select_3_to_uint16_dtype_0"), val = string("uint16")];
            tensor<uint16, [1]> cast_3 = cast(dtype = select_3_to_uint16_dtype_0, x = select_3)[name = string("cast_3")];
            tensor<fp16, [1, 448]> high_1_cast_fp16_cast_uint16 = gather(axis = high_1_axis_1, batch_dims = high_1_batch_dims_1, indices = cast_3, validate_indices = high_1_validate_indices_1, x = t_embedding_positional_embedding_to_fp16)[name = string("high_1_cast_fp16_cast_uint16")];
            tensor<fp16, [1, 448]> var_110_cast_fp16 = sub(x = high_1_cast_fp16_cast_uint16, y = low_1_cast_fp16_cast_uint16)[name = string("op_110_cast_fp16")];
            tensor<int32, [1]> var_112_axes_1 = const()[name = string("op_112_axes_1"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1]> var_112_cast_fp16 = expand_dims(axes = var_112_axes_1, x = t_1_cast_fp16)[name = string("op_112_cast_fp16")];
            tensor<int32, [1]> var_114_axes_1 = const()[name = string("op_114_axes_1"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1, 1]> var_114 = expand_dims(axes = var_114_axes_1, x = cast_6)[name = string("op_114")];
            string var_114_promoted_to_fp16_dtype_0 = const()[name = string("op_114_promoted_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 1]> cast_2 = cast(dtype = var_114_promoted_to_fp16_dtype_0, x = var_114)[name = string("cast_2")];
            tensor<fp16, [1, 1]> var_115_cast_fp16 = sub(x = var_112_cast_fp16, y = cast_2)[name = string("op_115_cast_fp16")];
            tensor<fp16, [1, 448]> var_116_cast_fp16 = mul(x = var_110_cast_fp16, y = var_115_cast_fp16)[name = string("op_116_cast_fp16")];
            tensor<fp16, [1, 448]> input_5_cast_fp16 = add(x = low_1_cast_fp16_cast_uint16, y = var_116_cast_fp16)[name = string("input_5_cast_fp16")];
            tensor<fp16, [384, 448]> t_embedding_fc1_weight_to_fp16 = const()[name = string("t_embedding_fc1_weight_to_fp16"), val = tensor<fp16, [384, 448]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5504)))];
            tensor<fp16, [384]> t_embedding_fc1_bias_to_fp16 = const()[name = string("t_embedding_fc1_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(349632)))];
            tensor<fp16, [1, 384]> linear_3_cast_fp16 = linear(bias = t_embedding_fc1_bias_to_fp16, weight = t_embedding_fc1_weight_to_fp16, x = input_5_cast_fp16)[name = string("linear_3_cast_fp16")];
            tensor<fp16, [1, 384]> input4_1_cast_fp16 = silu(x = linear_3_cast_fp16)[name = string("input4_1_cast_fp16")];
            tensor<fp16, [384, 384]> t_embedding_fc2_weight_to_fp16 = const()[name = string("t_embedding_fc2_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(350464)))];
            tensor<fp16, [384]> t_embedding_fc2_bias_to_fp16 = const()[name = string("t_embedding_fc2_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(645440)))];
            tensor<fp16, [1, 384]> linear_4_cast_fp16 = linear(bias = t_embedding_fc2_bias_to_fp16, weight = t_embedding_fc2_weight_to_fp16, x = input4_1_cast_fp16)[name = string("linear_4_cast_fp16")];
            tensor<fp16, [1, 384]> var_125_cast_fp16 = silu(x = linear_4_cast_fp16)[name = string("op_125_cast_fp16")];
            tensor<fp16, [384, 384]> diffusion_embedding_input_weight_to_fp16 = const()[name = string("diffusion_embedding_input_weight_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(646272)))];
            tensor<fp16, [384]> diffusion_embedding_input_bias_to_fp16 = const()[name = string("diffusion_embedding_input_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(941248)))];
            tensor<fp16, [1, 384]> linear_5_cast_fp16 = linear(bias = diffusion_embedding_input_bias_to_fp16, weight = diffusion_embedding_input_weight_to_fp16, x = var_125_cast_fp16)[name = string("linear_5_cast_fp16")];
            tensor<int32, [1]> x1_1_axes_1 = const()[name = string("x1_1_axes_1"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 384, 1]> x1_1_cast_fp16 = expand_dims(axes = x1_1_axes_1, x = linear_5_cast_fp16)[name = string("x1_1_cast_fp16")];
            int32 var_135 = const()[name = string("op_135"), val = int32(1)];
            tensor<int32, [1]> var_139 = const()[name = string("op_139"), val = tensor<int32, [1]>([32])];
            tensor<int32, [1]> var_141 = const()[name = string("op_141"), val = tensor<int32, [1]>([1])];
            string input0_7_pad_type_1 = const()[name = string("input0_7_pad_type_1"), val = string("custom")];
            tensor<int32, [2]> input0_7_pad_1 = const()[name = string("input0_7_pad_1"), val = tensor<int32, [2]>([32, 32])];
            string audio_to_fp16_dtype_0 = const()[name = string("audio_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [384, 4, 64]> front_end_0_weight_to_fp16 = const()[name = string("front_end_0_weight_to_fp16"), val = tensor<fp16, [384, 4, 64]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(942080)))];
            tensor<fp16, [1, 4, 1024]> cast_1 = cast(dtype = audio_to_fp16_dtype_0, x = audio)[name = string("cast_1")];
            tensor<fp16, [1, 4, 1056]> cast_1_state_input = read_state(input = cast_1_state);
            tensor<fp16, [1, 4, 1056]> cast_1_state_updated = slice_update(begin = tensor<int32, [3]>([0, 0, 32]), end = tensor<int32, [3]>([1, 4, 1056]), end_mask = tensor<bool, [3]>([false, false, false]), update = cast_1, x = cast_1_state_input);
            tensor<fp16, [1, 384, 32]> input0_7_cast_fp16 = conv(dilations = var_141, groups = var_135, pad = tensor<int32, [2]>([0, 0]), pad_type = input0_7_pad_type_1, strides = var_139, weight = front_end_0_weight_to_fp16, x = cast_1_state_updated);
            write_state(data = cast_1_state_updated, input = cast_1_state);
            tensor<fp16, [1, 384, 32]> var_144_cast_fp16 = relu(x = input0_7_cast_fp16)[name = string("op_144_cast_fp16")];
            tensor<fp16, [1, 384, 32]> x_5_cast_fp16 = add(x = var_144_cast_fp16, y = x1_1_cast_fp16)[name = string("x_5_cast_fp16")];
            bool var_148 = const()[name = string("op_148"), val = bool(true)];
            tensor<int32, [1]> var_153 = const()[name = string("op_153"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_3_cast_fp16 = reduce_mean(axes = var_153, keep_dims = var_148, x = x_5_cast_fp16)[name = string("mean_y_3_cast_fp16")];
            tensor<fp16, [1, 384, 32]> var_155_cast_fp16 = sub(x = x_5_cast_fp16, y = mean_y_3_cast_fp16)[name = string("op_155_cast_fp16")];
            tensor<fp16, [1, 384, 32]> var_156_cast_fp16 = square(x = var_155_cast_fp16);
            tensor<int32, [1]> var_157 = const()[name = string("op_157"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_158_cast_fp16 = reduce_mean(axes = var_157, keep_dims = var_148, x = var_156_cast_fp16)[name = string("op_158_cast_fp16")];
            fp16 var_159_to_fp16 = const()[name = string("op_159_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_160_cast_fp16 = add(x = var_158_cast_fp16, y = var_159_to_fp16)[name = string("op_160_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_3_cast_fp16 = sqrt(x = var_160_cast_fp16)[name = string("std_y_3_cast_fp16")];
            tensor<fp16, [1, 384, 1]> front_norm_norm_gamma_to_fp16 = const()[name = string("front_norm_norm_gamma_to_fp16"), val = tensor<fp16, [1, 384, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(1138752)))];
            tensor<fp16, [1, 384, 32]> var_163_cast_fp16 = mul(x = front_norm_norm_gamma_to_fp16, y = var_155_cast_fp16)[name = string("op_163_cast_fp16")];
            tensor<fp16, [1, 384, 32]> var_164_cast_fp16 = real_div(x = var_163_cast_fp16, y = std_y_3_cast_fp16)[name = string("op_164_cast_fp16")];
            tensor<fp16, [1, 384, 1]> front_norm_norm_beta_to_fp16 = const()[name = string("front_norm_norm_beta_to_fp16"), val = tensor<fp16, [1, 384, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(1139584)))];
            tensor<fp16, [1, 384, 32]> input_7_cast_fp16 = add(x = var_164_cast_fp16, y = front_norm_norm_beta_to_fp16)[name = string("input_7_cast_fp16")];
            int32 var_167 = const()[name = string("op_167"), val = int32(1)];
            tensor<int32, [1]> var_172 = const()[name = string("op_172"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_174 = const()[name = string("op_174"), val = tensor<int32, [1]>([1])];
            string x_7_pad_type_0 = const()[name = string("x_7_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> x_7_pad_0 = const()[name = string("x_7_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 384, 1]> to_latent_weight_to_fp16 = const()[name = string("to_latent_weight_to_fp16"), val = tensor<fp16, [448, 384, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(1140416)))];
            tensor<fp16, [1, 448, 32]> x_7_cast_fp16 = conv(dilations = var_174, groups = var_167, pad = x_7_pad_0, pad_type = x_7_pad_type_0, strides = var_172, weight = to_latent_weight_to_fp16, x = input_7_cast_fp16)[name = string("x_7_cast_fp16")];
            tensor<fp16, [448, 384]> diffusion_embedding_tcn_weight_to_fp16 = const()[name = string("diffusion_embedding_tcn_weight_to_fp16"), val = tensor<fp16, [448, 384]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(1484544)))];
            tensor<fp16, [448]> diffusion_embedding_tcn_bias_to_fp16 = const()[name = string("diffusion_embedding_tcn_bias_to_fp16"), val = tensor<fp16, [448]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(1828672)))];
            tensor<fp16, [1, 448]> linear_6_cast_fp16 = linear(bias = diffusion_embedding_tcn_bias_to_fp16, weight = diffusion_embedding_tcn_weight_to_fp16, x = var_125_cast_fp16)[name = string("linear_6_cast_fp16")];
            tensor<int32, [1]> diffusion_embedding_1_axes_0 = const()[name = string("diffusion_embedding_1_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 448, 1]> diffusion_embedding_1_cast_fp16 = expand_dims(axes = diffusion_embedding_1_axes_0, x = linear_6_cast_fp16)[name = string("diffusion_embedding_1_cast_fp16")];
            bool var_185 = const()[name = string("op_185"), val = bool(true)];
            int32 var_188 = const()[name = string("op_188"), val = int32(448)];
            int32 var_189 = const()[name = string("op_189"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_9_cast_fp16 = add(x = x_7_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_9_cast_fp16")];
            tensor<int32, [1]> var_199 = const()[name = string("op_199"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_201 = const()[name = string("op_201"), val = tensor<int32, [1]>([1])];
            string input0_9_pad_type_0 = const()[name = string("input0_9_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_9_pad_0 = const()[name = string("input0_9_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_0_tcn_0_weight_to_fp16 = const()[name = string("sep_module_0_tcn_0_weight_to_fp16"), val = tensor<fp16, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(1829632)))];
            tensor<fp16, [1, 448, 32]> input0_9_cast_fp16 = conv(dilations = var_201, groups = var_189, pad = input0_9_pad_0, pad_type = input0_9_pad_type_0, strides = var_199, weight = sep_module_0_tcn_0_weight_to_fp16, x = input_9_cast_fp16)[name = string("input0_9_cast_fp16")];
            fp32 var_205_alpha_1 = const()[name = string("op_205_alpha_1"), val = fp32(0x1.1de9e8p-2)];
            tensor<fp16, [1, 448, 32]> var_205_cast_fp16 = leaky_relu(alpha = fp16(0x1.1ep-2), x = input0_9_cast_fp16);
            tensor<int32, [1]> var_209 = const()[name = string("op_209"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_5_cast_fp16 = reduce_mean(axes = var_209, keep_dims = var_185, x = var_205_cast_fp16)[name = string("mean_y_5_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_211_cast_fp16 = sub(x = var_205_cast_fp16, y = mean_y_5_cast_fp16)[name = string("op_211_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_212_cast_fp16 = square(x = var_211_cast_fp16);
            tensor<int32, [1]> var_213 = const()[name = string("op_213"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_214_cast_fp16 = reduce_mean(axes = var_213, keep_dims = var_185, x = var_212_cast_fp16)[name = string("op_214_cast_fp16")];
            fp16 var_215_to_fp16 = const()[name = string("op_215_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_216_cast_fp16 = add(x = var_214_cast_fp16, y = var_215_to_fp16)[name = string("op_216_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_5_cast_fp16 = sqrt(x = var_216_cast_fp16)[name = string("std_y_5_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_0_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_0_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(2231104)))];
            tensor<fp16, [1, 448, 32]> var_219_cast_fp16 = mul(x = sep_module_0_tcn_2_norm_gamma_to_fp16, y = var_211_cast_fp16)[name = string("op_219_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_220_cast_fp16 = real_div(x = var_219_cast_fp16, y = std_y_5_cast_fp16)[name = string("op_220_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_0_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_0_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(2232064)))];
            tensor<fp16, [1, 448, 32]> input_11_cast_fp16 = add(x = var_220_cast_fp16, y = sep_module_0_tcn_2_norm_beta_to_fp16)[name = string("input_11_cast_fp16")];
            tensor<int32, [6]> input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 1])];
            string input_13_mode_0 = const()[name = string("input_13_mode_0"), val = string("constant")];
            fp16 const_0_to_fp16 = const()[name = string("const_0_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 34]> input_11_cast_fp16_state_input = read_state(input = input_11_cast_fp16_state);
            tensor<fp16, [1, 448, 34]> input_13_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 2]), end = tensor<int32, [3]>([1, 448, 34]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_11_cast_fp16, x = input_11_cast_fp16_state_input);
            write_state(data = input_13_cast_fp16, input = input_11_cast_fp16_state);
            tensor<int32, [1]> var_225 = const()[name = string("op_225"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_227 = const()[name = string("op_227"), val = tensor<int32, [1]>([1])];
            string input_15_pad_type_0 = const()[name = string("input_15_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_15_pad_0 = const()[name = string("input_15_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_0_tcn_4_weight_to_fp16 = const()[name = string("sep_module_0_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(2233024)))];
            tensor<fp16, [1, 448, 32]> input_15_cast_fp16 = conv(dilations = var_227, groups = var_188, pad = input_15_pad_0, pad_type = input_15_pad_type_0, strides = var_225, weight = sep_module_0_tcn_4_weight_to_fp16, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")];
            fp32 var_231_alpha_1 = const()[name = string("op_231_alpha_1"), val = fp32(-0x1.2d0644p-2)];
            tensor<fp16, [1, 448, 32]> var_231_cast_fp16 = leaky_relu(alpha = fp16(-0x1.2dp-2), x = input_15_cast_fp16);
            tensor<int32, [1]> var_235 = const()[name = string("op_235"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_7_cast_fp16 = reduce_mean(axes = var_235, keep_dims = var_185, x = var_231_cast_fp16)[name = string("mean_y_7_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_237_cast_fp16 = sub(x = var_231_cast_fp16, y = mean_y_7_cast_fp16)[name = string("op_237_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_238_cast_fp16 = square(x = var_237_cast_fp16);
            tensor<int32, [1]> var_239 = const()[name = string("op_239"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_240_cast_fp16 = reduce_mean(axes = var_239, keep_dims = var_185, x = var_238_cast_fp16)[name = string("op_240_cast_fp16")];
            fp16 var_241_to_fp16 = const()[name = string("op_241_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_242_cast_fp16 = add(x = var_240_cast_fp16, y = var_241_to_fp16)[name = string("op_242_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_7_cast_fp16 = sqrt(x = var_242_cast_fp16)[name = string("std_y_7_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_0_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_0_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(2235776)))];
            tensor<fp16, [1, 448, 32]> var_245_cast_fp16 = mul(x = sep_module_0_tcn_6_norm_gamma_to_fp16, y = var_237_cast_fp16)[name = string("op_245_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_246_cast_fp16 = real_div(x = var_245_cast_fp16, y = std_y_7_cast_fp16)[name = string("op_246_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_0_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_0_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(2236736)))];
            tensor<fp16, [1, 448, 32]> y_2_cast_fp16 = add(x = var_246_cast_fp16, y = sep_module_0_tcn_6_norm_beta_to_fp16)[name = string("y_2_cast_fp16")];
            tensor<fp16, [1, 448, 33]> x_7_cast_fp16_state_input = read_state(input = x_7_cast_fp16_state);
            tensor<fp16, [1, 448, 33]> x_7_cast_fp16_state_updated = slice_update(begin = tensor<int32, [3]>([0, 0, 1]), end = tensor<int32, [3]>([1, 448, 33]), end_mask = tensor<bool, [3]>([false, false, false]), update = x_7_cast_fp16, x = x_7_cast_fp16_state_input);
            write_state(data = x_7_cast_fp16_state_updated, input = x_7_cast_fp16_state);
            tensor<fp16, [1, 448, 32]> x_7_cast_fp16_delayed = slice_by_size(begin = tensor<int32, [3]>([0, 0, 0]), size = tensor<int32, [3]>([1, 448, 32]), x = x_7_cast_fp16_state_updated);
            tensor<fp16, [1, 448, 32]> x_9_cast_fp16 = add(x = x_7_cast_fp16_delayed, y = y_2_cast_fp16)[name = string("x_9_cast_fp16")];
            bool var_252 = const()[name = string("op_252"), val = bool(true)];
            int32 var_256 = const()[name = string("op_256"), val = int32(448)];
            int32 var_257 = const()[name = string("op_257"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_17_cast_fp16 = add(x = x_9_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_17_cast_fp16")];
            tensor<int32, [1]> var_267 = const()[name = string("op_267"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_269 = const()[name = string("op_269"), val = tensor<int32, [1]>([1])];
            string input0_11_pad_type_0 = const()[name = string("input0_11_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_11_pad_0 = const()[name = string("input0_11_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_1_tcn_0_weight_to_fp16 = const()[name = string("sep_module_1_tcn_0_weight_to_fp16"), val = tensor<fp16, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(2237696)))];
            tensor<fp16, [1, 448, 32]> input0_11_cast_fp16 = conv(dilations = var_269, groups = var_257, pad = input0_11_pad_0, pad_type = input0_11_pad_type_0, strides = var_267, weight = sep_module_1_tcn_0_weight_to_fp16, x = input_17_cast_fp16)[name = string("input0_11_cast_fp16")];
            fp32 var_273_alpha_1 = const()[name = string("op_273_alpha_1"), val = fp32(0x1.b28c8cp-2)];
            tensor<fp16, [1, 448, 32]> var_273_cast_fp16 = leaky_relu(alpha = fp16(0x1.b28p-2), x = input0_11_cast_fp16);
            tensor<int32, [1]> var_277 = const()[name = string("op_277"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_9_cast_fp16 = reduce_mean(axes = var_277, keep_dims = var_252, x = var_273_cast_fp16)[name = string("mean_y_9_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_279_cast_fp16 = sub(x = var_273_cast_fp16, y = mean_y_9_cast_fp16)[name = string("op_279_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_280_cast_fp16 = square(x = var_279_cast_fp16);
            tensor<int32, [1]> var_281 = const()[name = string("op_281"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_282_cast_fp16 = reduce_mean(axes = var_281, keep_dims = var_252, x = var_280_cast_fp16)[name = string("op_282_cast_fp16")];
            fp16 var_283_to_fp16 = const()[name = string("op_283_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_284_cast_fp16 = add(x = var_282_cast_fp16, y = var_283_to_fp16)[name = string("op_284_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_9_cast_fp16 = sqrt(x = var_284_cast_fp16)[name = string("std_y_9_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_1_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_1_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(2639168)))];
            tensor<fp16, [1, 448, 32]> var_287_cast_fp16 = mul(x = sep_module_1_tcn_2_norm_gamma_to_fp16, y = var_279_cast_fp16)[name = string("op_287_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_288_cast_fp16 = real_div(x = var_287_cast_fp16, y = std_y_9_cast_fp16)[name = string("op_288_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_1_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_1_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(2640128)))];
            tensor<fp16, [1, 448, 32]> input_19_cast_fp16 = add(x = var_288_cast_fp16, y = sep_module_1_tcn_2_norm_beta_to_fp16)[name = string("input_19_cast_fp16")];
            tensor<int32, [6]> input_21_pad_0 = const()[name = string("input_21_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_21_mode_0 = const()[name = string("input_21_mode_0"), val = string("constant")];
            fp16 const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 36]> input_19_cast_fp16_state_input = read_state(input = input_19_cast_fp16_state);
            tensor<fp16, [1, 448, 36]> input_21_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 4]), end = tensor<int32, [3]>([1, 448, 36]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_19_cast_fp16, x = input_19_cast_fp16_state_input);
            write_state(data = input_21_cast_fp16, input = input_19_cast_fp16_state);
            tensor<int32, [1]> var_293 = const()[name = string("op_293"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_295 = const()[name = string("op_295"), val = tensor<int32, [1]>([2])];
            string input_23_pad_type_0 = const()[name = string("input_23_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_23_pad_0 = const()[name = string("input_23_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_1_tcn_4_weight_to_fp16 = const()[name = string("sep_module_1_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(2641088)))];
            tensor<fp16, [1, 448, 32]> input_23_cast_fp16 = conv(dilations = var_295, groups = var_256, pad = input_23_pad_0, pad_type = input_23_pad_type_0, strides = var_293, weight = sep_module_1_tcn_4_weight_to_fp16, x = input_21_cast_fp16)[name = string("input_23_cast_fp16")];
            fp32 var_299_alpha_1 = const()[name = string("op_299_alpha_1"), val = fp32(-0x1.3f336ap-2)];
            tensor<fp16, [1, 448, 32]> var_299_cast_fp16 = leaky_relu(alpha = fp16(-0x1.3f4p-2), x = input_23_cast_fp16);
            tensor<int32, [1]> var_303 = const()[name = string("op_303"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_11_cast_fp16 = reduce_mean(axes = var_303, keep_dims = var_252, x = var_299_cast_fp16)[name = string("mean_y_11_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_305_cast_fp16 = sub(x = var_299_cast_fp16, y = mean_y_11_cast_fp16)[name = string("op_305_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_306_cast_fp16 = square(x = var_305_cast_fp16);
            tensor<int32, [1]> var_307 = const()[name = string("op_307"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_308_cast_fp16 = reduce_mean(axes = var_307, keep_dims = var_252, x = var_306_cast_fp16)[name = string("op_308_cast_fp16")];
            fp16 var_309_to_fp16 = const()[name = string("op_309_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_310_cast_fp16 = add(x = var_308_cast_fp16, y = var_309_to_fp16)[name = string("op_310_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_11_cast_fp16 = sqrt(x = var_310_cast_fp16)[name = string("std_y_11_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_1_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_1_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(2643840)))];
            tensor<fp16, [1, 448, 32]> var_313_cast_fp16 = mul(x = sep_module_1_tcn_6_norm_gamma_to_fp16, y = var_305_cast_fp16)[name = string("op_313_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_314_cast_fp16 = real_div(x = var_313_cast_fp16, y = std_y_11_cast_fp16)[name = string("op_314_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_1_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_1_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(2644800)))];
            tensor<fp16, [1, 448, 32]> y_4_cast_fp16 = add(x = var_314_cast_fp16, y = sep_module_1_tcn_6_norm_beta_to_fp16)[name = string("y_4_cast_fp16")];
            tensor<fp16, [1, 448, 34]> x_9_cast_fp16_state_input = read_state(input = x_9_cast_fp16_state);
            tensor<fp16, [1, 448, 34]> x_9_cast_fp16_state_updated = slice_update(begin = tensor<int32, [3]>([0, 0, 2]), end = tensor<int32, [3]>([1, 448, 34]), end_mask = tensor<bool, [3]>([false, false, false]), update = x_9_cast_fp16, x = x_9_cast_fp16_state_input);
            write_state(data = x_9_cast_fp16_state_updated, input = x_9_cast_fp16_state);
            tensor<fp16, [1, 448, 32]> x_9_cast_fp16_delayed = slice_by_size(begin = tensor<int32, [3]>([0, 0, 0]), size = tensor<int32, [3]>([1, 448, 32]), x = x_9_cast_fp16_state_updated);
            tensor<fp16, [1, 448, 32]> x_11_cast_fp16 = add(x = x_9_cast_fp16_delayed, y = y_4_cast_fp16)[name = string("x_11_cast_fp16")];
            bool var_320 = const()[name = string("op_320"), val = bool(true)];
            int32 var_324 = const()[name = string("op_324"), val = int32(448)];
            int32 var_325 = const()[name = string("op_325"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_25_cast_fp16 = add(x = x_11_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_25_cast_fp16")];
            tensor<int32, [1]> var_335 = const()[name = string("op_335"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_337 = const()[name = string("op_337"), val = tensor<int32, [1]>([1])];
            string input0_13_pad_type_0 = const()[name = string("input0_13_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_13_pad_0 = const()[name = string("input0_13_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_2_tcn_0_weight_to_fp16 = const()[name = string("sep_module_2_tcn_0_weight_to_fp16"), val = tensor<fp16, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(2645760)))];
            tensor<fp16, [1, 448, 32]> input0_13_cast_fp16 = conv(dilations = var_337, groups = var_325, pad = input0_13_pad_0, pad_type = input0_13_pad_type_0, strides = var_335, weight = sep_module_2_tcn_0_weight_to_fp16, x = input_25_cast_fp16)[name = string("input0_13_cast_fp16")];
            fp32 var_341_alpha_1 = const()[name = string("op_341_alpha_1"), val = fp32(0x1.f598e8p-2)];
            tensor<fp16, [1, 448, 32]> var_341_cast_fp16 = leaky_relu(alpha = fp16(0x1.f58p-2), x = input0_13_cast_fp16);
            tensor<int32, [1]> var_345 = const()[name = string("op_345"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_13_cast_fp16 = reduce_mean(axes = var_345, keep_dims = var_320, x = var_341_cast_fp16)[name = string("mean_y_13_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_347_cast_fp16 = sub(x = var_341_cast_fp16, y = mean_y_13_cast_fp16)[name = string("op_347_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_348_cast_fp16 = square(x = var_347_cast_fp16);
            tensor<int32, [1]> var_349 = const()[name = string("op_349"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_350_cast_fp16 = reduce_mean(axes = var_349, keep_dims = var_320, x = var_348_cast_fp16)[name = string("op_350_cast_fp16")];
            fp16 var_351_to_fp16 = const()[name = string("op_351_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_352_cast_fp16 = add(x = var_350_cast_fp16, y = var_351_to_fp16)[name = string("op_352_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_13_cast_fp16 = sqrt(x = var_352_cast_fp16)[name = string("std_y_13_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_2_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_2_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3047232)))];
            tensor<fp16, [1, 448, 32]> var_355_cast_fp16 = mul(x = sep_module_2_tcn_2_norm_gamma_to_fp16, y = var_347_cast_fp16)[name = string("op_355_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_356_cast_fp16 = real_div(x = var_355_cast_fp16, y = std_y_13_cast_fp16)[name = string("op_356_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_2_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_2_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3048192)))];
            tensor<fp16, [1, 448, 32]> input_27_cast_fp16 = add(x = var_356_cast_fp16, y = sep_module_2_tcn_2_norm_beta_to_fp16)[name = string("input_27_cast_fp16")];
            tensor<int32, [6]> input_29_pad_0 = const()[name = string("input_29_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_29_mode_0 = const()[name = string("input_29_mode_0"), val = string("constant")];
            fp16 const_2_to_fp16 = const()[name = string("const_2_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 40]> input_27_cast_fp16_state_input = read_state(input = input_27_cast_fp16_state);
            tensor<fp16, [1, 448, 40]> input_29_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 8]), end = tensor<int32, [3]>([1, 448, 40]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_27_cast_fp16, x = input_27_cast_fp16_state_input);
            write_state(data = input_29_cast_fp16, input = input_27_cast_fp16_state);
            tensor<int32, [1]> var_361 = const()[name = string("op_361"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_363 = const()[name = string("op_363"), val = tensor<int32, [1]>([4])];
            string input_31_pad_type_0 = const()[name = string("input_31_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_31_pad_0 = const()[name = string("input_31_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_2_tcn_4_weight_to_fp16 = const()[name = string("sep_module_2_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3049152)))];
            tensor<fp16, [1, 448, 32]> input_31_cast_fp16 = conv(dilations = var_363, groups = var_324, pad = input_31_pad_0, pad_type = input_31_pad_type_0, strides = var_361, weight = sep_module_2_tcn_4_weight_to_fp16, x = input_29_cast_fp16)[name = string("input_31_cast_fp16")];
            fp32 var_367_alpha_1 = const()[name = string("op_367_alpha_1"), val = fp32(-0x1.83468p-3)];
            tensor<fp16, [1, 448, 32]> var_367_cast_fp16 = leaky_relu(alpha = fp16(-0x1.834p-3), x = input_31_cast_fp16);
            tensor<int32, [1]> var_371 = const()[name = string("op_371"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_15_cast_fp16 = reduce_mean(axes = var_371, keep_dims = var_320, x = var_367_cast_fp16)[name = string("mean_y_15_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_373_cast_fp16 = sub(x = var_367_cast_fp16, y = mean_y_15_cast_fp16)[name = string("op_373_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_374_cast_fp16 = square(x = var_373_cast_fp16);
            tensor<int32, [1]> var_375 = const()[name = string("op_375"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_376_cast_fp16 = reduce_mean(axes = var_375, keep_dims = var_320, x = var_374_cast_fp16)[name = string("op_376_cast_fp16")];
            fp16 var_377_to_fp16 = const()[name = string("op_377_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_378_cast_fp16 = add(x = var_376_cast_fp16, y = var_377_to_fp16)[name = string("op_378_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_15_cast_fp16 = sqrt(x = var_378_cast_fp16)[name = string("std_y_15_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_2_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_2_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3051904)))];
            tensor<fp16, [1, 448, 32]> var_381_cast_fp16 = mul(x = sep_module_2_tcn_6_norm_gamma_to_fp16, y = var_373_cast_fp16)[name = string("op_381_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_382_cast_fp16 = real_div(x = var_381_cast_fp16, y = std_y_15_cast_fp16)[name = string("op_382_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_2_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_2_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3052864)))];
            tensor<fp16, [1, 448, 32]> y_6_cast_fp16 = add(x = var_382_cast_fp16, y = sep_module_2_tcn_6_norm_beta_to_fp16)[name = string("y_6_cast_fp16")];
            tensor<fp16, [1, 448, 36]> x_11_cast_fp16_state_input = read_state(input = x_11_cast_fp16_state);
            tensor<fp16, [1, 448, 36]> x_11_cast_fp16_state_updated = slice_update(begin = tensor<int32, [3]>([0, 0, 4]), end = tensor<int32, [3]>([1, 448, 36]), end_mask = tensor<bool, [3]>([false, false, false]), update = x_11_cast_fp16, x = x_11_cast_fp16_state_input);
            write_state(data = x_11_cast_fp16_state_updated, input = x_11_cast_fp16_state);
            tensor<fp16, [1, 448, 32]> x_11_cast_fp16_delayed = slice_by_size(begin = tensor<int32, [3]>([0, 0, 0]), size = tensor<int32, [3]>([1, 448, 32]), x = x_11_cast_fp16_state_updated);
            tensor<fp16, [1, 448, 32]> x_13_cast_fp16 = add(x = x_11_cast_fp16_delayed, y = y_6_cast_fp16)[name = string("x_13_cast_fp16")];
            bool var_388 = const()[name = string("op_388"), val = bool(true)];
            int32 var_392 = const()[name = string("op_392"), val = int32(448)];
            int32 var_393 = const()[name = string("op_393"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_33_cast_fp16 = add(x = x_13_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_33_cast_fp16")];
            tensor<int32, [1]> var_403 = const()[name = string("op_403"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_405 = const()[name = string("op_405"), val = tensor<int32, [1]>([1])];
            string input0_15_pad_type_0 = const()[name = string("input0_15_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_15_pad_0 = const()[name = string("input0_15_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_3_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3053824))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3154240))))[name = string("sep_module_3_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_15_cast_fp16 = conv(dilations = var_405, groups = var_393, pad = input0_15_pad_0, pad_type = input0_15_pad_type_0, strides = var_403, weight = sep_module_3_tcn_0_weight_to_fp16_palettized, x = input_33_cast_fp16)[name = string("input0_15_cast_fp16")];
            fp32 var_409_alpha_1 = const()[name = string("op_409_alpha_1"), val = fp32(0x1.e6974ep-2)];
            tensor<fp16, [1, 448, 32]> var_409_cast_fp16 = leaky_relu(alpha = fp16(0x1.e68p-2), x = input0_15_cast_fp16);
            tensor<int32, [1]> var_413 = const()[name = string("op_413"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_17_cast_fp16 = reduce_mean(axes = var_413, keep_dims = var_388, x = var_409_cast_fp16)[name = string("mean_y_17_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_415_cast_fp16 = sub(x = var_409_cast_fp16, y = mean_y_17_cast_fp16)[name = string("op_415_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_416_cast_fp16 = square(x = var_415_cast_fp16);
            tensor<int32, [1]> var_417 = const()[name = string("op_417"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_418_cast_fp16 = reduce_mean(axes = var_417, keep_dims = var_388, x = var_416_cast_fp16)[name = string("op_418_cast_fp16")];
            fp16 var_419_to_fp16 = const()[name = string("op_419_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_420_cast_fp16 = add(x = var_418_cast_fp16, y = var_419_to_fp16)[name = string("op_420_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_17_cast_fp16 = sqrt(x = var_420_cast_fp16)[name = string("std_y_17_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_3_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_3_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3154368)))];
            tensor<fp16, [1, 448, 32]> var_423_cast_fp16 = mul(x = sep_module_3_tcn_2_norm_gamma_to_fp16, y = var_415_cast_fp16)[name = string("op_423_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_424_cast_fp16 = real_div(x = var_423_cast_fp16, y = std_y_17_cast_fp16)[name = string("op_424_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_3_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_3_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3155328)))];
            tensor<fp16, [1, 448, 32]> input_35_cast_fp16 = add(x = var_424_cast_fp16, y = sep_module_3_tcn_2_norm_beta_to_fp16)[name = string("input_35_cast_fp16")];
            tensor<int32, [6]> input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 8, 8])];
            string input_37_mode_0 = const()[name = string("input_37_mode_0"), val = string("constant")];
            fp16 const_3_to_fp16 = const()[name = string("const_3_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 48]> input_35_cast_fp16_state_input = read_state(input = input_35_cast_fp16_state);
            tensor<fp16, [1, 448, 48]> input_37_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 16]), end = tensor<int32, [3]>([1, 448, 48]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_35_cast_fp16, x = input_35_cast_fp16_state_input);
            write_state(data = input_37_cast_fp16, input = input_35_cast_fp16_state);
            tensor<int32, [1]> var_429 = const()[name = string("op_429"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_431 = const()[name = string("op_431"), val = tensor<int32, [1]>([8])];
            string input_39_pad_type_0 = const()[name = string("input_39_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_39_pad_0 = const()[name = string("input_39_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_3_tcn_4_weight_to_fp16 = const()[name = string("sep_module_3_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3156288)))];
            tensor<fp16, [1, 448, 32]> input_39_cast_fp16 = conv(dilations = var_431, groups = var_392, pad = input_39_pad_0, pad_type = input_39_pad_type_0, strides = var_429, weight = sep_module_3_tcn_4_weight_to_fp16, x = input_37_cast_fp16)[name = string("input_39_cast_fp16")];
            fp32 var_435_alpha_1 = const()[name = string("op_435_alpha_1"), val = fp32(0x1.fa3752p-4)];
            tensor<fp16, [1, 448, 32]> var_435_cast_fp16 = leaky_relu(alpha = fp16(0x1.fa4p-4), x = input_39_cast_fp16);
            tensor<int32, [1]> var_439 = const()[name = string("op_439"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_19_cast_fp16 = reduce_mean(axes = var_439, keep_dims = var_388, x = var_435_cast_fp16)[name = string("mean_y_19_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_441_cast_fp16 = sub(x = var_435_cast_fp16, y = mean_y_19_cast_fp16)[name = string("op_441_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_442_cast_fp16 = square(x = var_441_cast_fp16);
            tensor<int32, [1]> var_443 = const()[name = string("op_443"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_444_cast_fp16 = reduce_mean(axes = var_443, keep_dims = var_388, x = var_442_cast_fp16)[name = string("op_444_cast_fp16")];
            fp16 var_445_to_fp16 = const()[name = string("op_445_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_446_cast_fp16 = add(x = var_444_cast_fp16, y = var_445_to_fp16)[name = string("op_446_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_19_cast_fp16 = sqrt(x = var_446_cast_fp16)[name = string("std_y_19_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_3_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_3_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3159040)))];
            tensor<fp16, [1, 448, 32]> var_449_cast_fp16 = mul(x = sep_module_3_tcn_6_norm_gamma_to_fp16, y = var_441_cast_fp16)[name = string("op_449_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_450_cast_fp16 = real_div(x = var_449_cast_fp16, y = std_y_19_cast_fp16)[name = string("op_450_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_3_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_3_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3160000)))];
            tensor<fp16, [1, 448, 32]> y_8_cast_fp16 = add(x = var_450_cast_fp16, y = sep_module_3_tcn_6_norm_beta_to_fp16)[name = string("y_8_cast_fp16")];
            tensor<fp16, [1, 448, 40]> x_13_cast_fp16_state_input = read_state(input = x_13_cast_fp16_state);
            tensor<fp16, [1, 448, 40]> x_13_cast_fp16_state_updated = slice_update(begin = tensor<int32, [3]>([0, 0, 8]), end = tensor<int32, [3]>([1, 448, 40]), end_mask = tensor<bool, [3]>([false, false, false]), update = x_13_cast_fp16, x = x_13_cast_fp16_state_input);
            write_state(data = x_13_cast_fp16_state_updated, input = x_13_cast_fp16_state);
            tensor<fp16, [1, 448, 32]> x_13_cast_fp16_delayed = slice_by_size(begin = tensor<int32, [3]>([0, 0, 0]), size = tensor<int32, [3]>([1, 448, 32]), x = x_13_cast_fp16_state_updated);
            tensor<fp16, [1, 448, 32]> x_15_cast_fp16 = add(x = x_13_cast_fp16_delayed, y = y_8_cast_fp16)[name = string("x_15_cast_fp16")];
            bool var_456 = const()[name = string("op_456"), val = bool(true)];
            int32 var_460 = const()[name = string("op_460"), val = int32(448)];
            int32 var_461 = const()[name = string("op_461"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_41_cast_fp16 = add(x = x_15_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_41_cast_fp16")];
            tensor<int32, [1]> var_471 = const()[name = string("op_471"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_473 = const()[name = string("op_473"), val = tensor<int32, [1]>([1])];
            string input0_17_pad_type_0 = const()[name = string("input0_17_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_17_pad_0 = const()[name = string("input0_17_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_4_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3160960))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3261376))))[name = string("sep_module_4_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_17_cast_fp16 = conv(dilations = var_473, groups = var_461, pad = input0_17_pad_0, pad_type = input0_17_pad_type_0, strides = var_471, weight = sep_module_4_tcn_0_weight_to_fp16_palettized, x = input_41_cast_fp16)[name = string("input0_17_cast_fp16")];
            fp32 var_477_alpha_1 = const()[name = string("op_477_alpha_1"), val = fp32(0x1.b81c32p-2)];
            tensor<fp16, [1, 448, 32]> var_477_cast_fp16 = leaky_relu(alpha = fp16(0x1.b8p-2), x = input0_17_cast_fp16);
            tensor<int32, [1]> var_481 = const()[name = string("op_481"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_21_cast_fp16 = reduce_mean(axes = var_481, keep_dims = var_456, x = var_477_cast_fp16)[name = string("mean_y_21_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_483_cast_fp16 = sub(x = var_477_cast_fp16, y = mean_y_21_cast_fp16)[name = string("op_483_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_484_cast_fp16 = square(x = var_483_cast_fp16);
            tensor<int32, [1]> var_485 = const()[name = string("op_485"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_486_cast_fp16 = reduce_mean(axes = var_485, keep_dims = var_456, x = var_484_cast_fp16)[name = string("op_486_cast_fp16")];
            fp16 var_487_to_fp16 = const()[name = string("op_487_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_488_cast_fp16 = add(x = var_486_cast_fp16, y = var_487_to_fp16)[name = string("op_488_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_21_cast_fp16 = sqrt(x = var_488_cast_fp16)[name = string("std_y_21_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_4_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_4_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3261504)))];
            tensor<fp16, [1, 448, 32]> var_491_cast_fp16 = mul(x = sep_module_4_tcn_2_norm_gamma_to_fp16, y = var_483_cast_fp16)[name = string("op_491_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_492_cast_fp16 = real_div(x = var_491_cast_fp16, y = std_y_21_cast_fp16)[name = string("op_492_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_4_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_4_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3262464)))];
            tensor<fp16, [1, 448, 32]> input_43_cast_fp16 = add(x = var_492_cast_fp16, y = sep_module_4_tcn_2_norm_beta_to_fp16)[name = string("input_43_cast_fp16")];
            tensor<int32, [6]> input_45_pad_0 = const()[name = string("input_45_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 16, 16])];
            string input_45_mode_0 = const()[name = string("input_45_mode_0"), val = string("constant")];
            fp16 const_4_to_fp16 = const()[name = string("const_4_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 64]> input_43_cast_fp16_state_input = read_state(input = input_43_cast_fp16_state);
            tensor<fp16, [1, 448, 64]> input_45_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 32]), end = tensor<int32, [3]>([1, 448, 64]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_43_cast_fp16, x = input_43_cast_fp16_state_input);
            write_state(data = input_45_cast_fp16, input = input_43_cast_fp16_state);
            tensor<int32, [1]> var_497 = const()[name = string("op_497"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_499 = const()[name = string("op_499"), val = tensor<int32, [1]>([16])];
            string input_47_pad_type_0 = const()[name = string("input_47_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_47_pad_0 = const()[name = string("input_47_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_4_tcn_4_weight_to_fp16 = const()[name = string("sep_module_4_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3263424)))];
            tensor<fp16, [1, 448, 32]> input_47_cast_fp16 = conv(dilations = var_499, groups = var_460, pad = input_47_pad_0, pad_type = input_47_pad_type_0, strides = var_497, weight = sep_module_4_tcn_4_weight_to_fp16, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")];
            fp32 var_503_alpha_1 = const()[name = string("op_503_alpha_1"), val = fp32(-0x1.4023bcp-1)];
            tensor<fp16, [1, 448, 32]> var_503_cast_fp16 = leaky_relu(alpha = fp16(-0x1.404p-1), x = input_47_cast_fp16);
            tensor<int32, [1]> var_507 = const()[name = string("op_507"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_23_cast_fp16 = reduce_mean(axes = var_507, keep_dims = var_456, x = var_503_cast_fp16)[name = string("mean_y_23_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_509_cast_fp16 = sub(x = var_503_cast_fp16, y = mean_y_23_cast_fp16)[name = string("op_509_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_510_cast_fp16 = square(x = var_509_cast_fp16);
            tensor<int32, [1]> var_511 = const()[name = string("op_511"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_512_cast_fp16 = reduce_mean(axes = var_511, keep_dims = var_456, x = var_510_cast_fp16)[name = string("op_512_cast_fp16")];
            fp16 var_513_to_fp16 = const()[name = string("op_513_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_514_cast_fp16 = add(x = var_512_cast_fp16, y = var_513_to_fp16)[name = string("op_514_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_23_cast_fp16 = sqrt(x = var_514_cast_fp16)[name = string("std_y_23_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_4_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_4_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3266176)))];
            tensor<fp16, [1, 448, 32]> var_517_cast_fp16 = mul(x = sep_module_4_tcn_6_norm_gamma_to_fp16, y = var_509_cast_fp16)[name = string("op_517_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_518_cast_fp16 = real_div(x = var_517_cast_fp16, y = std_y_23_cast_fp16)[name = string("op_518_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_4_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_4_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3267136)))];
            tensor<fp16, [1, 448, 32]> y_10_cast_fp16 = add(x = var_518_cast_fp16, y = sep_module_4_tcn_6_norm_beta_to_fp16)[name = string("y_10_cast_fp16")];
            tensor<fp16, [1, 448, 48]> x_15_cast_fp16_state_input = read_state(input = x_15_cast_fp16_state);
            tensor<fp16, [1, 448, 48]> x_15_cast_fp16_state_updated = slice_update(begin = tensor<int32, [3]>([0, 0, 16]), end = tensor<int32, [3]>([1, 448, 48]), end_mask = tensor<bool, [3]>([false, false, false]), update = x_15_cast_fp16, x = x_15_cast_fp16_state_input);
            write_state(data = x_15_cast_fp16_state_updated, input = x_15_cast_fp16_state);
            tensor<fp16, [1, 448, 32]> x_15_cast_fp16_delayed = slice_by_size(begin = tensor<int32, [3]>([0, 0, 0]), size = tensor<int32, [3]>([1, 448, 32]), x = x_15_cast_fp16_state_updated);
            tensor<fp16, [1, 448, 32]> x_17_cast_fp16 = add(x = x_15_cast_fp16_delayed, y = y_10_cast_fp16)[name = string("x_17_cast_fp16")];
            bool var_524 = const()[name = string("op_524"), val = bool(true)];
            int32 var_528 = const()[name = string("op_528"), val = int32(448)];
            int32 var_529 = const()[name = string("op_529"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_49_cast_fp16 = add(x = x_17_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_49_cast_fp16")];
            tensor<int32, [1]> var_539 = const()[name = string("op_539"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_541 = const()[name = string("op_541"), val = tensor<int32, [1]>([1])];
            string input0_19_pad_type_0 = const()[name = string("input0_19_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_19_pad_0 = const()[name = string("input0_19_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_5_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3268096))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3368512))))[name = string("sep_module_5_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_19_cast_fp16 = conv(dilations = var_541, groups = var_529, pad = input0_19_pad_0, pad_type = input0_19_pad_type_0, strides = var_539, weight = sep_module_5_tcn_0_weight_to_fp16_palettized, x = input_49_cast_fp16)[name = string("input0_19_cast_fp16")];
            fp32 var_545_alpha_1 = const()[name = string("op_545_alpha_1"), val = fp32(0x1.29755cp-1)];
            tensor<fp16, [1, 448, 32]> var_545_cast_fp16 = leaky_relu(alpha = fp16(0x1.298p-1), x = input0_19_cast_fp16);
            tensor<int32, [1]> var_549 = const()[name = string("op_549"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_25_cast_fp16 = reduce_mean(axes = var_549, keep_dims = var_524, x = var_545_cast_fp16)[name = string("mean_y_25_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_551_cast_fp16 = sub(x = var_545_cast_fp16, y = mean_y_25_cast_fp16)[name = string("op_551_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_552_cast_fp16 = square(x = var_551_cast_fp16);
            tensor<int32, [1]> var_553 = const()[name = string("op_553"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_554_cast_fp16 = reduce_mean(axes = var_553, keep_dims = var_524, x = var_552_cast_fp16)[name = string("op_554_cast_fp16")];
            fp16 var_555_to_fp16 = const()[name = string("op_555_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_556_cast_fp16 = add(x = var_554_cast_fp16, y = var_555_to_fp16)[name = string("op_556_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_25_cast_fp16 = sqrt(x = var_556_cast_fp16)[name = string("std_y_25_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_5_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_5_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3368640)))];
            tensor<fp16, [1, 448, 32]> var_559_cast_fp16 = mul(x = sep_module_5_tcn_2_norm_gamma_to_fp16, y = var_551_cast_fp16)[name = string("op_559_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_560_cast_fp16 = real_div(x = var_559_cast_fp16, y = std_y_25_cast_fp16)[name = string("op_560_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_5_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_5_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3369600)))];
            tensor<fp16, [1, 448, 32]> input_51_cast_fp16 = add(x = var_560_cast_fp16, y = sep_module_5_tcn_2_norm_beta_to_fp16)[name = string("input_51_cast_fp16")];
            tensor<int32, [6]> input_53_pad_0 = const()[name = string("input_53_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 32, 32])];
            string input_53_mode_0 = const()[name = string("input_53_mode_0"), val = string("constant")];
            fp16 const_5_to_fp16 = const()[name = string("const_5_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 96]> input_51_cast_fp16_state_input = read_state(input = input_51_cast_fp16_state);
            tensor<fp16, [1, 448, 96]> input_53_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 64]), end = tensor<int32, [3]>([1, 448, 96]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_51_cast_fp16, x = input_51_cast_fp16_state_input);
            write_state(data = input_53_cast_fp16, input = input_51_cast_fp16_state);
            tensor<int32, [1]> var_565 = const()[name = string("op_565"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_567 = const()[name = string("op_567"), val = tensor<int32, [1]>([32])];
            string input_55_pad_type_0 = const()[name = string("input_55_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_55_pad_0 = const()[name = string("input_55_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_5_tcn_4_weight_to_fp16 = const()[name = string("sep_module_5_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3370560)))];
            tensor<fp16, [1, 448, 32]> input_55_cast_fp16 = conv(dilations = var_567, groups = var_528, pad = input_55_pad_0, pad_type = input_55_pad_type_0, strides = var_565, weight = sep_module_5_tcn_4_weight_to_fp16, x = input_53_cast_fp16)[name = string("input_55_cast_fp16")];
            fp32 var_571_alpha_1 = const()[name = string("op_571_alpha_1"), val = fp32(-0x1.40e7c6p-2)];
            tensor<fp16, [1, 448, 32]> var_571_cast_fp16 = leaky_relu(alpha = fp16(-0x1.41p-2), x = input_55_cast_fp16);
            tensor<int32, [1]> var_575 = const()[name = string("op_575"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_27_cast_fp16 = reduce_mean(axes = var_575, keep_dims = var_524, x = var_571_cast_fp16)[name = string("mean_y_27_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_577_cast_fp16 = sub(x = var_571_cast_fp16, y = mean_y_27_cast_fp16)[name = string("op_577_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_578_cast_fp16 = square(x = var_577_cast_fp16);
            tensor<int32, [1]> var_579 = const()[name = string("op_579"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_580_cast_fp16 = reduce_mean(axes = var_579, keep_dims = var_524, x = var_578_cast_fp16)[name = string("op_580_cast_fp16")];
            fp16 var_581_to_fp16 = const()[name = string("op_581_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_582_cast_fp16 = add(x = var_580_cast_fp16, y = var_581_to_fp16)[name = string("op_582_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_27_cast_fp16 = sqrt(x = var_582_cast_fp16)[name = string("std_y_27_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_5_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_5_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3373312)))];
            tensor<fp16, [1, 448, 32]> var_585_cast_fp16 = mul(x = sep_module_5_tcn_6_norm_gamma_to_fp16, y = var_577_cast_fp16)[name = string("op_585_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_586_cast_fp16 = real_div(x = var_585_cast_fp16, y = std_y_27_cast_fp16)[name = string("op_586_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_5_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_5_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3374272)))];
            tensor<fp16, [1, 448, 32]> y_12_cast_fp16 = add(x = var_586_cast_fp16, y = sep_module_5_tcn_6_norm_beta_to_fp16)[name = string("y_12_cast_fp16")];
            tensor<fp16, [1, 448, 64]> x_17_cast_fp16_state_input = read_state(input = x_17_cast_fp16_state);
            tensor<fp16, [1, 448, 64]> x_17_cast_fp16_state_updated = slice_update(begin = tensor<int32, [3]>([0, 0, 32]), end = tensor<int32, [3]>([1, 448, 64]), end_mask = tensor<bool, [3]>([false, false, false]), update = x_17_cast_fp16, x = x_17_cast_fp16_state_input);
            write_state(data = x_17_cast_fp16_state_updated, input = x_17_cast_fp16_state);
            tensor<fp16, [1, 448, 32]> x_17_cast_fp16_delayed = slice_by_size(begin = tensor<int32, [3]>([0, 0, 0]), size = tensor<int32, [3]>([1, 448, 32]), x = x_17_cast_fp16_state_updated);
            tensor<fp16, [1, 448, 32]> x_19_cast_fp16 = add(x = x_17_cast_fp16_delayed, y = y_12_cast_fp16)[name = string("x_19_cast_fp16")];
            bool var_592 = const()[name = string("op_592"), val = bool(true)];
            int32 var_596 = const()[name = string("op_596"), val = int32(448)];
            int32 var_597 = const()[name = string("op_597"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_57_cast_fp16 = add(x = x_19_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_57_cast_fp16")];
            tensor<int32, [1]> var_607 = const()[name = string("op_607"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_609 = const()[name = string("op_609"), val = tensor<int32, [1]>([1])];
            string input0_21_pad_type_0 = const()[name = string("input0_21_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_21_pad_0 = const()[name = string("input0_21_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_6_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3375232))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3475648))))[name = string("sep_module_6_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_21_cast_fp16 = conv(dilations = var_609, groups = var_597, pad = input0_21_pad_0, pad_type = input0_21_pad_type_0, strides = var_607, weight = sep_module_6_tcn_0_weight_to_fp16_palettized, x = input_57_cast_fp16)[name = string("input0_21_cast_fp16")];
            fp32 var_613_alpha_1 = const()[name = string("op_613_alpha_1"), val = fp32(0x1.58c914p-1)];
            tensor<fp16, [1, 448, 32]> var_613_cast_fp16 = leaky_relu(alpha = fp16(0x1.58cp-1), x = input0_21_cast_fp16);
            tensor<int32, [1]> var_617 = const()[name = string("op_617"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_29_cast_fp16 = reduce_mean(axes = var_617, keep_dims = var_592, x = var_613_cast_fp16)[name = string("mean_y_29_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_619_cast_fp16 = sub(x = var_613_cast_fp16, y = mean_y_29_cast_fp16)[name = string("op_619_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_620_cast_fp16 = square(x = var_619_cast_fp16);
            tensor<int32, [1]> var_621 = const()[name = string("op_621"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_622_cast_fp16 = reduce_mean(axes = var_621, keep_dims = var_592, x = var_620_cast_fp16)[name = string("op_622_cast_fp16")];
            fp16 var_623_to_fp16 = const()[name = string("op_623_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_624_cast_fp16 = add(x = var_622_cast_fp16, y = var_623_to_fp16)[name = string("op_624_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_29_cast_fp16 = sqrt(x = var_624_cast_fp16)[name = string("std_y_29_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_6_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_6_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3475776)))];
            tensor<fp16, [1, 448, 32]> var_627_cast_fp16 = mul(x = sep_module_6_tcn_2_norm_gamma_to_fp16, y = var_619_cast_fp16)[name = string("op_627_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_628_cast_fp16 = real_div(x = var_627_cast_fp16, y = std_y_29_cast_fp16)[name = string("op_628_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_6_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_6_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3476736)))];
            tensor<fp16, [1, 448, 32]> input_59_cast_fp16 = add(x = var_628_cast_fp16, y = sep_module_6_tcn_2_norm_beta_to_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, 64, 64])];
            string input_61_mode_0 = const()[name = string("input_61_mode_0"), val = string("constant")];
            fp16 const_6_to_fp16 = const()[name = string("const_6_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 160]> input_59_cast_fp16_state_input = read_state(input = input_59_cast_fp16_state);
            tensor<fp16, [1, 448, 160]> input_61_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 128]), end = tensor<int32, [3]>([1, 448, 160]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_59_cast_fp16, x = input_59_cast_fp16_state_input);
            write_state(data = input_61_cast_fp16, input = input_59_cast_fp16_state);
            tensor<int32, [1]> var_633 = const()[name = string("op_633"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_635 = const()[name = string("op_635"), val = tensor<int32, [1]>([64])];
            string input_63_pad_type_0 = const()[name = string("input_63_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_63_pad_0 = const()[name = string("input_63_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_6_tcn_4_weight_to_fp16 = const()[name = string("sep_module_6_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3477696)))];
            tensor<fp16, [1, 448, 32]> input_63_cast_fp16 = conv(dilations = var_635, groups = var_596, pad = input_63_pad_0, pad_type = input_63_pad_type_0, strides = var_633, weight = sep_module_6_tcn_4_weight_to_fp16, x = input_61_cast_fp16)[name = string("input_63_cast_fp16")];
            fp32 var_639_alpha_1 = const()[name = string("op_639_alpha_1"), val = fp32(-0x1.583c4cp-2)];
            tensor<fp16, [1, 448, 32]> var_639_cast_fp16 = leaky_relu(alpha = fp16(-0x1.584p-2), x = input_63_cast_fp16);
            tensor<int32, [1]> var_643 = const()[name = string("op_643"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_31_cast_fp16 = reduce_mean(axes = var_643, keep_dims = var_592, x = var_639_cast_fp16)[name = string("mean_y_31_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_645_cast_fp16 = sub(x = var_639_cast_fp16, y = mean_y_31_cast_fp16)[name = string("op_645_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_646_cast_fp16 = square(x = var_645_cast_fp16);
            tensor<int32, [1]> var_647 = const()[name = string("op_647"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_648_cast_fp16 = reduce_mean(axes = var_647, keep_dims = var_592, x = var_646_cast_fp16)[name = string("op_648_cast_fp16")];
            fp16 var_649_to_fp16 = const()[name = string("op_649_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_650_cast_fp16 = add(x = var_648_cast_fp16, y = var_649_to_fp16)[name = string("op_650_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_31_cast_fp16 = sqrt(x = var_650_cast_fp16)[name = string("std_y_31_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_6_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_6_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3480448)))];
            tensor<fp16, [1, 448, 32]> var_653_cast_fp16 = mul(x = sep_module_6_tcn_6_norm_gamma_to_fp16, y = var_645_cast_fp16)[name = string("op_653_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_654_cast_fp16 = real_div(x = var_653_cast_fp16, y = std_y_31_cast_fp16)[name = string("op_654_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_6_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_6_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3481408)))];
            tensor<fp16, [1, 448, 32]> y_14_cast_fp16 = add(x = var_654_cast_fp16, y = sep_module_6_tcn_6_norm_beta_to_fp16)[name = string("y_14_cast_fp16")];
            tensor<fp16, [1, 448, 96]> x_19_cast_fp16_state_input = read_state(input = x_19_cast_fp16_state);
            tensor<fp16, [1, 448, 96]> x_19_cast_fp16_state_updated = slice_update(begin = tensor<int32, [3]>([0, 0, 64]), end = tensor<int32, [3]>([1, 448, 96]), end_mask = tensor<bool, [3]>([false, false, false]), update = x_19_cast_fp16, x = x_19_cast_fp16_state_input);
            write_state(data = x_19_cast_fp16_state_updated, input = x_19_cast_fp16_state);
            tensor<fp16, [1, 448, 32]> x_19_cast_fp16_delayed = slice_by_size(begin = tensor<int32, [3]>([0, 0, 0]), size = tensor<int32, [3]>([1, 448, 32]), x = x_19_cast_fp16_state_updated);
            tensor<fp16, [1, 448, 32]> x_21_cast_fp16 = add(x = x_19_cast_fp16_delayed, y = y_14_cast_fp16)[name = string("x_21_cast_fp16")];
            bool var_660 = const()[name = string("op_660"), val = bool(true)];
            int32 var_664 = const()[name = string("op_664"), val = int32(448)];
            int32 var_665 = const()[name = string("op_665"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_65_cast_fp16 = add(x = x_21_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_65_cast_fp16")];
            tensor<int32, [1]> var_675 = const()[name = string("op_675"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_677 = const()[name = string("op_677"), val = tensor<int32, [1]>([1])];
            string input0_23_pad_type_0 = const()[name = string("input0_23_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_23_pad_0 = const()[name = string("input0_23_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_7_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3482368))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3582784))))[name = string("sep_module_7_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_23_cast_fp16 = conv(dilations = var_677, groups = var_665, pad = input0_23_pad_0, pad_type = input0_23_pad_type_0, strides = var_675, weight = sep_module_7_tcn_0_weight_to_fp16_palettized, x = input_65_cast_fp16)[name = string("input0_23_cast_fp16")];
            fp32 var_681_alpha_1 = const()[name = string("op_681_alpha_1"), val = fp32(0x1.38257ap-1)];
            tensor<fp16, [1, 448, 32]> var_681_cast_fp16 = leaky_relu(alpha = fp16(0x1.384p-1), x = input0_23_cast_fp16);
            tensor<int32, [1]> var_685 = const()[name = string("op_685"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_33_cast_fp16 = reduce_mean(axes = var_685, keep_dims = var_660, x = var_681_cast_fp16)[name = string("mean_y_33_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_687_cast_fp16 = sub(x = var_681_cast_fp16, y = mean_y_33_cast_fp16)[name = string("op_687_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_688_cast_fp16 = square(x = var_687_cast_fp16);
            tensor<int32, [1]> var_689 = const()[name = string("op_689"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_690_cast_fp16 = reduce_mean(axes = var_689, keep_dims = var_660, x = var_688_cast_fp16)[name = string("op_690_cast_fp16")];
            fp16 var_691_to_fp16 = const()[name = string("op_691_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_692_cast_fp16 = add(x = var_690_cast_fp16, y = var_691_to_fp16)[name = string("op_692_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_33_cast_fp16 = sqrt(x = var_692_cast_fp16)[name = string("std_y_33_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_7_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_7_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3582912)))];
            tensor<fp16, [1, 448, 32]> var_695_cast_fp16 = mul(x = sep_module_7_tcn_2_norm_gamma_to_fp16, y = var_687_cast_fp16)[name = string("op_695_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_696_cast_fp16 = real_div(x = var_695_cast_fp16, y = std_y_33_cast_fp16)[name = string("op_696_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_7_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_7_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3583872)))];
            tensor<fp16, [1, 448, 32]> input_67_cast_fp16 = add(x = var_696_cast_fp16, y = sep_module_7_tcn_2_norm_beta_to_fp16)[name = string("input_67_cast_fp16")];
            tensor<int32, [6]> input_69_pad_0 = const()[name = string("input_69_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 128, 128])];
            string input_69_mode_0 = const()[name = string("input_69_mode_0"), val = string("constant")];
            fp16 const_7_to_fp16 = const()[name = string("const_7_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 288]> input_67_cast_fp16_state_input = read_state(input = input_67_cast_fp16_state);
            tensor<fp16, [1, 448, 288]> input_69_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 256]), end = tensor<int32, [3]>([1, 448, 288]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_67_cast_fp16, x = input_67_cast_fp16_state_input);
            write_state(data = input_69_cast_fp16, input = input_67_cast_fp16_state);
            tensor<int32, [1]> var_701 = const()[name = string("op_701"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_703 = const()[name = string("op_703"), val = tensor<int32, [1]>([128])];
            string input_71_pad_type_0 = const()[name = string("input_71_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_71_pad_0 = const()[name = string("input_71_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_7_tcn_4_weight_to_fp16 = const()[name = string("sep_module_7_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3584832)))];
            tensor<fp16, [1, 448, 32]> input_71_cast_fp16 = conv(dilations = var_703, groups = var_664, pad = input_71_pad_0, pad_type = input_71_pad_type_0, strides = var_701, weight = sep_module_7_tcn_4_weight_to_fp16, x = input_69_cast_fp16)[name = string("input_71_cast_fp16")];
            fp32 var_707_alpha_1 = const()[name = string("op_707_alpha_1"), val = fp32(-0x1.d9b8eap-3)];
            tensor<fp16, [1, 448, 32]> var_707_cast_fp16 = leaky_relu(alpha = fp16(-0x1.d9cp-3), x = input_71_cast_fp16);
            tensor<int32, [1]> var_711 = const()[name = string("op_711"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_35_cast_fp16 = reduce_mean(axes = var_711, keep_dims = var_660, x = var_707_cast_fp16)[name = string("mean_y_35_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_713_cast_fp16 = sub(x = var_707_cast_fp16, y = mean_y_35_cast_fp16)[name = string("op_713_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_714_cast_fp16 = square(x = var_713_cast_fp16);
            tensor<int32, [1]> var_715 = const()[name = string("op_715"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_716_cast_fp16 = reduce_mean(axes = var_715, keep_dims = var_660, x = var_714_cast_fp16)[name = string("op_716_cast_fp16")];
            fp16 var_717_to_fp16 = const()[name = string("op_717_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_718_cast_fp16 = add(x = var_716_cast_fp16, y = var_717_to_fp16)[name = string("op_718_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_35_cast_fp16 = sqrt(x = var_718_cast_fp16)[name = string("std_y_35_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_7_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_7_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3587584)))];
            tensor<fp16, [1, 448, 32]> var_721_cast_fp16 = mul(x = sep_module_7_tcn_6_norm_gamma_to_fp16, y = var_713_cast_fp16)[name = string("op_721_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_722_cast_fp16 = real_div(x = var_721_cast_fp16, y = std_y_35_cast_fp16)[name = string("op_722_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_7_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_7_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3588544)))];
            tensor<fp16, [1, 448, 32]> y_16_cast_fp16 = add(x = var_722_cast_fp16, y = sep_module_7_tcn_6_norm_beta_to_fp16)[name = string("y_16_cast_fp16")];
            tensor<fp16, [1, 448, 160]> x_21_cast_fp16_state_input = read_state(input = x_21_cast_fp16_state);
            tensor<fp16, [1, 448, 160]> x_21_cast_fp16_state_updated = slice_update(begin = tensor<int32, [3]>([0, 0, 128]), end = tensor<int32, [3]>([1, 448, 160]), end_mask = tensor<bool, [3]>([false, false, false]), update = x_21_cast_fp16, x = x_21_cast_fp16_state_input);
            write_state(data = x_21_cast_fp16_state_updated, input = x_21_cast_fp16_state);
            tensor<fp16, [1, 448, 32]> x_21_cast_fp16_delayed = slice_by_size(begin = tensor<int32, [3]>([0, 0, 0]), size = tensor<int32, [3]>([1, 448, 32]), x = x_21_cast_fp16_state_updated);
            tensor<fp16, [1, 448, 32]> x_23_cast_fp16 = add(x = x_21_cast_fp16_delayed, y = y_16_cast_fp16)[name = string("x_23_cast_fp16")];
            bool var_728 = const()[name = string("op_728"), val = bool(true)];
            int32 var_732 = const()[name = string("op_732"), val = int32(448)];
            int32 var_734 = const()[name = string("op_734"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_73_cast_fp16 = add(x = x_23_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_73_cast_fp16")];
            tensor<int32, [1]> var_744 = const()[name = string("op_744"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_746 = const()[name = string("op_746"), val = tensor<int32, [1]>([1])];
            string input0_25_pad_type_0 = const()[name = string("input0_25_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_25_pad_0 = const()[name = string("input0_25_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_8_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3589504))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3689920))))[name = string("sep_module_8_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_25_cast_fp16 = conv(dilations = var_746, groups = var_734, pad = input0_25_pad_0, pad_type = input0_25_pad_type_0, strides = var_744, weight = sep_module_8_tcn_0_weight_to_fp16_palettized, x = input_73_cast_fp16)[name = string("input0_25_cast_fp16")];
            fp32 var_750_alpha_1 = const()[name = string("op_750_alpha_1"), val = fp32(0x1.b819dcp-2)];
            tensor<fp16, [1, 448, 32]> var_750_cast_fp16 = leaky_relu(alpha = fp16(0x1.b8p-2), x = input0_25_cast_fp16);
            tensor<int32, [1]> var_754 = const()[name = string("op_754"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_37_cast_fp16 = reduce_mean(axes = var_754, keep_dims = var_728, x = var_750_cast_fp16)[name = string("mean_y_37_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_756_cast_fp16 = sub(x = var_750_cast_fp16, y = mean_y_37_cast_fp16)[name = string("op_756_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_757_cast_fp16 = square(x = var_756_cast_fp16);
            tensor<int32, [1]> var_758 = const()[name = string("op_758"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_759_cast_fp16 = reduce_mean(axes = var_758, keep_dims = var_728, x = var_757_cast_fp16)[name = string("op_759_cast_fp16")];
            fp16 var_760_to_fp16 = const()[name = string("op_760_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_761_cast_fp16 = add(x = var_759_cast_fp16, y = var_760_to_fp16)[name = string("op_761_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_37_cast_fp16 = sqrt(x = var_761_cast_fp16)[name = string("std_y_37_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_8_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_8_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3690048)))];
            tensor<fp16, [1, 448, 32]> var_764_cast_fp16 = mul(x = sep_module_8_tcn_2_norm_gamma_to_fp16, y = var_756_cast_fp16)[name = string("op_764_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_765_cast_fp16 = real_div(x = var_764_cast_fp16, y = std_y_37_cast_fp16)[name = string("op_765_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_8_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_8_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3691008)))];
            tensor<fp16, [1, 448, 32]> input_75_cast_fp16 = add(x = var_765_cast_fp16, y = sep_module_8_tcn_2_norm_beta_to_fp16)[name = string("input_75_cast_fp16")];
            tensor<int32, [6]> input_77_pad_0 = const()[name = string("input_77_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 512, 0])];
            string input_77_mode_0 = const()[name = string("input_77_mode_0"), val = string("constant")];
            fp16 const_8_to_fp16 = const()[name = string("const_8_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 544]> input_75_cast_fp16_state_input = read_state(input = input_75_cast_fp16_state);
            tensor<fp16, [1, 448, 544]> input_77_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 512]), end = tensor<int32, [3]>([1, 448, 544]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_75_cast_fp16, x = input_75_cast_fp16_state_input);
            write_state(data = input_77_cast_fp16, input = input_75_cast_fp16_state);
            tensor<int32, [1]> var_770 = const()[name = string("op_770"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_772 = const()[name = string("op_772"), val = tensor<int32, [1]>([256])];
            string input_79_pad_type_0 = const()[name = string("input_79_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_79_pad_0 = const()[name = string("input_79_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_8_tcn_4_weight_to_fp16 = const()[name = string("sep_module_8_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3691968)))];
            tensor<fp16, [1, 448, 32]> input_79_cast_fp16 = conv(dilations = var_772, groups = var_732, pad = input_79_pad_0, pad_type = input_79_pad_type_0, strides = var_770, weight = sep_module_8_tcn_4_weight_to_fp16, x = input_77_cast_fp16)[name = string("input_79_cast_fp16")];
            fp32 var_776_alpha_1 = const()[name = string("op_776_alpha_1"), val = fp32(-0x1.24c8ccp-3)];
            tensor<fp16, [1, 448, 32]> var_776_cast_fp16 = leaky_relu(alpha = fp16(-0x1.24cp-3), x = input_79_cast_fp16);
            tensor<int32, [1]> var_780 = const()[name = string("op_780"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_39_cast_fp16 = reduce_mean(axes = var_780, keep_dims = var_728, x = var_776_cast_fp16)[name = string("mean_y_39_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_782_cast_fp16 = sub(x = var_776_cast_fp16, y = mean_y_39_cast_fp16)[name = string("op_782_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_783_cast_fp16 = square(x = var_782_cast_fp16);
            tensor<int32, [1]> var_784 = const()[name = string("op_784"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_785_cast_fp16 = reduce_mean(axes = var_784, keep_dims = var_728, x = var_783_cast_fp16)[name = string("op_785_cast_fp16")];
            fp16 var_786_to_fp16 = const()[name = string("op_786_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_787_cast_fp16 = add(x = var_785_cast_fp16, y = var_786_to_fp16)[name = string("op_787_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_39_cast_fp16 = sqrt(x = var_787_cast_fp16)[name = string("std_y_39_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_8_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_8_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3694720)))];
            tensor<fp16, [1, 448, 32]> var_790_cast_fp16 = mul(x = sep_module_8_tcn_6_norm_gamma_to_fp16, y = var_782_cast_fp16)[name = string("op_790_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_791_cast_fp16 = real_div(x = var_790_cast_fp16, y = std_y_39_cast_fp16)[name = string("op_791_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_8_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_8_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3695680)))];
            tensor<fp16, [1, 448, 32]> y_18_cast_fp16 = add(x = var_791_cast_fp16, y = sep_module_8_tcn_6_norm_beta_to_fp16)[name = string("y_18_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_25_cast_fp16 = add(x = x_23_cast_fp16, y = y_18_cast_fp16)[name = string("x_25_cast_fp16")];
            bool var_797 = const()[name = string("op_797"), val = bool(true)];
            int32 var_801 = const()[name = string("op_801"), val = int32(448)];
            int32 var_802 = const()[name = string("op_802"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_81_cast_fp16 = add(x = x_25_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_81_cast_fp16")];
            tensor<int32, [1]> var_812 = const()[name = string("op_812"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_814 = const()[name = string("op_814"), val = tensor<int32, [1]>([1])];
            string input0_27_pad_type_0 = const()[name = string("input0_27_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_27_pad_0 = const()[name = string("input0_27_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_9_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3696640))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3797056))))[name = string("sep_module_9_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_27_cast_fp16 = conv(dilations = var_814, groups = var_802, pad = input0_27_pad_0, pad_type = input0_27_pad_type_0, strides = var_812, weight = sep_module_9_tcn_0_weight_to_fp16_palettized, x = input_81_cast_fp16)[name = string("input0_27_cast_fp16")];
            fp32 var_818_alpha_1 = const()[name = string("op_818_alpha_1"), val = fp32(-0x1.27cc4ap-2)];
            tensor<fp16, [1, 448, 32]> var_818_cast_fp16 = leaky_relu(alpha = fp16(-0x1.27cp-2), x = input0_27_cast_fp16);
            tensor<int32, [1]> var_822 = const()[name = string("op_822"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_41_cast_fp16 = reduce_mean(axes = var_822, keep_dims = var_797, x = var_818_cast_fp16)[name = string("mean_y_41_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_824_cast_fp16 = sub(x = var_818_cast_fp16, y = mean_y_41_cast_fp16)[name = string("op_824_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_825_cast_fp16 = square(x = var_824_cast_fp16);
            tensor<int32, [1]> var_826 = const()[name = string("op_826"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_827_cast_fp16 = reduce_mean(axes = var_826, keep_dims = var_797, x = var_825_cast_fp16)[name = string("op_827_cast_fp16")];
            fp16 var_828_to_fp16 = const()[name = string("op_828_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_829_cast_fp16 = add(x = var_827_cast_fp16, y = var_828_to_fp16)[name = string("op_829_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_41_cast_fp16 = sqrt(x = var_829_cast_fp16)[name = string("std_y_41_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_9_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_9_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3797184)))];
            tensor<fp16, [1, 448, 32]> var_832_cast_fp16 = mul(x = sep_module_9_tcn_2_norm_gamma_to_fp16, y = var_824_cast_fp16)[name = string("op_832_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_833_cast_fp16 = real_div(x = var_832_cast_fp16, y = std_y_41_cast_fp16)[name = string("op_833_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_9_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_9_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3798144)))];
            tensor<fp16, [1, 448, 32]> input_83_cast_fp16 = add(x = var_833_cast_fp16, y = sep_module_9_tcn_2_norm_beta_to_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, 0])];
            string input_85_mode_0 = const()[name = string("input_85_mode_0"), val = string("constant")];
            fp16 const_9_to_fp16 = const()[name = string("const_9_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 34]> input_83_cast_fp16_state_input = read_state(input = input_83_cast_fp16_state);
            tensor<fp16, [1, 448, 34]> input_85_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 2]), end = tensor<int32, [3]>([1, 448, 34]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_83_cast_fp16, x = input_83_cast_fp16_state_input);
            write_state(data = input_85_cast_fp16, input = input_83_cast_fp16_state);
            tensor<int32, [1]> var_838 = const()[name = string("op_838"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_840 = const()[name = string("op_840"), val = tensor<int32, [1]>([1])];
            string input_87_pad_type_0 = const()[name = string("input_87_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_87_pad_0 = const()[name = string("input_87_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_9_tcn_4_weight_to_fp16 = const()[name = string("sep_module_9_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3799104)))];
            tensor<fp16, [1, 448, 32]> input_87_cast_fp16 = conv(dilations = var_840, groups = var_801, pad = input_87_pad_0, pad_type = input_87_pad_type_0, strides = var_838, weight = sep_module_9_tcn_4_weight_to_fp16, x = input_85_cast_fp16)[name = string("input_87_cast_fp16")];
            fp32 var_844_alpha_1 = const()[name = string("op_844_alpha_1"), val = fp32(0x1.bb6d42p-1)];
            tensor<fp16, [1, 448, 32]> var_844_cast_fp16 = leaky_relu(alpha = fp16(0x1.bb8p-1), x = input_87_cast_fp16);
            tensor<int32, [1]> var_848 = const()[name = string("op_848"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_43_cast_fp16 = reduce_mean(axes = var_848, keep_dims = var_797, x = var_844_cast_fp16)[name = string("mean_y_43_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_850_cast_fp16 = sub(x = var_844_cast_fp16, y = mean_y_43_cast_fp16)[name = string("op_850_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_851_cast_fp16 = square(x = var_850_cast_fp16);
            tensor<int32, [1]> var_852 = const()[name = string("op_852"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_853_cast_fp16 = reduce_mean(axes = var_852, keep_dims = var_797, x = var_851_cast_fp16)[name = string("op_853_cast_fp16")];
            fp16 var_854_to_fp16 = const()[name = string("op_854_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_855_cast_fp16 = add(x = var_853_cast_fp16, y = var_854_to_fp16)[name = string("op_855_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_43_cast_fp16 = sqrt(x = var_855_cast_fp16)[name = string("std_y_43_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_9_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_9_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3801856)))];
            tensor<fp16, [1, 448, 32]> var_858_cast_fp16 = mul(x = sep_module_9_tcn_6_norm_gamma_to_fp16, y = var_850_cast_fp16)[name = string("op_858_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_859_cast_fp16 = real_div(x = var_858_cast_fp16, y = std_y_43_cast_fp16)[name = string("op_859_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_9_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_9_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3802816)))];
            tensor<fp16, [1, 448, 32]> y_20_cast_fp16 = add(x = var_859_cast_fp16, y = sep_module_9_tcn_6_norm_beta_to_fp16)[name = string("y_20_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_27_cast_fp16 = add(x = x_25_cast_fp16, y = y_20_cast_fp16)[name = string("x_27_cast_fp16")];
            bool var_865 = const()[name = string("op_865"), val = bool(true)];
            int32 var_869 = const()[name = string("op_869"), val = int32(448)];
            int32 var_871 = const()[name = string("op_871"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_89_cast_fp16 = add(x = x_27_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_89_cast_fp16")];
            tensor<int32, [1]> var_881 = const()[name = string("op_881"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_883 = const()[name = string("op_883"), val = tensor<int32, [1]>([1])];
            string input0_29_pad_type_0 = const()[name = string("input0_29_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_29_pad_0 = const()[name = string("input0_29_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_10_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3803776))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3904192))))[name = string("sep_module_10_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_29_cast_fp16 = conv(dilations = var_883, groups = var_871, pad = input0_29_pad_0, pad_type = input0_29_pad_type_0, strides = var_881, weight = sep_module_10_tcn_0_weight_to_fp16_palettized, x = input_89_cast_fp16)[name = string("input0_29_cast_fp16")];
            fp32 var_887_alpha_1 = const()[name = string("op_887_alpha_1"), val = fp32(0x1.dd1642p-2)];
            tensor<fp16, [1, 448, 32]> var_887_cast_fp16 = leaky_relu(alpha = fp16(0x1.ddp-2), x = input0_29_cast_fp16);
            tensor<int32, [1]> var_891 = const()[name = string("op_891"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_45_cast_fp16 = reduce_mean(axes = var_891, keep_dims = var_865, x = var_887_cast_fp16)[name = string("mean_y_45_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_893_cast_fp16 = sub(x = var_887_cast_fp16, y = mean_y_45_cast_fp16)[name = string("op_893_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_894_cast_fp16 = square(x = var_893_cast_fp16);
            tensor<int32, [1]> var_895 = const()[name = string("op_895"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_896_cast_fp16 = reduce_mean(axes = var_895, keep_dims = var_865, x = var_894_cast_fp16)[name = string("op_896_cast_fp16")];
            fp16 var_897_to_fp16 = const()[name = string("op_897_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_898_cast_fp16 = add(x = var_896_cast_fp16, y = var_897_to_fp16)[name = string("op_898_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_45_cast_fp16 = sqrt(x = var_898_cast_fp16)[name = string("std_y_45_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_10_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_10_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3904320)))];
            tensor<fp16, [1, 448, 32]> var_901_cast_fp16 = mul(x = sep_module_10_tcn_2_norm_gamma_to_fp16, y = var_893_cast_fp16)[name = string("op_901_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_902_cast_fp16 = real_div(x = var_901_cast_fp16, y = std_y_45_cast_fp16)[name = string("op_902_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_10_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_10_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3905280)))];
            tensor<fp16, [1, 448, 32]> input_91_cast_fp16 = add(x = var_902_cast_fp16, y = sep_module_10_tcn_2_norm_beta_to_fp16)[name = string("input_91_cast_fp16")];
            tensor<int32, [6]> input_93_pad_0 = const()[name = string("input_93_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 0])];
            string input_93_mode_0 = const()[name = string("input_93_mode_0"), val = string("constant")];
            fp16 const_10_to_fp16 = const()[name = string("const_10_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 36]> input_91_cast_fp16_state_input = read_state(input = input_91_cast_fp16_state);
            tensor<fp16, [1, 448, 36]> input_93_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 4]), end = tensor<int32, [3]>([1, 448, 36]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_91_cast_fp16, x = input_91_cast_fp16_state_input);
            write_state(data = input_93_cast_fp16, input = input_91_cast_fp16_state);
            tensor<int32, [1]> var_907 = const()[name = string("op_907"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_909 = const()[name = string("op_909"), val = tensor<int32, [1]>([2])];
            string input_95_pad_type_0 = const()[name = string("input_95_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_95_pad_0 = const()[name = string("input_95_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_10_tcn_4_weight_to_fp16 = const()[name = string("sep_module_10_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3906240)))];
            tensor<fp16, [1, 448, 32]> input_95_cast_fp16 = conv(dilations = var_909, groups = var_869, pad = input_95_pad_0, pad_type = input_95_pad_type_0, strides = var_907, weight = sep_module_10_tcn_4_weight_to_fp16, x = input_93_cast_fp16)[name = string("input_95_cast_fp16")];
            fp32 var_913_alpha_1 = const()[name = string("op_913_alpha_1"), val = fp32(0x1.21aa5cp-1)];
            tensor<fp16, [1, 448, 32]> var_913_cast_fp16 = leaky_relu(alpha = fp16(0x1.21cp-1), x = input_95_cast_fp16);
            tensor<int32, [1]> var_917 = const()[name = string("op_917"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_47_cast_fp16 = reduce_mean(axes = var_917, keep_dims = var_865, x = var_913_cast_fp16)[name = string("mean_y_47_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_919_cast_fp16 = sub(x = var_913_cast_fp16, y = mean_y_47_cast_fp16)[name = string("op_919_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_920_cast_fp16 = square(x = var_919_cast_fp16);
            tensor<int32, [1]> var_921 = const()[name = string("op_921"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_922_cast_fp16 = reduce_mean(axes = var_921, keep_dims = var_865, x = var_920_cast_fp16)[name = string("op_922_cast_fp16")];
            fp16 var_923_to_fp16 = const()[name = string("op_923_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_924_cast_fp16 = add(x = var_922_cast_fp16, y = var_923_to_fp16)[name = string("op_924_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_47_cast_fp16 = sqrt(x = var_924_cast_fp16)[name = string("std_y_47_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_10_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_10_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3908992)))];
            tensor<fp16, [1, 448, 32]> var_927_cast_fp16 = mul(x = sep_module_10_tcn_6_norm_gamma_to_fp16, y = var_919_cast_fp16)[name = string("op_927_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_928_cast_fp16 = real_div(x = var_927_cast_fp16, y = std_y_47_cast_fp16)[name = string("op_928_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_10_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_10_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3909952)))];
            tensor<fp16, [1, 448, 32]> y_22_cast_fp16 = add(x = var_928_cast_fp16, y = sep_module_10_tcn_6_norm_beta_to_fp16)[name = string("y_22_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_29_cast_fp16 = add(x = x_27_cast_fp16, y = y_22_cast_fp16)[name = string("x_29_cast_fp16")];
            bool var_934 = const()[name = string("op_934"), val = bool(true)];
            int32 var_938 = const()[name = string("op_938"), val = int32(448)];
            int32 var_940 = const()[name = string("op_940"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_97_cast_fp16 = add(x = x_29_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_97_cast_fp16")];
            tensor<int32, [1]> var_950 = const()[name = string("op_950"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_952 = const()[name = string("op_952"), val = tensor<int32, [1]>([1])];
            string input0_31_pad_type_0 = const()[name = string("input0_31_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_31_pad_0 = const()[name = string("input0_31_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_11_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(3910912))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4011328))))[name = string("sep_module_11_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_31_cast_fp16 = conv(dilations = var_952, groups = var_940, pad = input0_31_pad_0, pad_type = input0_31_pad_type_0, strides = var_950, weight = sep_module_11_tcn_0_weight_to_fp16_palettized, x = input_97_cast_fp16)[name = string("input0_31_cast_fp16")];
            fp32 var_956_alpha_1 = const()[name = string("op_956_alpha_1"), val = fp32(-0x1.0aabaap-2)];
            tensor<fp16, [1, 448, 32]> var_956_cast_fp16 = leaky_relu(alpha = fp16(-0x1.0acp-2), x = input0_31_cast_fp16);
            tensor<int32, [1]> var_960 = const()[name = string("op_960"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_49_cast_fp16 = reduce_mean(axes = var_960, keep_dims = var_934, x = var_956_cast_fp16)[name = string("mean_y_49_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_962_cast_fp16 = sub(x = var_956_cast_fp16, y = mean_y_49_cast_fp16)[name = string("op_962_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_963_cast_fp16 = square(x = var_962_cast_fp16);
            tensor<int32, [1]> var_964 = const()[name = string("op_964"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_965_cast_fp16 = reduce_mean(axes = var_964, keep_dims = var_934, x = var_963_cast_fp16)[name = string("op_965_cast_fp16")];
            fp16 var_966_to_fp16 = const()[name = string("op_966_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_967_cast_fp16 = add(x = var_965_cast_fp16, y = var_966_to_fp16)[name = string("op_967_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_49_cast_fp16 = sqrt(x = var_967_cast_fp16)[name = string("std_y_49_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_11_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_11_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4011456)))];
            tensor<fp16, [1, 448, 32]> var_970_cast_fp16 = mul(x = sep_module_11_tcn_2_norm_gamma_to_fp16, y = var_962_cast_fp16)[name = string("op_970_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_971_cast_fp16 = real_div(x = var_970_cast_fp16, y = std_y_49_cast_fp16)[name = string("op_971_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_11_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_11_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4012416)))];
            tensor<fp16, [1, 448, 32]> input_99_cast_fp16 = add(x = var_971_cast_fp16, y = sep_module_11_tcn_2_norm_beta_to_fp16)[name = string("input_99_cast_fp16")];
            tensor<int32, [6]> input_101_pad_0 = const()[name = string("input_101_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 8, 0])];
            string input_101_mode_0 = const()[name = string("input_101_mode_0"), val = string("constant")];
            fp16 const_11_to_fp16 = const()[name = string("const_11_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 40]> input_99_cast_fp16_state_input = read_state(input = input_99_cast_fp16_state);
            tensor<fp16, [1, 448, 40]> input_101_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 8]), end = tensor<int32, [3]>([1, 448, 40]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_99_cast_fp16, x = input_99_cast_fp16_state_input);
            write_state(data = input_101_cast_fp16, input = input_99_cast_fp16_state);
            tensor<int32, [1]> var_976 = const()[name = string("op_976"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_978 = const()[name = string("op_978"), val = tensor<int32, [1]>([4])];
            string input_103_pad_type_0 = const()[name = string("input_103_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_103_pad_0 = const()[name = string("input_103_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_11_tcn_4_weight_to_fp16 = const()[name = string("sep_module_11_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4013376)))];
            tensor<fp16, [1, 448, 32]> input_103_cast_fp16 = conv(dilations = var_978, groups = var_938, pad = input_103_pad_0, pad_type = input_103_pad_type_0, strides = var_976, weight = sep_module_11_tcn_4_weight_to_fp16, x = input_101_cast_fp16)[name = string("input_103_cast_fp16")];
            fp32 var_982_alpha_1 = const()[name = string("op_982_alpha_1"), val = fp32(0x1.92c246p-3)];
            tensor<fp16, [1, 448, 32]> var_982_cast_fp16 = leaky_relu(alpha = fp16(0x1.92cp-3), x = input_103_cast_fp16);
            tensor<int32, [1]> var_986 = const()[name = string("op_986"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_51_cast_fp16 = reduce_mean(axes = var_986, keep_dims = var_934, x = var_982_cast_fp16)[name = string("mean_y_51_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_988_cast_fp16 = sub(x = var_982_cast_fp16, y = mean_y_51_cast_fp16)[name = string("op_988_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_989_cast_fp16 = square(x = var_988_cast_fp16);
            tensor<int32, [1]> var_990 = const()[name = string("op_990"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_991_cast_fp16 = reduce_mean(axes = var_990, keep_dims = var_934, x = var_989_cast_fp16)[name = string("op_991_cast_fp16")];
            fp16 var_992_to_fp16 = const()[name = string("op_992_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_993_cast_fp16 = add(x = var_991_cast_fp16, y = var_992_to_fp16)[name = string("op_993_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_51_cast_fp16 = sqrt(x = var_993_cast_fp16)[name = string("std_y_51_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_11_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_11_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4016128)))];
            tensor<fp16, [1, 448, 32]> var_996_cast_fp16 = mul(x = sep_module_11_tcn_6_norm_gamma_to_fp16, y = var_988_cast_fp16)[name = string("op_996_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_997_cast_fp16 = real_div(x = var_996_cast_fp16, y = std_y_51_cast_fp16)[name = string("op_997_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_11_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_11_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4017088)))];
            tensor<fp16, [1, 448, 32]> y_24_cast_fp16 = add(x = var_997_cast_fp16, y = sep_module_11_tcn_6_norm_beta_to_fp16)[name = string("y_24_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_31_cast_fp16 = add(x = x_29_cast_fp16, y = y_24_cast_fp16)[name = string("x_31_cast_fp16")];
            bool var_1003 = const()[name = string("op_1003"), val = bool(true)];
            int32 var_1007 = const()[name = string("op_1007"), val = int32(448)];
            int32 var_1009 = const()[name = string("op_1009"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_105_cast_fp16 = add(x = x_31_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_105_cast_fp16")];
            tensor<int32, [1]> var_1019 = const()[name = string("op_1019"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1021 = const()[name = string("op_1021"), val = tensor<int32, [1]>([1])];
            string input0_33_pad_type_0 = const()[name = string("input0_33_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_33_pad_0 = const()[name = string("input0_33_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_12_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4018048))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4118464))))[name = string("sep_module_12_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_33_cast_fp16 = conv(dilations = var_1021, groups = var_1009, pad = input0_33_pad_0, pad_type = input0_33_pad_type_0, strides = var_1019, weight = sep_module_12_tcn_0_weight_to_fp16_palettized, x = input_105_cast_fp16)[name = string("input0_33_cast_fp16")];
            fp32 var_1025_alpha_1 = const()[name = string("op_1025_alpha_1"), val = fp32(0x1.cc226cp-2)];
            tensor<fp16, [1, 448, 32]> var_1025_cast_fp16 = leaky_relu(alpha = fp16(0x1.cc4p-2), x = input0_33_cast_fp16);
            tensor<int32, [1]> var_1029 = const()[name = string("op_1029"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_53_cast_fp16 = reduce_mean(axes = var_1029, keep_dims = var_1003, x = var_1025_cast_fp16)[name = string("mean_y_53_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1031_cast_fp16 = sub(x = var_1025_cast_fp16, y = mean_y_53_cast_fp16)[name = string("op_1031_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1032_cast_fp16 = square(x = var_1031_cast_fp16);
            tensor<int32, [1]> var_1033 = const()[name = string("op_1033"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1034_cast_fp16 = reduce_mean(axes = var_1033, keep_dims = var_1003, x = var_1032_cast_fp16)[name = string("op_1034_cast_fp16")];
            fp16 var_1035_to_fp16 = const()[name = string("op_1035_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1036_cast_fp16 = add(x = var_1034_cast_fp16, y = var_1035_to_fp16)[name = string("op_1036_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_53_cast_fp16 = sqrt(x = var_1036_cast_fp16)[name = string("std_y_53_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_12_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_12_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4118592)))];
            tensor<fp16, [1, 448, 32]> var_1039_cast_fp16 = mul(x = sep_module_12_tcn_2_norm_gamma_to_fp16, y = var_1031_cast_fp16)[name = string("op_1039_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1040_cast_fp16 = real_div(x = var_1039_cast_fp16, y = std_y_53_cast_fp16)[name = string("op_1040_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_12_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_12_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4119552)))];
            tensor<fp16, [1, 448, 32]> input_107_cast_fp16 = add(x = var_1040_cast_fp16, y = sep_module_12_tcn_2_norm_beta_to_fp16)[name = string("input_107_cast_fp16")];
            tensor<int32, [6]> input_109_pad_0 = const()[name = string("input_109_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 16, 0])];
            string input_109_mode_0 = const()[name = string("input_109_mode_0"), val = string("constant")];
            fp16 const_12_to_fp16 = const()[name = string("const_12_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 48]> input_107_cast_fp16_state_input = read_state(input = input_107_cast_fp16_state);
            tensor<fp16, [1, 448, 48]> input_109_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 16]), end = tensor<int32, [3]>([1, 448, 48]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_107_cast_fp16, x = input_107_cast_fp16_state_input);
            write_state(data = input_109_cast_fp16, input = input_107_cast_fp16_state);
            tensor<int32, [1]> var_1045 = const()[name = string("op_1045"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1047 = const()[name = string("op_1047"), val = tensor<int32, [1]>([8])];
            string input_111_pad_type_0 = const()[name = string("input_111_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_111_pad_0 = const()[name = string("input_111_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_12_tcn_4_weight_to_fp16 = const()[name = string("sep_module_12_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4120512)))];
            tensor<fp16, [1, 448, 32]> input_111_cast_fp16 = conv(dilations = var_1047, groups = var_1007, pad = input_111_pad_0, pad_type = input_111_pad_type_0, strides = var_1045, weight = sep_module_12_tcn_4_weight_to_fp16, x = input_109_cast_fp16)[name = string("input_111_cast_fp16")];
            fp32 var_1051_alpha_1 = const()[name = string("op_1051_alpha_1"), val = fp32(0x1.af352p-7)];
            tensor<fp16, [1, 448, 32]> var_1051_cast_fp16 = leaky_relu(alpha = fp16(0x1.af4p-7), x = input_111_cast_fp16);
            tensor<int32, [1]> var_1055 = const()[name = string("op_1055"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_55_cast_fp16 = reduce_mean(axes = var_1055, keep_dims = var_1003, x = var_1051_cast_fp16)[name = string("mean_y_55_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1057_cast_fp16 = sub(x = var_1051_cast_fp16, y = mean_y_55_cast_fp16)[name = string("op_1057_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1058_cast_fp16 = square(x = var_1057_cast_fp16);
            tensor<int32, [1]> var_1059 = const()[name = string("op_1059"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1060_cast_fp16 = reduce_mean(axes = var_1059, keep_dims = var_1003, x = var_1058_cast_fp16)[name = string("op_1060_cast_fp16")];
            fp16 var_1061_to_fp16 = const()[name = string("op_1061_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1062_cast_fp16 = add(x = var_1060_cast_fp16, y = var_1061_to_fp16)[name = string("op_1062_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_55_cast_fp16 = sqrt(x = var_1062_cast_fp16)[name = string("std_y_55_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_12_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_12_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4123264)))];
            tensor<fp16, [1, 448, 32]> var_1065_cast_fp16 = mul(x = sep_module_12_tcn_6_norm_gamma_to_fp16, y = var_1057_cast_fp16)[name = string("op_1065_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1066_cast_fp16 = real_div(x = var_1065_cast_fp16, y = std_y_55_cast_fp16)[name = string("op_1066_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_12_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_12_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4124224)))];
            tensor<fp16, [1, 448, 32]> y_26_cast_fp16 = add(x = var_1066_cast_fp16, y = sep_module_12_tcn_6_norm_beta_to_fp16)[name = string("y_26_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_33_cast_fp16 = add(x = x_31_cast_fp16, y = y_26_cast_fp16)[name = string("x_33_cast_fp16")];
            bool var_1072 = const()[name = string("op_1072"), val = bool(true)];
            int32 var_1076 = const()[name = string("op_1076"), val = int32(448)];
            int32 var_1078 = const()[name = string("op_1078"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_113_cast_fp16 = add(x = x_33_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_113_cast_fp16")];
            tensor<int32, [1]> var_1088 = const()[name = string("op_1088"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1090 = const()[name = string("op_1090"), val = tensor<int32, [1]>([1])];
            string input0_35_pad_type_0 = const()[name = string("input0_35_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_35_pad_0 = const()[name = string("input0_35_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_13_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4125184))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4225600))))[name = string("sep_module_13_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_35_cast_fp16 = conv(dilations = var_1090, groups = var_1078, pad = input0_35_pad_0, pad_type = input0_35_pad_type_0, strides = var_1088, weight = sep_module_13_tcn_0_weight_to_fp16_palettized, x = input_113_cast_fp16)[name = string("input0_35_cast_fp16")];
            fp32 var_1094_alpha_1 = const()[name = string("op_1094_alpha_1"), val = fp32(0x1.8f680ap-2)];
            tensor<fp16, [1, 448, 32]> var_1094_cast_fp16 = leaky_relu(alpha = fp16(0x1.8f8p-2), x = input0_35_cast_fp16);
            tensor<int32, [1]> var_1098 = const()[name = string("op_1098"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_57_cast_fp16 = reduce_mean(axes = var_1098, keep_dims = var_1072, x = var_1094_cast_fp16)[name = string("mean_y_57_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1100_cast_fp16 = sub(x = var_1094_cast_fp16, y = mean_y_57_cast_fp16)[name = string("op_1100_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1101_cast_fp16 = square(x = var_1100_cast_fp16);
            tensor<int32, [1]> var_1102 = const()[name = string("op_1102"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1103_cast_fp16 = reduce_mean(axes = var_1102, keep_dims = var_1072, x = var_1101_cast_fp16)[name = string("op_1103_cast_fp16")];
            fp16 var_1104_to_fp16 = const()[name = string("op_1104_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1105_cast_fp16 = add(x = var_1103_cast_fp16, y = var_1104_to_fp16)[name = string("op_1105_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_57_cast_fp16 = sqrt(x = var_1105_cast_fp16)[name = string("std_y_57_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_13_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_13_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4225728)))];
            tensor<fp16, [1, 448, 32]> var_1108_cast_fp16 = mul(x = sep_module_13_tcn_2_norm_gamma_to_fp16, y = var_1100_cast_fp16)[name = string("op_1108_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1109_cast_fp16 = real_div(x = var_1108_cast_fp16, y = std_y_57_cast_fp16)[name = string("op_1109_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_13_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_13_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4226688)))];
            tensor<fp16, [1, 448, 32]> input_115_cast_fp16 = add(x = var_1109_cast_fp16, y = sep_module_13_tcn_2_norm_beta_to_fp16)[name = string("input_115_cast_fp16")];
            tensor<int32, [6]> input_117_pad_0 = const()[name = string("input_117_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 32, 0])];
            string input_117_mode_0 = const()[name = string("input_117_mode_0"), val = string("constant")];
            fp16 const_13_to_fp16 = const()[name = string("const_13_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 64]> input_115_cast_fp16_state_input = read_state(input = input_115_cast_fp16_state);
            tensor<fp16, [1, 448, 64]> input_117_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 32]), end = tensor<int32, [3]>([1, 448, 64]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_115_cast_fp16, x = input_115_cast_fp16_state_input);
            write_state(data = input_117_cast_fp16, input = input_115_cast_fp16_state);
            tensor<int32, [1]> var_1114 = const()[name = string("op_1114"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1116 = const()[name = string("op_1116"), val = tensor<int32, [1]>([16])];
            string input_119_pad_type_0 = const()[name = string("input_119_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_119_pad_0 = const()[name = string("input_119_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_13_tcn_4_weight_to_fp16 = const()[name = string("sep_module_13_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4227648)))];
            tensor<fp16, [1, 448, 32]> input_119_cast_fp16 = conv(dilations = var_1116, groups = var_1076, pad = input_119_pad_0, pad_type = input_119_pad_type_0, strides = var_1114, weight = sep_module_13_tcn_4_weight_to_fp16, x = input_117_cast_fp16)[name = string("input_119_cast_fp16")];
            fp32 var_1120_alpha_1 = const()[name = string("op_1120_alpha_1"), val = fp32(-0x1.154838p-3)];
            tensor<fp16, [1, 448, 32]> var_1120_cast_fp16 = leaky_relu(alpha = fp16(-0x1.154p-3), x = input_119_cast_fp16);
            tensor<int32, [1]> var_1124 = const()[name = string("op_1124"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_59_cast_fp16 = reduce_mean(axes = var_1124, keep_dims = var_1072, x = var_1120_cast_fp16)[name = string("mean_y_59_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1126_cast_fp16 = sub(x = var_1120_cast_fp16, y = mean_y_59_cast_fp16)[name = string("op_1126_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1127_cast_fp16 = square(x = var_1126_cast_fp16);
            tensor<int32, [1]> var_1128 = const()[name = string("op_1128"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1129_cast_fp16 = reduce_mean(axes = var_1128, keep_dims = var_1072, x = var_1127_cast_fp16)[name = string("op_1129_cast_fp16")];
            fp16 var_1130_to_fp16 = const()[name = string("op_1130_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1131_cast_fp16 = add(x = var_1129_cast_fp16, y = var_1130_to_fp16)[name = string("op_1131_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_59_cast_fp16 = sqrt(x = var_1131_cast_fp16)[name = string("std_y_59_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_13_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_13_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4230400)))];
            tensor<fp16, [1, 448, 32]> var_1134_cast_fp16 = mul(x = sep_module_13_tcn_6_norm_gamma_to_fp16, y = var_1126_cast_fp16)[name = string("op_1134_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1135_cast_fp16 = real_div(x = var_1134_cast_fp16, y = std_y_59_cast_fp16)[name = string("op_1135_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_13_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_13_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4231360)))];
            tensor<fp16, [1, 448, 32]> y_28_cast_fp16 = add(x = var_1135_cast_fp16, y = sep_module_13_tcn_6_norm_beta_to_fp16)[name = string("y_28_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_35_cast_fp16 = add(x = x_33_cast_fp16, y = y_28_cast_fp16)[name = string("x_35_cast_fp16")];
            bool var_1141 = const()[name = string("op_1141"), val = bool(true)];
            int32 var_1145 = const()[name = string("op_1145"), val = int32(448)];
            int32 var_1147 = const()[name = string("op_1147"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_121_cast_fp16 = add(x = x_35_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_121_cast_fp16")];
            tensor<int32, [1]> var_1157 = const()[name = string("op_1157"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1159 = const()[name = string("op_1159"), val = tensor<int32, [1]>([1])];
            string input0_37_pad_type_0 = const()[name = string("input0_37_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_37_pad_0 = const()[name = string("input0_37_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_14_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4232320))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4332736))))[name = string("sep_module_14_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_37_cast_fp16 = conv(dilations = var_1159, groups = var_1147, pad = input0_37_pad_0, pad_type = input0_37_pad_type_0, strides = var_1157, weight = sep_module_14_tcn_0_weight_to_fp16_palettized, x = input_121_cast_fp16)[name = string("input0_37_cast_fp16")];
            fp32 var_1163_alpha_1 = const()[name = string("op_1163_alpha_1"), val = fp32(0x1.c9bb7cp-2)];
            tensor<fp16, [1, 448, 32]> var_1163_cast_fp16 = leaky_relu(alpha = fp16(0x1.c9cp-2), x = input0_37_cast_fp16);
            tensor<int32, [1]> var_1167 = const()[name = string("op_1167"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_61_cast_fp16 = reduce_mean(axes = var_1167, keep_dims = var_1141, x = var_1163_cast_fp16)[name = string("mean_y_61_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1169_cast_fp16 = sub(x = var_1163_cast_fp16, y = mean_y_61_cast_fp16)[name = string("op_1169_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1170_cast_fp16 = square(x = var_1169_cast_fp16);
            tensor<int32, [1]> var_1171 = const()[name = string("op_1171"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1172_cast_fp16 = reduce_mean(axes = var_1171, keep_dims = var_1141, x = var_1170_cast_fp16)[name = string("op_1172_cast_fp16")];
            fp16 var_1173_to_fp16 = const()[name = string("op_1173_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1174_cast_fp16 = add(x = var_1172_cast_fp16, y = var_1173_to_fp16)[name = string("op_1174_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_61_cast_fp16 = sqrt(x = var_1174_cast_fp16)[name = string("std_y_61_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_14_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_14_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4332864)))];
            tensor<fp16, [1, 448, 32]> var_1177_cast_fp16 = mul(x = sep_module_14_tcn_2_norm_gamma_to_fp16, y = var_1169_cast_fp16)[name = string("op_1177_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1178_cast_fp16 = real_div(x = var_1177_cast_fp16, y = std_y_61_cast_fp16)[name = string("op_1178_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_14_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_14_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4333824)))];
            tensor<fp16, [1, 448, 32]> input_123_cast_fp16 = add(x = var_1178_cast_fp16, y = sep_module_14_tcn_2_norm_beta_to_fp16)[name = string("input_123_cast_fp16")];
            tensor<int32, [6]> input_125_pad_0 = const()[name = string("input_125_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 64, 0])];
            string input_125_mode_0 = const()[name = string("input_125_mode_0"), val = string("constant")];
            fp16 const_14_to_fp16 = const()[name = string("const_14_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 96]> input_123_cast_fp16_state_input = read_state(input = input_123_cast_fp16_state);
            tensor<fp16, [1, 448, 96]> input_125_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 64]), end = tensor<int32, [3]>([1, 448, 96]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_123_cast_fp16, x = input_123_cast_fp16_state_input);
            write_state(data = input_125_cast_fp16, input = input_123_cast_fp16_state);
            tensor<int32, [1]> var_1183 = const()[name = string("op_1183"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1185 = const()[name = string("op_1185"), val = tensor<int32, [1]>([32])];
            string input_127_pad_type_0 = const()[name = string("input_127_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_127_pad_0 = const()[name = string("input_127_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_14_tcn_4_weight_to_fp16 = const()[name = string("sep_module_14_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4334784)))];
            tensor<fp16, [1, 448, 32]> input_127_cast_fp16 = conv(dilations = var_1185, groups = var_1145, pad = input_127_pad_0, pad_type = input_127_pad_type_0, strides = var_1183, weight = sep_module_14_tcn_4_weight_to_fp16, x = input_125_cast_fp16)[name = string("input_127_cast_fp16")];
            fp32 var_1189_alpha_1 = const()[name = string("op_1189_alpha_1"), val = fp32(-0x1.d0fec2p-7)];
            tensor<fp16, [1, 448, 32]> var_1189_cast_fp16 = leaky_relu(alpha = fp16(-0x1.d1p-7), x = input_127_cast_fp16);
            tensor<int32, [1]> var_1193 = const()[name = string("op_1193"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_63_cast_fp16 = reduce_mean(axes = var_1193, keep_dims = var_1141, x = var_1189_cast_fp16)[name = string("mean_y_63_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1195_cast_fp16 = sub(x = var_1189_cast_fp16, y = mean_y_63_cast_fp16)[name = string("op_1195_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1196_cast_fp16 = square(x = var_1195_cast_fp16);
            tensor<int32, [1]> var_1197 = const()[name = string("op_1197"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1198_cast_fp16 = reduce_mean(axes = var_1197, keep_dims = var_1141, x = var_1196_cast_fp16)[name = string("op_1198_cast_fp16")];
            fp16 var_1199_to_fp16 = const()[name = string("op_1199_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1200_cast_fp16 = add(x = var_1198_cast_fp16, y = var_1199_to_fp16)[name = string("op_1200_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_63_cast_fp16 = sqrt(x = var_1200_cast_fp16)[name = string("std_y_63_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_14_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_14_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4337536)))];
            tensor<fp16, [1, 448, 32]> var_1203_cast_fp16 = mul(x = sep_module_14_tcn_6_norm_gamma_to_fp16, y = var_1195_cast_fp16)[name = string("op_1203_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1204_cast_fp16 = real_div(x = var_1203_cast_fp16, y = std_y_63_cast_fp16)[name = string("op_1204_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_14_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_14_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4338496)))];
            tensor<fp16, [1, 448, 32]> y_30_cast_fp16 = add(x = var_1204_cast_fp16, y = sep_module_14_tcn_6_norm_beta_to_fp16)[name = string("y_30_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_37_cast_fp16 = add(x = x_35_cast_fp16, y = y_30_cast_fp16)[name = string("x_37_cast_fp16")];
            bool var_1210 = const()[name = string("op_1210"), val = bool(true)];
            int32 var_1214 = const()[name = string("op_1214"), val = int32(448)];
            int32 var_1216 = const()[name = string("op_1216"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_129_cast_fp16 = add(x = x_37_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_129_cast_fp16")];
            tensor<int32, [1]> var_1226 = const()[name = string("op_1226"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1228 = const()[name = string("op_1228"), val = tensor<int32, [1]>([1])];
            string input0_39_pad_type_0 = const()[name = string("input0_39_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_39_pad_0 = const()[name = string("input0_39_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_15_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4339456))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4439872))))[name = string("sep_module_15_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_39_cast_fp16 = conv(dilations = var_1228, groups = var_1216, pad = input0_39_pad_0, pad_type = input0_39_pad_type_0, strides = var_1226, weight = sep_module_15_tcn_0_weight_to_fp16_palettized, x = input_129_cast_fp16)[name = string("input0_39_cast_fp16")];
            fp32 var_1232_alpha_1 = const()[name = string("op_1232_alpha_1"), val = fp32(0x1.2fbae2p-2)];
            tensor<fp16, [1, 448, 32]> var_1232_cast_fp16 = leaky_relu(alpha = fp16(0x1.2fcp-2), x = input0_39_cast_fp16);
            tensor<int32, [1]> var_1236 = const()[name = string("op_1236"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_65_cast_fp16 = reduce_mean(axes = var_1236, keep_dims = var_1210, x = var_1232_cast_fp16)[name = string("mean_y_65_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1238_cast_fp16 = sub(x = var_1232_cast_fp16, y = mean_y_65_cast_fp16)[name = string("op_1238_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1239_cast_fp16 = square(x = var_1238_cast_fp16);
            tensor<int32, [1]> var_1240 = const()[name = string("op_1240"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1241_cast_fp16 = reduce_mean(axes = var_1240, keep_dims = var_1210, x = var_1239_cast_fp16)[name = string("op_1241_cast_fp16")];
            fp16 var_1242_to_fp16 = const()[name = string("op_1242_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1243_cast_fp16 = add(x = var_1241_cast_fp16, y = var_1242_to_fp16)[name = string("op_1243_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_65_cast_fp16 = sqrt(x = var_1243_cast_fp16)[name = string("std_y_65_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_15_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_15_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4440000)))];
            tensor<fp16, [1, 448, 32]> var_1246_cast_fp16 = mul(x = sep_module_15_tcn_2_norm_gamma_to_fp16, y = var_1238_cast_fp16)[name = string("op_1246_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1247_cast_fp16 = real_div(x = var_1246_cast_fp16, y = std_y_65_cast_fp16)[name = string("op_1247_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_15_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_15_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4440960)))];
            tensor<fp16, [1, 448, 32]> input_131_cast_fp16 = add(x = var_1247_cast_fp16, y = sep_module_15_tcn_2_norm_beta_to_fp16)[name = string("input_131_cast_fp16")];
            tensor<int32, [6]> input_133_pad_0 = const()[name = string("input_133_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 128, 0])];
            string input_133_mode_0 = const()[name = string("input_133_mode_0"), val = string("constant")];
            fp16 const_15_to_fp16 = const()[name = string("const_15_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 160]> input_131_cast_fp16_state_input = read_state(input = input_131_cast_fp16_state);
            tensor<fp16, [1, 448, 160]> input_133_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 128]), end = tensor<int32, [3]>([1, 448, 160]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_131_cast_fp16, x = input_131_cast_fp16_state_input);
            write_state(data = input_133_cast_fp16, input = input_131_cast_fp16_state);
            tensor<int32, [1]> var_1252 = const()[name = string("op_1252"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1254 = const()[name = string("op_1254"), val = tensor<int32, [1]>([64])];
            string input_135_pad_type_0 = const()[name = string("input_135_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_135_pad_0 = const()[name = string("input_135_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_15_tcn_4_weight_to_fp16 = const()[name = string("sep_module_15_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4441920)))];
            tensor<fp16, [1, 448, 32]> input_135_cast_fp16 = conv(dilations = var_1254, groups = var_1214, pad = input_135_pad_0, pad_type = input_135_pad_type_0, strides = var_1252, weight = sep_module_15_tcn_4_weight_to_fp16, x = input_133_cast_fp16)[name = string("input_135_cast_fp16")];
            fp32 var_1258_alpha_1 = const()[name = string("op_1258_alpha_1"), val = fp32(-0x1.12f4bcp-2)];
            tensor<fp16, [1, 448, 32]> var_1258_cast_fp16 = leaky_relu(alpha = fp16(-0x1.13p-2), x = input_135_cast_fp16);
            tensor<int32, [1]> var_1262 = const()[name = string("op_1262"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_67_cast_fp16 = reduce_mean(axes = var_1262, keep_dims = var_1210, x = var_1258_cast_fp16)[name = string("mean_y_67_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1264_cast_fp16 = sub(x = var_1258_cast_fp16, y = mean_y_67_cast_fp16)[name = string("op_1264_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1265_cast_fp16 = square(x = var_1264_cast_fp16);
            tensor<int32, [1]> var_1266 = const()[name = string("op_1266"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1267_cast_fp16 = reduce_mean(axes = var_1266, keep_dims = var_1210, x = var_1265_cast_fp16)[name = string("op_1267_cast_fp16")];
            fp16 var_1268_to_fp16 = const()[name = string("op_1268_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1269_cast_fp16 = add(x = var_1267_cast_fp16, y = var_1268_to_fp16)[name = string("op_1269_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_67_cast_fp16 = sqrt(x = var_1269_cast_fp16)[name = string("std_y_67_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_15_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_15_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4444672)))];
            tensor<fp16, [1, 448, 32]> var_1272_cast_fp16 = mul(x = sep_module_15_tcn_6_norm_gamma_to_fp16, y = var_1264_cast_fp16)[name = string("op_1272_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1273_cast_fp16 = real_div(x = var_1272_cast_fp16, y = std_y_67_cast_fp16)[name = string("op_1273_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_15_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_15_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4445632)))];
            tensor<fp16, [1, 448, 32]> y_32_cast_fp16 = add(x = var_1273_cast_fp16, y = sep_module_15_tcn_6_norm_beta_to_fp16)[name = string("y_32_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_39_cast_fp16 = add(x = x_37_cast_fp16, y = y_32_cast_fp16)[name = string("x_39_cast_fp16")];
            bool var_1279 = const()[name = string("op_1279"), val = bool(true)];
            int32 var_1283 = const()[name = string("op_1283"), val = int32(448)];
            int32 var_1285 = const()[name = string("op_1285"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_137_cast_fp16 = add(x = x_39_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_137_cast_fp16")];
            tensor<int32, [1]> var_1295 = const()[name = string("op_1295"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1297 = const()[name = string("op_1297"), val = tensor<int32, [1]>([1])];
            string input0_41_pad_type_0 = const()[name = string("input0_41_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_41_pad_0 = const()[name = string("input0_41_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_16_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4446592))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4547008))))[name = string("sep_module_16_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_41_cast_fp16 = conv(dilations = var_1297, groups = var_1285, pad = input0_41_pad_0, pad_type = input0_41_pad_type_0, strides = var_1295, weight = sep_module_16_tcn_0_weight_to_fp16_palettized, x = input_137_cast_fp16)[name = string("input0_41_cast_fp16")];
            fp32 var_1301_alpha_1 = const()[name = string("op_1301_alpha_1"), val = fp32(0x1.f188b2p-2)];
            tensor<fp16, [1, 448, 32]> var_1301_cast_fp16 = leaky_relu(alpha = fp16(0x1.f18p-2), x = input0_41_cast_fp16);
            tensor<int32, [1]> var_1305 = const()[name = string("op_1305"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_69_cast_fp16 = reduce_mean(axes = var_1305, keep_dims = var_1279, x = var_1301_cast_fp16)[name = string("mean_y_69_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1307_cast_fp16 = sub(x = var_1301_cast_fp16, y = mean_y_69_cast_fp16)[name = string("op_1307_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1308_cast_fp16 = square(x = var_1307_cast_fp16);
            tensor<int32, [1]> var_1309 = const()[name = string("op_1309"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1310_cast_fp16 = reduce_mean(axes = var_1309, keep_dims = var_1279, x = var_1308_cast_fp16)[name = string("op_1310_cast_fp16")];
            fp16 var_1311_to_fp16 = const()[name = string("op_1311_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1312_cast_fp16 = add(x = var_1310_cast_fp16, y = var_1311_to_fp16)[name = string("op_1312_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_69_cast_fp16 = sqrt(x = var_1312_cast_fp16)[name = string("std_y_69_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_16_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_16_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4547136)))];
            tensor<fp16, [1, 448, 32]> var_1315_cast_fp16 = mul(x = sep_module_16_tcn_2_norm_gamma_to_fp16, y = var_1307_cast_fp16)[name = string("op_1315_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1316_cast_fp16 = real_div(x = var_1315_cast_fp16, y = std_y_69_cast_fp16)[name = string("op_1316_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_16_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_16_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4548096)))];
            tensor<fp16, [1, 448, 32]> input_139_cast_fp16 = add(x = var_1316_cast_fp16, y = sep_module_16_tcn_2_norm_beta_to_fp16)[name = string("input_139_cast_fp16")];
            tensor<int32, [6]> input_141_pad_0 = const()[name = string("input_141_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 256, 0])];
            string input_141_mode_0 = const()[name = string("input_141_mode_0"), val = string("constant")];
            fp16 const_16_to_fp16 = const()[name = string("const_16_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 288]> input_139_cast_fp16_state_input = read_state(input = input_139_cast_fp16_state);
            tensor<fp16, [1, 448, 288]> input_141_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 256]), end = tensor<int32, [3]>([1, 448, 288]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_139_cast_fp16, x = input_139_cast_fp16_state_input);
            write_state(data = input_141_cast_fp16, input = input_139_cast_fp16_state);
            tensor<int32, [1]> var_1321 = const()[name = string("op_1321"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1323 = const()[name = string("op_1323"), val = tensor<int32, [1]>([128])];
            string input_143_pad_type_0 = const()[name = string("input_143_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_143_pad_0 = const()[name = string("input_143_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_16_tcn_4_weight_to_fp16 = const()[name = string("sep_module_16_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4549056)))];
            tensor<fp16, [1, 448, 32]> input_143_cast_fp16 = conv(dilations = var_1323, groups = var_1283, pad = input_143_pad_0, pad_type = input_143_pad_type_0, strides = var_1321, weight = sep_module_16_tcn_4_weight_to_fp16, x = input_141_cast_fp16)[name = string("input_143_cast_fp16")];
            fp32 var_1327_alpha_1 = const()[name = string("op_1327_alpha_1"), val = fp32(-0x1.17df32p-3)];
            tensor<fp16, [1, 448, 32]> var_1327_cast_fp16 = leaky_relu(alpha = fp16(-0x1.17cp-3), x = input_143_cast_fp16);
            tensor<int32, [1]> var_1331 = const()[name = string("op_1331"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_71_cast_fp16 = reduce_mean(axes = var_1331, keep_dims = var_1279, x = var_1327_cast_fp16)[name = string("mean_y_71_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1333_cast_fp16 = sub(x = var_1327_cast_fp16, y = mean_y_71_cast_fp16)[name = string("op_1333_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1334_cast_fp16 = square(x = var_1333_cast_fp16);
            tensor<int32, [1]> var_1335 = const()[name = string("op_1335"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1336_cast_fp16 = reduce_mean(axes = var_1335, keep_dims = var_1279, x = var_1334_cast_fp16)[name = string("op_1336_cast_fp16")];
            fp16 var_1337_to_fp16 = const()[name = string("op_1337_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1338_cast_fp16 = add(x = var_1336_cast_fp16, y = var_1337_to_fp16)[name = string("op_1338_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_71_cast_fp16 = sqrt(x = var_1338_cast_fp16)[name = string("std_y_71_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_16_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_16_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4551808)))];
            tensor<fp16, [1, 448, 32]> var_1341_cast_fp16 = mul(x = sep_module_16_tcn_6_norm_gamma_to_fp16, y = var_1333_cast_fp16)[name = string("op_1341_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1342_cast_fp16 = real_div(x = var_1341_cast_fp16, y = std_y_71_cast_fp16)[name = string("op_1342_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_16_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_16_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4552768)))];
            tensor<fp16, [1, 448, 32]> y_34_cast_fp16 = add(x = var_1342_cast_fp16, y = sep_module_16_tcn_6_norm_beta_to_fp16)[name = string("y_34_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_41_cast_fp16 = add(x = x_39_cast_fp16, y = y_34_cast_fp16)[name = string("x_41_cast_fp16")];
            bool var_1348 = const()[name = string("op_1348"), val = bool(true)];
            int32 var_1352 = const()[name = string("op_1352"), val = int32(448)];
            int32 var_1354 = const()[name = string("op_1354"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_145_cast_fp16 = add(x = x_41_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_145_cast_fp16")];
            tensor<int32, [1]> var_1364 = const()[name = string("op_1364"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1366 = const()[name = string("op_1366"), val = tensor<int32, [1]>([1])];
            string input0_43_pad_type_0 = const()[name = string("input0_43_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_43_pad_0 = const()[name = string("input0_43_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_17_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4553728))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4654144))))[name = string("sep_module_17_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_43_cast_fp16 = conv(dilations = var_1366, groups = var_1354, pad = input0_43_pad_0, pad_type = input0_43_pad_type_0, strides = var_1364, weight = sep_module_17_tcn_0_weight_to_fp16_palettized, x = input_145_cast_fp16)[name = string("input0_43_cast_fp16")];
            fp32 var_1370_alpha_1 = const()[name = string("op_1370_alpha_1"), val = fp32(0x1.d96ab6p-2)];
            tensor<fp16, [1, 448, 32]> var_1370_cast_fp16 = leaky_relu(alpha = fp16(0x1.d98p-2), x = input0_43_cast_fp16);
            tensor<int32, [1]> var_1374 = const()[name = string("op_1374"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_73_cast_fp16 = reduce_mean(axes = var_1374, keep_dims = var_1348, x = var_1370_cast_fp16)[name = string("mean_y_73_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1376_cast_fp16 = sub(x = var_1370_cast_fp16, y = mean_y_73_cast_fp16)[name = string("op_1376_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1377_cast_fp16 = square(x = var_1376_cast_fp16);
            tensor<int32, [1]> var_1378 = const()[name = string("op_1378"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1379_cast_fp16 = reduce_mean(axes = var_1378, keep_dims = var_1348, x = var_1377_cast_fp16)[name = string("op_1379_cast_fp16")];
            fp16 var_1380_to_fp16 = const()[name = string("op_1380_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1381_cast_fp16 = add(x = var_1379_cast_fp16, y = var_1380_to_fp16)[name = string("op_1381_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_73_cast_fp16 = sqrt(x = var_1381_cast_fp16)[name = string("std_y_73_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_17_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_17_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4654272)))];
            tensor<fp16, [1, 448, 32]> var_1384_cast_fp16 = mul(x = sep_module_17_tcn_2_norm_gamma_to_fp16, y = var_1376_cast_fp16)[name = string("op_1384_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1385_cast_fp16 = real_div(x = var_1384_cast_fp16, y = std_y_73_cast_fp16)[name = string("op_1385_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_17_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_17_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4655232)))];
            tensor<fp16, [1, 448, 32]> input_147_cast_fp16 = add(x = var_1385_cast_fp16, y = sep_module_17_tcn_2_norm_beta_to_fp16)[name = string("input_147_cast_fp16")];
            tensor<int32, [6]> input_149_pad_0 = const()[name = string("input_149_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 512, 0])];
            string input_149_mode_0 = const()[name = string("input_149_mode_0"), val = string("constant")];
            fp16 const_17_to_fp16 = const()[name = string("const_17_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 544]> input_147_cast_fp16_state_input = read_state(input = input_147_cast_fp16_state);
            tensor<fp16, [1, 448, 544]> input_149_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 512]), end = tensor<int32, [3]>([1, 448, 544]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_147_cast_fp16, x = input_147_cast_fp16_state_input);
            write_state(data = input_149_cast_fp16, input = input_147_cast_fp16_state);
            tensor<int32, [1]> var_1390 = const()[name = string("op_1390"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1392 = const()[name = string("op_1392"), val = tensor<int32, [1]>([256])];
            string input_151_pad_type_0 = const()[name = string("input_151_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_151_pad_0 = const()[name = string("input_151_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_17_tcn_4_weight_to_fp16 = const()[name = string("sep_module_17_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4656192)))];
            tensor<fp16, [1, 448, 32]> input_151_cast_fp16 = conv(dilations = var_1392, groups = var_1352, pad = input_151_pad_0, pad_type = input_151_pad_type_0, strides = var_1390, weight = sep_module_17_tcn_4_weight_to_fp16, x = input_149_cast_fp16)[name = string("input_151_cast_fp16")];
            fp32 var_1396_alpha_1 = const()[name = string("op_1396_alpha_1"), val = fp32(0x1.45df9ep-3)];
            tensor<fp16, [1, 448, 32]> var_1396_cast_fp16 = leaky_relu(alpha = fp16(0x1.45cp-3), x = input_151_cast_fp16);
            tensor<int32, [1]> var_1400 = const()[name = string("op_1400"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_75_cast_fp16 = reduce_mean(axes = var_1400, keep_dims = var_1348, x = var_1396_cast_fp16)[name = string("mean_y_75_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1402_cast_fp16 = sub(x = var_1396_cast_fp16, y = mean_y_75_cast_fp16)[name = string("op_1402_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1403_cast_fp16 = square(x = var_1402_cast_fp16);
            tensor<int32, [1]> var_1404 = const()[name = string("op_1404"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1405_cast_fp16 = reduce_mean(axes = var_1404, keep_dims = var_1348, x = var_1403_cast_fp16)[name = string("op_1405_cast_fp16")];
            fp16 var_1406_to_fp16 = const()[name = string("op_1406_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1407_cast_fp16 = add(x = var_1405_cast_fp16, y = var_1406_to_fp16)[name = string("op_1407_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_75_cast_fp16 = sqrt(x = var_1407_cast_fp16)[name = string("std_y_75_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_17_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_17_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4658944)))];
            tensor<fp16, [1, 448, 32]> var_1410_cast_fp16 = mul(x = sep_module_17_tcn_6_norm_gamma_to_fp16, y = var_1402_cast_fp16)[name = string("op_1410_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1411_cast_fp16 = real_div(x = var_1410_cast_fp16, y = std_y_75_cast_fp16)[name = string("op_1411_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_17_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_17_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4659904)))];
            tensor<fp16, [1, 448, 32]> y_36_cast_fp16 = add(x = var_1411_cast_fp16, y = sep_module_17_tcn_6_norm_beta_to_fp16)[name = string("y_36_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_43_cast_fp16 = add(x = x_41_cast_fp16, y = y_36_cast_fp16)[name = string("x_43_cast_fp16")];
            bool var_1417 = const()[name = string("op_1417"), val = bool(true)];
            int32 var_1421 = const()[name = string("op_1421"), val = int32(448)];
            int32 var_1422 = const()[name = string("op_1422"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_153_cast_fp16 = add(x = x_43_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_153_cast_fp16")];
            tensor<int32, [1]> var_1432 = const()[name = string("op_1432"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1434 = const()[name = string("op_1434"), val = tensor<int32, [1]>([1])];
            string input0_45_pad_type_0 = const()[name = string("input0_45_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_45_pad_0 = const()[name = string("input0_45_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_18_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4660864))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4761280))))[name = string("sep_module_18_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_45_cast_fp16 = conv(dilations = var_1434, groups = var_1422, pad = input0_45_pad_0, pad_type = input0_45_pad_type_0, strides = var_1432, weight = sep_module_18_tcn_0_weight_to_fp16_palettized, x = input_153_cast_fp16)[name = string("input0_45_cast_fp16")];
            fp32 var_1438_alpha_1 = const()[name = string("op_1438_alpha_1"), val = fp32(-0x1.6df066p-2)];
            tensor<fp16, [1, 448, 32]> var_1438_cast_fp16 = leaky_relu(alpha = fp16(-0x1.6ep-2), x = input0_45_cast_fp16);
            tensor<int32, [1]> var_1442 = const()[name = string("op_1442"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_77_cast_fp16 = reduce_mean(axes = var_1442, keep_dims = var_1417, x = var_1438_cast_fp16)[name = string("mean_y_77_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1444_cast_fp16 = sub(x = var_1438_cast_fp16, y = mean_y_77_cast_fp16)[name = string("op_1444_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1445_cast_fp16 = square(x = var_1444_cast_fp16);
            tensor<int32, [1]> var_1446 = const()[name = string("op_1446"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1447_cast_fp16 = reduce_mean(axes = var_1446, keep_dims = var_1417, x = var_1445_cast_fp16)[name = string("op_1447_cast_fp16")];
            fp16 var_1448_to_fp16 = const()[name = string("op_1448_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1449_cast_fp16 = add(x = var_1447_cast_fp16, y = var_1448_to_fp16)[name = string("op_1449_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_77_cast_fp16 = sqrt(x = var_1449_cast_fp16)[name = string("std_y_77_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_18_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_18_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4761408)))];
            tensor<fp16, [1, 448, 32]> var_1452_cast_fp16 = mul(x = sep_module_18_tcn_2_norm_gamma_to_fp16, y = var_1444_cast_fp16)[name = string("op_1452_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1453_cast_fp16 = real_div(x = var_1452_cast_fp16, y = std_y_77_cast_fp16)[name = string("op_1453_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_18_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_18_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4762368)))];
            tensor<fp16, [1, 448, 32]> input_155_cast_fp16 = add(x = var_1453_cast_fp16, y = sep_module_18_tcn_2_norm_beta_to_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, 0])];
            string input_157_mode_0 = const()[name = string("input_157_mode_0"), val = string("constant")];
            fp16 const_18_to_fp16 = const()[name = string("const_18_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 34]> input_155_cast_fp16_state_input = read_state(input = input_155_cast_fp16_state);
            tensor<fp16, [1, 448, 34]> input_157_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 2]), end = tensor<int32, [3]>([1, 448, 34]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_155_cast_fp16, x = input_155_cast_fp16_state_input);
            write_state(data = input_157_cast_fp16, input = input_155_cast_fp16_state);
            tensor<int32, [1]> var_1458 = const()[name = string("op_1458"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1460 = const()[name = string("op_1460"), val = tensor<int32, [1]>([1])];
            string input_159_pad_type_0 = const()[name = string("input_159_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_159_pad_0 = const()[name = string("input_159_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_18_tcn_4_weight_to_fp16 = const()[name = string("sep_module_18_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4763328)))];
            tensor<fp16, [1, 448, 32]> input_159_cast_fp16 = conv(dilations = var_1460, groups = var_1421, pad = input_159_pad_0, pad_type = input_159_pad_type_0, strides = var_1458, weight = sep_module_18_tcn_4_weight_to_fp16, x = input_157_cast_fp16)[name = string("input_159_cast_fp16")];
            fp32 var_1464_alpha_1 = const()[name = string("op_1464_alpha_1"), val = fp32(0x1.f5ac06p-1)];
            tensor<fp16, [1, 448, 32]> var_1464_cast_fp16 = leaky_relu(alpha = fp16(0x1.f5cp-1), x = input_159_cast_fp16);
            tensor<int32, [1]> var_1468 = const()[name = string("op_1468"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_79_cast_fp16 = reduce_mean(axes = var_1468, keep_dims = var_1417, x = var_1464_cast_fp16)[name = string("mean_y_79_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1470_cast_fp16 = sub(x = var_1464_cast_fp16, y = mean_y_79_cast_fp16)[name = string("op_1470_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1471_cast_fp16 = square(x = var_1470_cast_fp16);
            tensor<int32, [1]> var_1472 = const()[name = string("op_1472"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1473_cast_fp16 = reduce_mean(axes = var_1472, keep_dims = var_1417, x = var_1471_cast_fp16)[name = string("op_1473_cast_fp16")];
            fp16 var_1474_to_fp16 = const()[name = string("op_1474_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1475_cast_fp16 = add(x = var_1473_cast_fp16, y = var_1474_to_fp16)[name = string("op_1475_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_79_cast_fp16 = sqrt(x = var_1475_cast_fp16)[name = string("std_y_79_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_18_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_18_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4766080)))];
            tensor<fp16, [1, 448, 32]> var_1478_cast_fp16 = mul(x = sep_module_18_tcn_6_norm_gamma_to_fp16, y = var_1470_cast_fp16)[name = string("op_1478_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1479_cast_fp16 = real_div(x = var_1478_cast_fp16, y = std_y_79_cast_fp16)[name = string("op_1479_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_18_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_18_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4767040)))];
            tensor<fp16, [1, 448, 32]> y_38_cast_fp16 = add(x = var_1479_cast_fp16, y = sep_module_18_tcn_6_norm_beta_to_fp16)[name = string("y_38_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_45_cast_fp16 = add(x = x_43_cast_fp16, y = y_38_cast_fp16)[name = string("x_45_cast_fp16")];
            bool var_1485 = const()[name = string("op_1485"), val = bool(true)];
            int32 var_1489 = const()[name = string("op_1489"), val = int32(448)];
            int32 var_1491 = const()[name = string("op_1491"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_161_cast_fp16 = add(x = x_45_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_161_cast_fp16")];
            tensor<int32, [1]> var_1501 = const()[name = string("op_1501"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1503 = const()[name = string("op_1503"), val = tensor<int32, [1]>([1])];
            string input0_47_pad_type_0 = const()[name = string("input0_47_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_47_pad_0 = const()[name = string("input0_47_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_19_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4768000))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4868416))))[name = string("sep_module_19_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_47_cast_fp16 = conv(dilations = var_1503, groups = var_1491, pad = input0_47_pad_0, pad_type = input0_47_pad_type_0, strides = var_1501, weight = sep_module_19_tcn_0_weight_to_fp16_palettized, x = input_161_cast_fp16)[name = string("input0_47_cast_fp16")];
            fp32 var_1507_alpha_1 = const()[name = string("op_1507_alpha_1"), val = fp32(-0x1.32ae4ap-2)];
            tensor<fp16, [1, 448, 32]> var_1507_cast_fp16 = leaky_relu(alpha = fp16(-0x1.32cp-2), x = input0_47_cast_fp16);
            tensor<int32, [1]> var_1511 = const()[name = string("op_1511"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_81_cast_fp16 = reduce_mean(axes = var_1511, keep_dims = var_1485, x = var_1507_cast_fp16)[name = string("mean_y_81_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1513_cast_fp16 = sub(x = var_1507_cast_fp16, y = mean_y_81_cast_fp16)[name = string("op_1513_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1514_cast_fp16 = square(x = var_1513_cast_fp16);
            tensor<int32, [1]> var_1515 = const()[name = string("op_1515"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1516_cast_fp16 = reduce_mean(axes = var_1515, keep_dims = var_1485, x = var_1514_cast_fp16)[name = string("op_1516_cast_fp16")];
            fp16 var_1517_to_fp16 = const()[name = string("op_1517_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1518_cast_fp16 = add(x = var_1516_cast_fp16, y = var_1517_to_fp16)[name = string("op_1518_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_81_cast_fp16 = sqrt(x = var_1518_cast_fp16)[name = string("std_y_81_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_19_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_19_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4868544)))];
            tensor<fp16, [1, 448, 32]> var_1521_cast_fp16 = mul(x = sep_module_19_tcn_2_norm_gamma_to_fp16, y = var_1513_cast_fp16)[name = string("op_1521_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1522_cast_fp16 = real_div(x = var_1521_cast_fp16, y = std_y_81_cast_fp16)[name = string("op_1522_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_19_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_19_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4869504)))];
            tensor<fp16, [1, 448, 32]> input_163_cast_fp16 = add(x = var_1522_cast_fp16, y = sep_module_19_tcn_2_norm_beta_to_fp16)[name = string("input_163_cast_fp16")];
            tensor<int32, [6]> input_165_pad_0 = const()[name = string("input_165_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 0])];
            string input_165_mode_0 = const()[name = string("input_165_mode_0"), val = string("constant")];
            fp16 const_19_to_fp16 = const()[name = string("const_19_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 36]> input_163_cast_fp16_state_input = read_state(input = input_163_cast_fp16_state);
            tensor<fp16, [1, 448, 36]> input_165_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 4]), end = tensor<int32, [3]>([1, 448, 36]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_163_cast_fp16, x = input_163_cast_fp16_state_input);
            write_state(data = input_165_cast_fp16, input = input_163_cast_fp16_state);
            tensor<int32, [1]> var_1527 = const()[name = string("op_1527"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1529 = const()[name = string("op_1529"), val = tensor<int32, [1]>([2])];
            string input_167_pad_type_0 = const()[name = string("input_167_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_167_pad_0 = const()[name = string("input_167_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_19_tcn_4_weight_to_fp16 = const()[name = string("sep_module_19_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4870464)))];
            tensor<fp16, [1, 448, 32]> input_167_cast_fp16 = conv(dilations = var_1529, groups = var_1489, pad = input_167_pad_0, pad_type = input_167_pad_type_0, strides = var_1527, weight = sep_module_19_tcn_4_weight_to_fp16, x = input_165_cast_fp16)[name = string("input_167_cast_fp16")];
            fp32 var_1533_alpha_1 = const()[name = string("op_1533_alpha_1"), val = fp32(0x1.eb9e38p-1)];
            tensor<fp16, [1, 448, 32]> var_1533_cast_fp16 = leaky_relu(alpha = fp16(0x1.eb8p-1), x = input_167_cast_fp16);
            tensor<int32, [1]> var_1537 = const()[name = string("op_1537"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_83_cast_fp16 = reduce_mean(axes = var_1537, keep_dims = var_1485, x = var_1533_cast_fp16)[name = string("mean_y_83_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1539_cast_fp16 = sub(x = var_1533_cast_fp16, y = mean_y_83_cast_fp16)[name = string("op_1539_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1540_cast_fp16 = square(x = var_1539_cast_fp16);
            tensor<int32, [1]> var_1541 = const()[name = string("op_1541"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1542_cast_fp16 = reduce_mean(axes = var_1541, keep_dims = var_1485, x = var_1540_cast_fp16)[name = string("op_1542_cast_fp16")];
            fp16 var_1543_to_fp16 = const()[name = string("op_1543_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1544_cast_fp16 = add(x = var_1542_cast_fp16, y = var_1543_to_fp16)[name = string("op_1544_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_83_cast_fp16 = sqrt(x = var_1544_cast_fp16)[name = string("std_y_83_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_19_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_19_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4873216)))];
            tensor<fp16, [1, 448, 32]> var_1547_cast_fp16 = mul(x = sep_module_19_tcn_6_norm_gamma_to_fp16, y = var_1539_cast_fp16)[name = string("op_1547_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1548_cast_fp16 = real_div(x = var_1547_cast_fp16, y = std_y_83_cast_fp16)[name = string("op_1548_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_19_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_19_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4874176)))];
            tensor<fp16, [1, 448, 32]> y_40_cast_fp16 = add(x = var_1548_cast_fp16, y = sep_module_19_tcn_6_norm_beta_to_fp16)[name = string("y_40_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_47_cast_fp16 = add(x = x_45_cast_fp16, y = y_40_cast_fp16)[name = string("x_47_cast_fp16")];
            bool var_1554 = const()[name = string("op_1554"), val = bool(true)];
            int32 var_1558 = const()[name = string("op_1558"), val = int32(448)];
            int32 var_1560 = const()[name = string("op_1560"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_169_cast_fp16 = add(x = x_47_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_169_cast_fp16")];
            tensor<int32, [1]> var_1570 = const()[name = string("op_1570"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1572 = const()[name = string("op_1572"), val = tensor<int32, [1]>([1])];
            string input0_49_pad_type_0 = const()[name = string("input0_49_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_49_pad_0 = const()[name = string("input0_49_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_20_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4875136))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4975552))))[name = string("sep_module_20_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_49_cast_fp16 = conv(dilations = var_1572, groups = var_1560, pad = input0_49_pad_0, pad_type = input0_49_pad_type_0, strides = var_1570, weight = sep_module_20_tcn_0_weight_to_fp16_palettized, x = input_169_cast_fp16)[name = string("input0_49_cast_fp16")];
            fp32 var_1576_alpha_1 = const()[name = string("op_1576_alpha_1"), val = fp32(0x1.8a564ap-4)];
            tensor<fp16, [1, 448, 32]> var_1576_cast_fp16 = leaky_relu(alpha = fp16(0x1.8a4p-4), x = input0_49_cast_fp16);
            tensor<int32, [1]> var_1580 = const()[name = string("op_1580"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_85_cast_fp16 = reduce_mean(axes = var_1580, keep_dims = var_1554, x = var_1576_cast_fp16)[name = string("mean_y_85_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1582_cast_fp16 = sub(x = var_1576_cast_fp16, y = mean_y_85_cast_fp16)[name = string("op_1582_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1583_cast_fp16 = square(x = var_1582_cast_fp16);
            tensor<int32, [1]> var_1584 = const()[name = string("op_1584"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1585_cast_fp16 = reduce_mean(axes = var_1584, keep_dims = var_1554, x = var_1583_cast_fp16)[name = string("op_1585_cast_fp16")];
            fp16 var_1586_to_fp16 = const()[name = string("op_1586_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1587_cast_fp16 = add(x = var_1585_cast_fp16, y = var_1586_to_fp16)[name = string("op_1587_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_85_cast_fp16 = sqrt(x = var_1587_cast_fp16)[name = string("std_y_85_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_20_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_20_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4975680)))];
            tensor<fp16, [1, 448, 32]> var_1590_cast_fp16 = mul(x = sep_module_20_tcn_2_norm_gamma_to_fp16, y = var_1582_cast_fp16)[name = string("op_1590_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1591_cast_fp16 = real_div(x = var_1590_cast_fp16, y = std_y_85_cast_fp16)[name = string("op_1591_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_20_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_20_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4976640)))];
            tensor<fp16, [1, 448, 32]> input_171_cast_fp16 = add(x = var_1591_cast_fp16, y = sep_module_20_tcn_2_norm_beta_to_fp16)[name = string("input_171_cast_fp16")];
            tensor<int32, [6]> input_173_pad_0 = const()[name = string("input_173_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 8, 0])];
            string input_173_mode_0 = const()[name = string("input_173_mode_0"), val = string("constant")];
            fp16 const_20_to_fp16 = const()[name = string("const_20_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 40]> input_171_cast_fp16_state_input = read_state(input = input_171_cast_fp16_state);
            tensor<fp16, [1, 448, 40]> input_173_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 8]), end = tensor<int32, [3]>([1, 448, 40]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_171_cast_fp16, x = input_171_cast_fp16_state_input);
            write_state(data = input_173_cast_fp16, input = input_171_cast_fp16_state);
            tensor<int32, [1]> var_1596 = const()[name = string("op_1596"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1598 = const()[name = string("op_1598"), val = tensor<int32, [1]>([4])];
            string input_175_pad_type_0 = const()[name = string("input_175_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_175_pad_0 = const()[name = string("input_175_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_20_tcn_4_weight_to_fp16 = const()[name = string("sep_module_20_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4977600)))];
            tensor<fp16, [1, 448, 32]> input_175_cast_fp16 = conv(dilations = var_1598, groups = var_1558, pad = input_175_pad_0, pad_type = input_175_pad_type_0, strides = var_1596, weight = sep_module_20_tcn_4_weight_to_fp16, x = input_173_cast_fp16)[name = string("input_175_cast_fp16")];
            fp32 var_1602_alpha_1 = const()[name = string("op_1602_alpha_1"), val = fp32(0x1.d3bbd2p-1)];
            tensor<fp16, [1, 448, 32]> var_1602_cast_fp16 = leaky_relu(alpha = fp16(0x1.d3cp-1), x = input_175_cast_fp16);
            tensor<int32, [1]> var_1606 = const()[name = string("op_1606"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_87_cast_fp16 = reduce_mean(axes = var_1606, keep_dims = var_1554, x = var_1602_cast_fp16)[name = string("mean_y_87_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1608_cast_fp16 = sub(x = var_1602_cast_fp16, y = mean_y_87_cast_fp16)[name = string("op_1608_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1609_cast_fp16 = square(x = var_1608_cast_fp16);
            tensor<int32, [1]> var_1610 = const()[name = string("op_1610"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1611_cast_fp16 = reduce_mean(axes = var_1610, keep_dims = var_1554, x = var_1609_cast_fp16)[name = string("op_1611_cast_fp16")];
            fp16 var_1612_to_fp16 = const()[name = string("op_1612_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1613_cast_fp16 = add(x = var_1611_cast_fp16, y = var_1612_to_fp16)[name = string("op_1613_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_87_cast_fp16 = sqrt(x = var_1613_cast_fp16)[name = string("std_y_87_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_20_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_20_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4980352)))];
            tensor<fp16, [1, 448, 32]> var_1616_cast_fp16 = mul(x = sep_module_20_tcn_6_norm_gamma_to_fp16, y = var_1608_cast_fp16)[name = string("op_1616_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1617_cast_fp16 = real_div(x = var_1616_cast_fp16, y = std_y_87_cast_fp16)[name = string("op_1617_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_20_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_20_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4981312)))];
            tensor<fp16, [1, 448, 32]> y_42_cast_fp16 = add(x = var_1617_cast_fp16, y = sep_module_20_tcn_6_norm_beta_to_fp16)[name = string("y_42_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_49_cast_fp16 = add(x = x_47_cast_fp16, y = y_42_cast_fp16)[name = string("x_49_cast_fp16")];
            bool var_1623 = const()[name = string("op_1623"), val = bool(true)];
            int32 var_1627 = const()[name = string("op_1627"), val = int32(448)];
            int32 var_1629 = const()[name = string("op_1629"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_177_cast_fp16 = add(x = x_49_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_177_cast_fp16")];
            tensor<int32, [1]> var_1639 = const()[name = string("op_1639"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1641 = const()[name = string("op_1641"), val = tensor<int32, [1]>([1])];
            string input0_51_pad_type_0 = const()[name = string("input0_51_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_51_pad_0 = const()[name = string("input0_51_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_21_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(4982272))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5082688))))[name = string("sep_module_21_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_51_cast_fp16 = conv(dilations = var_1641, groups = var_1629, pad = input0_51_pad_0, pad_type = input0_51_pad_type_0, strides = var_1639, weight = sep_module_21_tcn_0_weight_to_fp16_palettized, x = input_177_cast_fp16)[name = string("input0_51_cast_fp16")];
            fp32 var_1645_alpha_1 = const()[name = string("op_1645_alpha_1"), val = fp32(0x1.a4e94ap-2)];
            tensor<fp16, [1, 448, 32]> var_1645_cast_fp16 = leaky_relu(alpha = fp16(0x1.a5p-2), x = input0_51_cast_fp16);
            tensor<int32, [1]> var_1649 = const()[name = string("op_1649"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_89_cast_fp16 = reduce_mean(axes = var_1649, keep_dims = var_1623, x = var_1645_cast_fp16)[name = string("mean_y_89_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1651_cast_fp16 = sub(x = var_1645_cast_fp16, y = mean_y_89_cast_fp16)[name = string("op_1651_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1652_cast_fp16 = square(x = var_1651_cast_fp16);
            tensor<int32, [1]> var_1653 = const()[name = string("op_1653"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1654_cast_fp16 = reduce_mean(axes = var_1653, keep_dims = var_1623, x = var_1652_cast_fp16)[name = string("op_1654_cast_fp16")];
            fp16 var_1655_to_fp16 = const()[name = string("op_1655_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1656_cast_fp16 = add(x = var_1654_cast_fp16, y = var_1655_to_fp16)[name = string("op_1656_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_89_cast_fp16 = sqrt(x = var_1656_cast_fp16)[name = string("std_y_89_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_21_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_21_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5082816)))];
            tensor<fp16, [1, 448, 32]> var_1659_cast_fp16 = mul(x = sep_module_21_tcn_2_norm_gamma_to_fp16, y = var_1651_cast_fp16)[name = string("op_1659_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1660_cast_fp16 = real_div(x = var_1659_cast_fp16, y = std_y_89_cast_fp16)[name = string("op_1660_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_21_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_21_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5083776)))];
            tensor<fp16, [1, 448, 32]> input_179_cast_fp16 = add(x = var_1660_cast_fp16, y = sep_module_21_tcn_2_norm_beta_to_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, 16, 0])];
            string input_181_mode_0 = const()[name = string("input_181_mode_0"), val = string("constant")];
            fp16 const_21_to_fp16 = const()[name = string("const_21_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 48]> input_179_cast_fp16_state_input = read_state(input = input_179_cast_fp16_state);
            tensor<fp16, [1, 448, 48]> input_181_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 16]), end = tensor<int32, [3]>([1, 448, 48]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_179_cast_fp16, x = input_179_cast_fp16_state_input);
            write_state(data = input_181_cast_fp16, input = input_179_cast_fp16_state);
            tensor<int32, [1]> var_1665 = const()[name = string("op_1665"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1667 = const()[name = string("op_1667"), val = tensor<int32, [1]>([8])];
            string input_183_pad_type_0 = const()[name = string("input_183_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_183_pad_0 = const()[name = string("input_183_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_21_tcn_4_weight_to_fp16 = const()[name = string("sep_module_21_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5084736)))];
            tensor<fp16, [1, 448, 32]> input_183_cast_fp16 = conv(dilations = var_1667, groups = var_1627, pad = input_183_pad_0, pad_type = input_183_pad_type_0, strides = var_1665, weight = sep_module_21_tcn_4_weight_to_fp16, x = input_181_cast_fp16)[name = string("input_183_cast_fp16")];
            fp32 var_1671_alpha_1 = const()[name = string("op_1671_alpha_1"), val = fp32(-0x1.dc9914p-2)];
            tensor<fp16, [1, 448, 32]> var_1671_cast_fp16 = leaky_relu(alpha = fp16(-0x1.dc8p-2), x = input_183_cast_fp16);
            tensor<int32, [1]> var_1675 = const()[name = string("op_1675"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_91_cast_fp16 = reduce_mean(axes = var_1675, keep_dims = var_1623, x = var_1671_cast_fp16)[name = string("mean_y_91_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1677_cast_fp16 = sub(x = var_1671_cast_fp16, y = mean_y_91_cast_fp16)[name = string("op_1677_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1678_cast_fp16 = square(x = var_1677_cast_fp16);
            tensor<int32, [1]> var_1679 = const()[name = string("op_1679"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1680_cast_fp16 = reduce_mean(axes = var_1679, keep_dims = var_1623, x = var_1678_cast_fp16)[name = string("op_1680_cast_fp16")];
            fp16 var_1681_to_fp16 = const()[name = string("op_1681_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1682_cast_fp16 = add(x = var_1680_cast_fp16, y = var_1681_to_fp16)[name = string("op_1682_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_91_cast_fp16 = sqrt(x = var_1682_cast_fp16)[name = string("std_y_91_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_21_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_21_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5087488)))];
            tensor<fp16, [1, 448, 32]> var_1685_cast_fp16 = mul(x = sep_module_21_tcn_6_norm_gamma_to_fp16, y = var_1677_cast_fp16)[name = string("op_1685_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1686_cast_fp16 = real_div(x = var_1685_cast_fp16, y = std_y_91_cast_fp16)[name = string("op_1686_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_21_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_21_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5088448)))];
            tensor<fp16, [1, 448, 32]> y_44_cast_fp16 = add(x = var_1686_cast_fp16, y = sep_module_21_tcn_6_norm_beta_to_fp16)[name = string("y_44_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_51_cast_fp16 = add(x = x_49_cast_fp16, y = y_44_cast_fp16)[name = string("x_51_cast_fp16")];
            bool var_1692 = const()[name = string("op_1692"), val = bool(true)];
            int32 var_1696 = const()[name = string("op_1696"), val = int32(448)];
            int32 var_1698 = const()[name = string("op_1698"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_185_cast_fp16 = add(x = x_51_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_185_cast_fp16")];
            tensor<int32, [1]> var_1708 = const()[name = string("op_1708"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1710 = const()[name = string("op_1710"), val = tensor<int32, [1]>([1])];
            string input0_53_pad_type_0 = const()[name = string("input0_53_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_53_pad_0 = const()[name = string("input0_53_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_22_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5089408))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5189824))))[name = string("sep_module_22_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_53_cast_fp16 = conv(dilations = var_1710, groups = var_1698, pad = input0_53_pad_0, pad_type = input0_53_pad_type_0, strides = var_1708, weight = sep_module_22_tcn_0_weight_to_fp16_palettized, x = input_185_cast_fp16)[name = string("input0_53_cast_fp16")];
            fp32 var_1714_alpha_1 = const()[name = string("op_1714_alpha_1"), val = fp32(0x1.de43a6p-5)];
            tensor<fp16, [1, 448, 32]> var_1714_cast_fp16 = leaky_relu(alpha = fp16(0x1.de4p-5), x = input0_53_cast_fp16);
            tensor<int32, [1]> var_1718 = const()[name = string("op_1718"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_93_cast_fp16 = reduce_mean(axes = var_1718, keep_dims = var_1692, x = var_1714_cast_fp16)[name = string("mean_y_93_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1720_cast_fp16 = sub(x = var_1714_cast_fp16, y = mean_y_93_cast_fp16)[name = string("op_1720_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1721_cast_fp16 = square(x = var_1720_cast_fp16);
            tensor<int32, [1]> var_1722 = const()[name = string("op_1722"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1723_cast_fp16 = reduce_mean(axes = var_1722, keep_dims = var_1692, x = var_1721_cast_fp16)[name = string("op_1723_cast_fp16")];
            fp16 var_1724_to_fp16 = const()[name = string("op_1724_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1725_cast_fp16 = add(x = var_1723_cast_fp16, y = var_1724_to_fp16)[name = string("op_1725_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_93_cast_fp16 = sqrt(x = var_1725_cast_fp16)[name = string("std_y_93_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_22_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_22_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5189952)))];
            tensor<fp16, [1, 448, 32]> var_1728_cast_fp16 = mul(x = sep_module_22_tcn_2_norm_gamma_to_fp16, y = var_1720_cast_fp16)[name = string("op_1728_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1729_cast_fp16 = real_div(x = var_1728_cast_fp16, y = std_y_93_cast_fp16)[name = string("op_1729_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_22_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_22_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5190912)))];
            tensor<fp16, [1, 448, 32]> input_187_cast_fp16 = add(x = var_1729_cast_fp16, y = sep_module_22_tcn_2_norm_beta_to_fp16)[name = string("input_187_cast_fp16")];
            tensor<int32, [6]> input_189_pad_0 = const()[name = string("input_189_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 32, 0])];
            string input_189_mode_0 = const()[name = string("input_189_mode_0"), val = string("constant")];
            fp16 const_22_to_fp16 = const()[name = string("const_22_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 64]> input_187_cast_fp16_state_input = read_state(input = input_187_cast_fp16_state);
            tensor<fp16, [1, 448, 64]> input_189_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 32]), end = tensor<int32, [3]>([1, 448, 64]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_187_cast_fp16, x = input_187_cast_fp16_state_input);
            write_state(data = input_189_cast_fp16, input = input_187_cast_fp16_state);
            tensor<int32, [1]> var_1734 = const()[name = string("op_1734"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1736 = const()[name = string("op_1736"), val = tensor<int32, [1]>([16])];
            string input_191_pad_type_0 = const()[name = string("input_191_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_191_pad_0 = const()[name = string("input_191_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_22_tcn_4_weight_to_fp16 = const()[name = string("sep_module_22_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5191872)))];
            tensor<fp16, [1, 448, 32]> input_191_cast_fp16 = conv(dilations = var_1736, groups = var_1696, pad = input_191_pad_0, pad_type = input_191_pad_type_0, strides = var_1734, weight = sep_module_22_tcn_4_weight_to_fp16, x = input_189_cast_fp16)[name = string("input_191_cast_fp16")];
            fp32 var_1740_alpha_1 = const()[name = string("op_1740_alpha_1"), val = fp32(0x1.b57f7ap-2)];
            tensor<fp16, [1, 448, 32]> var_1740_cast_fp16 = leaky_relu(alpha = fp16(0x1.b58p-2), x = input_191_cast_fp16);
            tensor<int32, [1]> var_1744 = const()[name = string("op_1744"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_95_cast_fp16 = reduce_mean(axes = var_1744, keep_dims = var_1692, x = var_1740_cast_fp16)[name = string("mean_y_95_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1746_cast_fp16 = sub(x = var_1740_cast_fp16, y = mean_y_95_cast_fp16)[name = string("op_1746_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1747_cast_fp16 = square(x = var_1746_cast_fp16);
            tensor<int32, [1]> var_1748 = const()[name = string("op_1748"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1749_cast_fp16 = reduce_mean(axes = var_1748, keep_dims = var_1692, x = var_1747_cast_fp16)[name = string("op_1749_cast_fp16")];
            fp16 var_1750_to_fp16 = const()[name = string("op_1750_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1751_cast_fp16 = add(x = var_1749_cast_fp16, y = var_1750_to_fp16)[name = string("op_1751_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_95_cast_fp16 = sqrt(x = var_1751_cast_fp16)[name = string("std_y_95_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_22_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_22_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5194624)))];
            tensor<fp16, [1, 448, 32]> var_1754_cast_fp16 = mul(x = sep_module_22_tcn_6_norm_gamma_to_fp16, y = var_1746_cast_fp16)[name = string("op_1754_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1755_cast_fp16 = real_div(x = var_1754_cast_fp16, y = std_y_95_cast_fp16)[name = string("op_1755_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_22_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_22_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5195584)))];
            tensor<fp16, [1, 448, 32]> y_46_cast_fp16 = add(x = var_1755_cast_fp16, y = sep_module_22_tcn_6_norm_beta_to_fp16)[name = string("y_46_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_53_cast_fp16 = add(x = x_51_cast_fp16, y = y_46_cast_fp16)[name = string("x_53_cast_fp16")];
            bool var_1761 = const()[name = string("op_1761"), val = bool(true)];
            int32 var_1765 = const()[name = string("op_1765"), val = int32(448)];
            int32 var_1767 = const()[name = string("op_1767"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_193_cast_fp16 = add(x = x_53_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_193_cast_fp16")];
            tensor<int32, [1]> var_1777 = const()[name = string("op_1777"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1779 = const()[name = string("op_1779"), val = tensor<int32, [1]>([1])];
            string input0_55_pad_type_0 = const()[name = string("input0_55_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_55_pad_0 = const()[name = string("input0_55_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_23_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5196544))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5296960))))[name = string("sep_module_23_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_55_cast_fp16 = conv(dilations = var_1779, groups = var_1767, pad = input0_55_pad_0, pad_type = input0_55_pad_type_0, strides = var_1777, weight = sep_module_23_tcn_0_weight_to_fp16_palettized, x = input_193_cast_fp16)[name = string("input0_55_cast_fp16")];
            fp32 var_1783_alpha_1 = const()[name = string("op_1783_alpha_1"), val = fp32(0x1.b88384p-5)];
            tensor<fp16, [1, 448, 32]> var_1783_cast_fp16 = leaky_relu(alpha = fp16(0x1.b88p-5), x = input0_55_cast_fp16);
            tensor<int32, [1]> var_1787 = const()[name = string("op_1787"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_97_cast_fp16 = reduce_mean(axes = var_1787, keep_dims = var_1761, x = var_1783_cast_fp16)[name = string("mean_y_97_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1789_cast_fp16 = sub(x = var_1783_cast_fp16, y = mean_y_97_cast_fp16)[name = string("op_1789_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1790_cast_fp16 = square(x = var_1789_cast_fp16);
            tensor<int32, [1]> var_1791 = const()[name = string("op_1791"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1792_cast_fp16 = reduce_mean(axes = var_1791, keep_dims = var_1761, x = var_1790_cast_fp16)[name = string("op_1792_cast_fp16")];
            fp16 var_1793_to_fp16 = const()[name = string("op_1793_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1794_cast_fp16 = add(x = var_1792_cast_fp16, y = var_1793_to_fp16)[name = string("op_1794_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_97_cast_fp16 = sqrt(x = var_1794_cast_fp16)[name = string("std_y_97_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_23_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_23_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5297088)))];
            tensor<fp16, [1, 448, 32]> var_1797_cast_fp16 = mul(x = sep_module_23_tcn_2_norm_gamma_to_fp16, y = var_1789_cast_fp16)[name = string("op_1797_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1798_cast_fp16 = real_div(x = var_1797_cast_fp16, y = std_y_97_cast_fp16)[name = string("op_1798_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_23_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_23_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5298048)))];
            tensor<fp16, [1, 448, 32]> input_195_cast_fp16 = add(x = var_1798_cast_fp16, y = sep_module_23_tcn_2_norm_beta_to_fp16)[name = string("input_195_cast_fp16")];
            tensor<int32, [6]> input_197_pad_0 = const()[name = string("input_197_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 64, 0])];
            string input_197_mode_0 = const()[name = string("input_197_mode_0"), val = string("constant")];
            fp16 const_23_to_fp16 = const()[name = string("const_23_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 96]> input_195_cast_fp16_state_input = read_state(input = input_195_cast_fp16_state);
            tensor<fp16, [1, 448, 96]> input_197_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 64]), end = tensor<int32, [3]>([1, 448, 96]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_195_cast_fp16, x = input_195_cast_fp16_state_input);
            write_state(data = input_197_cast_fp16, input = input_195_cast_fp16_state);
            tensor<int32, [1]> var_1803 = const()[name = string("op_1803"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1805 = const()[name = string("op_1805"), val = tensor<int32, [1]>([32])];
            string input_199_pad_type_0 = const()[name = string("input_199_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_199_pad_0 = const()[name = string("input_199_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_23_tcn_4_weight_to_fp16 = const()[name = string("sep_module_23_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5299008)))];
            tensor<fp16, [1, 448, 32]> input_199_cast_fp16 = conv(dilations = var_1805, groups = var_1765, pad = input_199_pad_0, pad_type = input_199_pad_type_0, strides = var_1803, weight = sep_module_23_tcn_4_weight_to_fp16, x = input_197_cast_fp16)[name = string("input_199_cast_fp16")];
            fp32 var_1809_alpha_1 = const()[name = string("op_1809_alpha_1"), val = fp32(0x1.38845ep-3)];
            tensor<fp16, [1, 448, 32]> var_1809_cast_fp16 = leaky_relu(alpha = fp16(0x1.388p-3), x = input_199_cast_fp16);
            tensor<int32, [1]> var_1813 = const()[name = string("op_1813"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_99_cast_fp16 = reduce_mean(axes = var_1813, keep_dims = var_1761, x = var_1809_cast_fp16)[name = string("mean_y_99_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1815_cast_fp16 = sub(x = var_1809_cast_fp16, y = mean_y_99_cast_fp16)[name = string("op_1815_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1816_cast_fp16 = square(x = var_1815_cast_fp16);
            tensor<int32, [1]> var_1817 = const()[name = string("op_1817"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1818_cast_fp16 = reduce_mean(axes = var_1817, keep_dims = var_1761, x = var_1816_cast_fp16)[name = string("op_1818_cast_fp16")];
            fp16 var_1819_to_fp16 = const()[name = string("op_1819_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1820_cast_fp16 = add(x = var_1818_cast_fp16, y = var_1819_to_fp16)[name = string("op_1820_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_99_cast_fp16 = sqrt(x = var_1820_cast_fp16)[name = string("std_y_99_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_23_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_23_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5301760)))];
            tensor<fp16, [1, 448, 32]> var_1823_cast_fp16 = mul(x = sep_module_23_tcn_6_norm_gamma_to_fp16, y = var_1815_cast_fp16)[name = string("op_1823_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1824_cast_fp16 = real_div(x = var_1823_cast_fp16, y = std_y_99_cast_fp16)[name = string("op_1824_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_23_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_23_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5302720)))];
            tensor<fp16, [1, 448, 32]> y_48_cast_fp16 = add(x = var_1824_cast_fp16, y = sep_module_23_tcn_6_norm_beta_to_fp16)[name = string("y_48_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_55_cast_fp16 = add(x = x_53_cast_fp16, y = y_48_cast_fp16)[name = string("x_55_cast_fp16")];
            bool var_1830 = const()[name = string("op_1830"), val = bool(true)];
            int32 var_1834 = const()[name = string("op_1834"), val = int32(448)];
            int32 var_1836 = const()[name = string("op_1836"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_201_cast_fp16 = add(x = x_55_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_201_cast_fp16")];
            tensor<int32, [1]> var_1846 = const()[name = string("op_1846"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1848 = const()[name = string("op_1848"), val = tensor<int32, [1]>([1])];
            string input0_57_pad_type_0 = const()[name = string("input0_57_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_57_pad_0 = const()[name = string("input0_57_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_24_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5303680))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5404096))))[name = string("sep_module_24_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_57_cast_fp16 = conv(dilations = var_1848, groups = var_1836, pad = input0_57_pad_0, pad_type = input0_57_pad_type_0, strides = var_1846, weight = sep_module_24_tcn_0_weight_to_fp16_palettized, x = input_201_cast_fp16)[name = string("input0_57_cast_fp16")];
            fp32 var_1852_alpha_1 = const()[name = string("op_1852_alpha_1"), val = fp32(0x1.ee37e6p-2)];
            tensor<fp16, [1, 448, 32]> var_1852_cast_fp16 = leaky_relu(alpha = fp16(0x1.ee4p-2), x = input0_57_cast_fp16);
            tensor<int32, [1]> var_1856 = const()[name = string("op_1856"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_101_cast_fp16 = reduce_mean(axes = var_1856, keep_dims = var_1830, x = var_1852_cast_fp16)[name = string("mean_y_101_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1858_cast_fp16 = sub(x = var_1852_cast_fp16, y = mean_y_101_cast_fp16)[name = string("op_1858_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1859_cast_fp16 = square(x = var_1858_cast_fp16);
            tensor<int32, [1]> var_1860 = const()[name = string("op_1860"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1861_cast_fp16 = reduce_mean(axes = var_1860, keep_dims = var_1830, x = var_1859_cast_fp16)[name = string("op_1861_cast_fp16")];
            fp16 var_1862_to_fp16 = const()[name = string("op_1862_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1863_cast_fp16 = add(x = var_1861_cast_fp16, y = var_1862_to_fp16)[name = string("op_1863_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_101_cast_fp16 = sqrt(x = var_1863_cast_fp16)[name = string("std_y_101_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_24_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_24_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5404224)))];
            tensor<fp16, [1, 448, 32]> var_1866_cast_fp16 = mul(x = sep_module_24_tcn_2_norm_gamma_to_fp16, y = var_1858_cast_fp16)[name = string("op_1866_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1867_cast_fp16 = real_div(x = var_1866_cast_fp16, y = std_y_101_cast_fp16)[name = string("op_1867_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_24_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_24_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5405184)))];
            tensor<fp16, [1, 448, 32]> input_203_cast_fp16 = add(x = var_1867_cast_fp16, y = sep_module_24_tcn_2_norm_beta_to_fp16)[name = string("input_203_cast_fp16")];
            tensor<int32, [6]> input_205_pad_0 = const()[name = string("input_205_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 128, 0])];
            string input_205_mode_0 = const()[name = string("input_205_mode_0"), val = string("constant")];
            fp16 const_24_to_fp16 = const()[name = string("const_24_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 160]> input_203_cast_fp16_state_input = read_state(input = input_203_cast_fp16_state);
            tensor<fp16, [1, 448, 160]> input_205_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 128]), end = tensor<int32, [3]>([1, 448, 160]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_203_cast_fp16, x = input_203_cast_fp16_state_input);
            write_state(data = input_205_cast_fp16, input = input_203_cast_fp16_state);
            tensor<int32, [1]> var_1872 = const()[name = string("op_1872"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1874 = const()[name = string("op_1874"), val = tensor<int32, [1]>([64])];
            string input_207_pad_type_0 = const()[name = string("input_207_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_207_pad_0 = const()[name = string("input_207_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_24_tcn_4_weight_to_fp16 = const()[name = string("sep_module_24_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5406144)))];
            tensor<fp16, [1, 448, 32]> input_207_cast_fp16 = conv(dilations = var_1874, groups = var_1834, pad = input_207_pad_0, pad_type = input_207_pad_type_0, strides = var_1872, weight = sep_module_24_tcn_4_weight_to_fp16, x = input_205_cast_fp16)[name = string("input_207_cast_fp16")];
            fp32 var_1878_alpha_1 = const()[name = string("op_1878_alpha_1"), val = fp32(0x1.843ae4p-3)];
            tensor<fp16, [1, 448, 32]> var_1878_cast_fp16 = leaky_relu(alpha = fp16(0x1.844p-3), x = input_207_cast_fp16);
            tensor<int32, [1]> var_1882 = const()[name = string("op_1882"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_103_cast_fp16 = reduce_mean(axes = var_1882, keep_dims = var_1830, x = var_1878_cast_fp16)[name = string("mean_y_103_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1884_cast_fp16 = sub(x = var_1878_cast_fp16, y = mean_y_103_cast_fp16)[name = string("op_1884_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1885_cast_fp16 = square(x = var_1884_cast_fp16);
            tensor<int32, [1]> var_1886 = const()[name = string("op_1886"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1887_cast_fp16 = reduce_mean(axes = var_1886, keep_dims = var_1830, x = var_1885_cast_fp16)[name = string("op_1887_cast_fp16")];
            fp16 var_1888_to_fp16 = const()[name = string("op_1888_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1889_cast_fp16 = add(x = var_1887_cast_fp16, y = var_1888_to_fp16)[name = string("op_1889_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_103_cast_fp16 = sqrt(x = var_1889_cast_fp16)[name = string("std_y_103_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_24_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_24_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5408896)))];
            tensor<fp16, [1, 448, 32]> var_1892_cast_fp16 = mul(x = sep_module_24_tcn_6_norm_gamma_to_fp16, y = var_1884_cast_fp16)[name = string("op_1892_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1893_cast_fp16 = real_div(x = var_1892_cast_fp16, y = std_y_103_cast_fp16)[name = string("op_1893_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_24_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_24_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5409856)))];
            tensor<fp16, [1, 448, 32]> y_50_cast_fp16 = add(x = var_1893_cast_fp16, y = sep_module_24_tcn_6_norm_beta_to_fp16)[name = string("y_50_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_57_cast_fp16 = add(x = x_55_cast_fp16, y = y_50_cast_fp16)[name = string("x_57_cast_fp16")];
            bool var_1899 = const()[name = string("op_1899"), val = bool(true)];
            int32 var_1903 = const()[name = string("op_1903"), val = int32(448)];
            int32 var_1905 = const()[name = string("op_1905"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_209_cast_fp16 = add(x = x_57_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_209_cast_fp16")];
            tensor<int32, [1]> var_1915 = const()[name = string("op_1915"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1917 = const()[name = string("op_1917"), val = tensor<int32, [1]>([1])];
            string input0_59_pad_type_0 = const()[name = string("input0_59_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_59_pad_0 = const()[name = string("input0_59_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_25_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5410816))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5511232))))[name = string("sep_module_25_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_59_cast_fp16 = conv(dilations = var_1917, groups = var_1905, pad = input0_59_pad_0, pad_type = input0_59_pad_type_0, strides = var_1915, weight = sep_module_25_tcn_0_weight_to_fp16_palettized, x = input_209_cast_fp16)[name = string("input0_59_cast_fp16")];
            fp32 var_1921_alpha_1 = const()[name = string("op_1921_alpha_1"), val = fp32(0x1.dc0f66p-2)];
            tensor<fp16, [1, 448, 32]> var_1921_cast_fp16 = leaky_relu(alpha = fp16(0x1.dcp-2), x = input0_59_cast_fp16);
            tensor<int32, [1]> var_1925 = const()[name = string("op_1925"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_105_cast_fp16 = reduce_mean(axes = var_1925, keep_dims = var_1899, x = var_1921_cast_fp16)[name = string("mean_y_105_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1927_cast_fp16 = sub(x = var_1921_cast_fp16, y = mean_y_105_cast_fp16)[name = string("op_1927_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1928_cast_fp16 = square(x = var_1927_cast_fp16);
            tensor<int32, [1]> var_1929 = const()[name = string("op_1929"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1930_cast_fp16 = reduce_mean(axes = var_1929, keep_dims = var_1899, x = var_1928_cast_fp16)[name = string("op_1930_cast_fp16")];
            fp16 var_1931_to_fp16 = const()[name = string("op_1931_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1932_cast_fp16 = add(x = var_1930_cast_fp16, y = var_1931_to_fp16)[name = string("op_1932_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_105_cast_fp16 = sqrt(x = var_1932_cast_fp16)[name = string("std_y_105_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_25_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_25_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5511360)))];
            tensor<fp16, [1, 448, 32]> var_1935_cast_fp16 = mul(x = sep_module_25_tcn_2_norm_gamma_to_fp16, y = var_1927_cast_fp16)[name = string("op_1935_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1936_cast_fp16 = real_div(x = var_1935_cast_fp16, y = std_y_105_cast_fp16)[name = string("op_1936_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_25_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_25_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5512320)))];
            tensor<fp16, [1, 448, 32]> input_211_cast_fp16 = add(x = var_1936_cast_fp16, y = sep_module_25_tcn_2_norm_beta_to_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, 256, 0])];
            string input_213_mode_0 = const()[name = string("input_213_mode_0"), val = string("constant")];
            fp16 const_25_to_fp16 = const()[name = string("const_25_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 288]> input_211_cast_fp16_state_input = read_state(input = input_211_cast_fp16_state);
            tensor<fp16, [1, 448, 288]> input_213_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 256]), end = tensor<int32, [3]>([1, 448, 288]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_211_cast_fp16, x = input_211_cast_fp16_state_input);
            write_state(data = input_213_cast_fp16, input = input_211_cast_fp16_state);
            tensor<int32, [1]> var_1941 = const()[name = string("op_1941"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1943 = const()[name = string("op_1943"), val = tensor<int32, [1]>([128])];
            string input_215_pad_type_0 = const()[name = string("input_215_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_215_pad_0 = const()[name = string("input_215_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_25_tcn_4_weight_to_fp16 = const()[name = string("sep_module_25_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5513280)))];
            tensor<fp16, [1, 448, 32]> input_215_cast_fp16 = conv(dilations = var_1943, groups = var_1903, pad = input_215_pad_0, pad_type = input_215_pad_type_0, strides = var_1941, weight = sep_module_25_tcn_4_weight_to_fp16, x = input_213_cast_fp16)[name = string("input_215_cast_fp16")];
            fp32 var_1947_alpha_1 = const()[name = string("op_1947_alpha_1"), val = fp32(-0x1.92f154p-7)];
            tensor<fp16, [1, 448, 32]> var_1947_cast_fp16 = leaky_relu(alpha = fp16(-0x1.93p-7), x = input_215_cast_fp16);
            tensor<int32, [1]> var_1951 = const()[name = string("op_1951"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_107_cast_fp16 = reduce_mean(axes = var_1951, keep_dims = var_1899, x = var_1947_cast_fp16)[name = string("mean_y_107_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1953_cast_fp16 = sub(x = var_1947_cast_fp16, y = mean_y_107_cast_fp16)[name = string("op_1953_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1954_cast_fp16 = square(x = var_1953_cast_fp16);
            tensor<int32, [1]> var_1955 = const()[name = string("op_1955"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1956_cast_fp16 = reduce_mean(axes = var_1955, keep_dims = var_1899, x = var_1954_cast_fp16)[name = string("op_1956_cast_fp16")];
            fp16 var_1957_to_fp16 = const()[name = string("op_1957_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_1958_cast_fp16 = add(x = var_1956_cast_fp16, y = var_1957_to_fp16)[name = string("op_1958_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_107_cast_fp16 = sqrt(x = var_1958_cast_fp16)[name = string("std_y_107_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_25_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_25_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5516032)))];
            tensor<fp16, [1, 448, 32]> var_1961_cast_fp16 = mul(x = sep_module_25_tcn_6_norm_gamma_to_fp16, y = var_1953_cast_fp16)[name = string("op_1961_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1962_cast_fp16 = real_div(x = var_1961_cast_fp16, y = std_y_107_cast_fp16)[name = string("op_1962_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_25_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_25_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5516992)))];
            tensor<fp16, [1, 448, 32]> y_52_cast_fp16 = add(x = var_1962_cast_fp16, y = sep_module_25_tcn_6_norm_beta_to_fp16)[name = string("y_52_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_59_cast_fp16 = add(x = x_57_cast_fp16, y = y_52_cast_fp16)[name = string("x_59_cast_fp16")];
            bool var_1968 = const()[name = string("op_1968"), val = bool(true)];
            int32 var_1972 = const()[name = string("op_1972"), val = int32(448)];
            int32 var_1974 = const()[name = string("op_1974"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_217_cast_fp16 = add(x = x_59_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_217_cast_fp16")];
            tensor<int32, [1]> var_1984 = const()[name = string("op_1984"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_1986 = const()[name = string("op_1986"), val = tensor<int32, [1]>([1])];
            string input0_61_pad_type_0 = const()[name = string("input0_61_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_61_pad_0 = const()[name = string("input0_61_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_26_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5517952))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5618368))))[name = string("sep_module_26_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_61_cast_fp16 = conv(dilations = var_1986, groups = var_1974, pad = input0_61_pad_0, pad_type = input0_61_pad_type_0, strides = var_1984, weight = sep_module_26_tcn_0_weight_to_fp16_palettized, x = input_217_cast_fp16)[name = string("input0_61_cast_fp16")];
            fp32 var_1990_alpha_1 = const()[name = string("op_1990_alpha_1"), val = fp32(0x1.9fda08p-2)];
            tensor<fp16, [1, 448, 32]> var_1990_cast_fp16 = leaky_relu(alpha = fp16(0x1.9fcp-2), x = input0_61_cast_fp16);
            tensor<int32, [1]> var_1994 = const()[name = string("op_1994"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_109_cast_fp16 = reduce_mean(axes = var_1994, keep_dims = var_1968, x = var_1990_cast_fp16)[name = string("mean_y_109_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1996_cast_fp16 = sub(x = var_1990_cast_fp16, y = mean_y_109_cast_fp16)[name = string("op_1996_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_1997_cast_fp16 = square(x = var_1996_cast_fp16);
            tensor<int32, [1]> var_1998 = const()[name = string("op_1998"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_1999_cast_fp16 = reduce_mean(axes = var_1998, keep_dims = var_1968, x = var_1997_cast_fp16)[name = string("op_1999_cast_fp16")];
            fp16 var_2000_to_fp16 = const()[name = string("op_2000_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_2001_cast_fp16 = add(x = var_1999_cast_fp16, y = var_2000_to_fp16)[name = string("op_2001_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_109_cast_fp16 = sqrt(x = var_2001_cast_fp16)[name = string("std_y_109_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_26_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_26_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5618496)))];
            tensor<fp16, [1, 448, 32]> var_2004_cast_fp16 = mul(x = sep_module_26_tcn_2_norm_gamma_to_fp16, y = var_1996_cast_fp16)[name = string("op_2004_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2005_cast_fp16 = real_div(x = var_2004_cast_fp16, y = std_y_109_cast_fp16)[name = string("op_2005_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_26_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_26_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5619456)))];
            tensor<fp16, [1, 448, 32]> input_219_cast_fp16 = add(x = var_2005_cast_fp16, y = sep_module_26_tcn_2_norm_beta_to_fp16)[name = string("input_219_cast_fp16")];
            tensor<int32, [6]> input_221_pad_0 = const()[name = string("input_221_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 512, 0])];
            string input_221_mode_0 = const()[name = string("input_221_mode_0"), val = string("constant")];
            fp16 const_26_to_fp16 = const()[name = string("const_26_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 544]> input_219_cast_fp16_state_input = read_state(input = input_219_cast_fp16_state);
            tensor<fp16, [1, 448, 544]> input_221_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 512]), end = tensor<int32, [3]>([1, 448, 544]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_219_cast_fp16, x = input_219_cast_fp16_state_input);
            write_state(data = input_221_cast_fp16, input = input_219_cast_fp16_state);
            tensor<int32, [1]> var_2010 = const()[name = string("op_2010"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_2012 = const()[name = string("op_2012"), val = tensor<int32, [1]>([256])];
            string input_223_pad_type_0 = const()[name = string("input_223_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_223_pad_0 = const()[name = string("input_223_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_26_tcn_4_weight_to_fp16 = const()[name = string("sep_module_26_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5620416)))];
            tensor<fp16, [1, 448, 32]> input_223_cast_fp16 = conv(dilations = var_2012, groups = var_1972, pad = input_223_pad_0, pad_type = input_223_pad_type_0, strides = var_2010, weight = sep_module_26_tcn_4_weight_to_fp16, x = input_221_cast_fp16)[name = string("input_223_cast_fp16")];
            fp32 var_2016_alpha_1 = const()[name = string("op_2016_alpha_1"), val = fp32(0x1.9f6d02p-2)];
            tensor<fp16, [1, 448, 32]> var_2016_cast_fp16 = leaky_relu(alpha = fp16(0x1.9f8p-2), x = input_223_cast_fp16);
            tensor<int32, [1]> var_2020 = const()[name = string("op_2020"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_111_cast_fp16 = reduce_mean(axes = var_2020, keep_dims = var_1968, x = var_2016_cast_fp16)[name = string("mean_y_111_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2022_cast_fp16 = sub(x = var_2016_cast_fp16, y = mean_y_111_cast_fp16)[name = string("op_2022_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2023_cast_fp16 = square(x = var_2022_cast_fp16);
            tensor<int32, [1]> var_2024 = const()[name = string("op_2024"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_2025_cast_fp16 = reduce_mean(axes = var_2024, keep_dims = var_1968, x = var_2023_cast_fp16)[name = string("op_2025_cast_fp16")];
            fp16 var_2026_to_fp16 = const()[name = string("op_2026_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_2027_cast_fp16 = add(x = var_2025_cast_fp16, y = var_2026_to_fp16)[name = string("op_2027_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_111_cast_fp16 = sqrt(x = var_2027_cast_fp16)[name = string("std_y_111_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_26_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_26_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5623168)))];
            tensor<fp16, [1, 448, 32]> var_2030_cast_fp16 = mul(x = sep_module_26_tcn_6_norm_gamma_to_fp16, y = var_2022_cast_fp16)[name = string("op_2030_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2031_cast_fp16 = real_div(x = var_2030_cast_fp16, y = std_y_111_cast_fp16)[name = string("op_2031_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_26_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_26_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5624128)))];
            tensor<fp16, [1, 448, 32]> y_54_cast_fp16 = add(x = var_2031_cast_fp16, y = sep_module_26_tcn_6_norm_beta_to_fp16)[name = string("y_54_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_61_cast_fp16 = add(x = x_59_cast_fp16, y = y_54_cast_fp16)[name = string("x_61_cast_fp16")];
            bool var_2037 = const()[name = string("op_2037"), val = bool(true)];
            int32 var_2041 = const()[name = string("op_2041"), val = int32(448)];
            int32 var_2042 = const()[name = string("op_2042"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_225_cast_fp16 = add(x = x_61_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_225_cast_fp16")];
            tensor<int32, [1]> var_2052 = const()[name = string("op_2052"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_2054 = const()[name = string("op_2054"), val = tensor<int32, [1]>([1])];
            string input0_63_pad_type_0 = const()[name = string("input0_63_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_63_pad_0 = const()[name = string("input0_63_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_27_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5625088))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5725504))))[name = string("sep_module_27_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_63_cast_fp16 = conv(dilations = var_2054, groups = var_2042, pad = input0_63_pad_0, pad_type = input0_63_pad_type_0, strides = var_2052, weight = sep_module_27_tcn_0_weight_to_fp16_palettized, x = input_225_cast_fp16)[name = string("input0_63_cast_fp16")];
            fp32 var_2058_alpha_1 = const()[name = string("op_2058_alpha_1"), val = fp32(-0x1.76387ep-2)];
            tensor<fp16, [1, 448, 32]> var_2058_cast_fp16 = leaky_relu(alpha = fp16(-0x1.764p-2), x = input0_63_cast_fp16);
            tensor<int32, [1]> var_2062 = const()[name = string("op_2062"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_113_cast_fp16 = reduce_mean(axes = var_2062, keep_dims = var_2037, x = var_2058_cast_fp16)[name = string("mean_y_113_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2064_cast_fp16 = sub(x = var_2058_cast_fp16, y = mean_y_113_cast_fp16)[name = string("op_2064_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2065_cast_fp16 = square(x = var_2064_cast_fp16);
            tensor<int32, [1]> var_2066 = const()[name = string("op_2066"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_2067_cast_fp16 = reduce_mean(axes = var_2066, keep_dims = var_2037, x = var_2065_cast_fp16)[name = string("op_2067_cast_fp16")];
            fp16 var_2068_to_fp16 = const()[name = string("op_2068_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_2069_cast_fp16 = add(x = var_2067_cast_fp16, y = var_2068_to_fp16)[name = string("op_2069_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_113_cast_fp16 = sqrt(x = var_2069_cast_fp16)[name = string("std_y_113_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_27_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_27_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5725632)))];
            tensor<fp16, [1, 448, 32]> var_2072_cast_fp16 = mul(x = sep_module_27_tcn_2_norm_gamma_to_fp16, y = var_2064_cast_fp16)[name = string("op_2072_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2073_cast_fp16 = real_div(x = var_2072_cast_fp16, y = std_y_113_cast_fp16)[name = string("op_2073_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_27_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_27_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5726592)))];
            tensor<fp16, [1, 448, 32]> input_227_cast_fp16 = add(x = var_2073_cast_fp16, y = sep_module_27_tcn_2_norm_beta_to_fp16)[name = string("input_227_cast_fp16")];
            tensor<int32, [6]> input_229_pad_0 = const()[name = string("input_229_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string input_229_mode_0 = const()[name = string("input_229_mode_0"), val = string("constant")];
            fp16 const_27_to_fp16 = const()[name = string("const_27_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 34]> input_227_cast_fp16_state_input = read_state(input = input_227_cast_fp16_state);
            tensor<fp16, [1, 448, 34]> input_229_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 2]), end = tensor<int32, [3]>([1, 448, 34]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_227_cast_fp16, x = input_227_cast_fp16_state_input);
            write_state(data = input_229_cast_fp16, input = input_227_cast_fp16_state);
            tensor<int32, [1]> var_2078 = const()[name = string("op_2078"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_2080 = const()[name = string("op_2080"), val = tensor<int32, [1]>([1])];
            string input_231_pad_type_0 = const()[name = string("input_231_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_231_pad_0 = const()[name = string("input_231_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_27_tcn_4_weight_to_fp16 = const()[name = string("sep_module_27_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5727552)))];
            tensor<fp16, [1, 448, 32]> input_231_cast_fp16 = conv(dilations = var_2080, groups = var_2041, pad = input_231_pad_0, pad_type = input_231_pad_type_0, strides = var_2078, weight = sep_module_27_tcn_4_weight_to_fp16, x = input_229_cast_fp16)[name = string("input_231_cast_fp16")];
            fp32 var_2084_alpha_1 = const()[name = string("op_2084_alpha_1"), val = fp32(0x1.eef4fp-1)];
            tensor<fp16, [1, 448, 32]> var_2084_cast_fp16 = leaky_relu(alpha = fp16(0x1.efp-1), x = input_231_cast_fp16);
            tensor<int32, [1]> var_2088 = const()[name = string("op_2088"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_115_cast_fp16 = reduce_mean(axes = var_2088, keep_dims = var_2037, x = var_2084_cast_fp16)[name = string("mean_y_115_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2090_cast_fp16 = sub(x = var_2084_cast_fp16, y = mean_y_115_cast_fp16)[name = string("op_2090_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2091_cast_fp16 = square(x = var_2090_cast_fp16);
            tensor<int32, [1]> var_2092 = const()[name = string("op_2092"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_2093_cast_fp16 = reduce_mean(axes = var_2092, keep_dims = var_2037, x = var_2091_cast_fp16)[name = string("op_2093_cast_fp16")];
            fp16 var_2094_to_fp16 = const()[name = string("op_2094_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_2095_cast_fp16 = add(x = var_2093_cast_fp16, y = var_2094_to_fp16)[name = string("op_2095_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_115_cast_fp16 = sqrt(x = var_2095_cast_fp16)[name = string("std_y_115_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_27_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_27_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5730304)))];
            tensor<fp16, [1, 448, 32]> var_2098_cast_fp16 = mul(x = sep_module_27_tcn_6_norm_gamma_to_fp16, y = var_2090_cast_fp16)[name = string("op_2098_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2099_cast_fp16 = real_div(x = var_2098_cast_fp16, y = std_y_115_cast_fp16)[name = string("op_2099_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_27_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_27_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5731264)))];
            tensor<fp16, [1, 448, 32]> y_56_cast_fp16 = add(x = var_2099_cast_fp16, y = sep_module_27_tcn_6_norm_beta_to_fp16)[name = string("y_56_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_63_cast_fp16 = add(x = x_61_cast_fp16, y = y_56_cast_fp16)[name = string("x_63_cast_fp16")];
            bool var_2105 = const()[name = string("op_2105"), val = bool(true)];
            int32 var_2109 = const()[name = string("op_2109"), val = int32(448)];
            int32 var_2111 = const()[name = string("op_2111"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_233_cast_fp16 = add(x = x_63_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_233_cast_fp16")];
            tensor<int32, [1]> var_2121 = const()[name = string("op_2121"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_2123 = const()[name = string("op_2123"), val = tensor<int32, [1]>([1])];
            string input0_65_pad_type_0 = const()[name = string("input0_65_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_65_pad_0 = const()[name = string("input0_65_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_28_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5732224))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5832640))))[name = string("sep_module_28_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_65_cast_fp16 = conv(dilations = var_2123, groups = var_2111, pad = input0_65_pad_0, pad_type = input0_65_pad_type_0, strides = var_2121, weight = sep_module_28_tcn_0_weight_to_fp16_palettized, x = input_233_cast_fp16)[name = string("input0_65_cast_fp16")];
            fp32 var_2127_alpha_1 = const()[name = string("op_2127_alpha_1"), val = fp32(-0x1.53fe8ep-3)];
            tensor<fp16, [1, 448, 32]> var_2127_cast_fp16 = leaky_relu(alpha = fp16(-0x1.54p-3), x = input0_65_cast_fp16);
            tensor<int32, [1]> var_2131 = const()[name = string("op_2131"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_117_cast_fp16 = reduce_mean(axes = var_2131, keep_dims = var_2105, x = var_2127_cast_fp16)[name = string("mean_y_117_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2133_cast_fp16 = sub(x = var_2127_cast_fp16, y = mean_y_117_cast_fp16)[name = string("op_2133_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2134_cast_fp16 = square(x = var_2133_cast_fp16);
            tensor<int32, [1]> var_2135 = const()[name = string("op_2135"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_2136_cast_fp16 = reduce_mean(axes = var_2135, keep_dims = var_2105, x = var_2134_cast_fp16)[name = string("op_2136_cast_fp16")];
            fp16 var_2137_to_fp16 = const()[name = string("op_2137_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_2138_cast_fp16 = add(x = var_2136_cast_fp16, y = var_2137_to_fp16)[name = string("op_2138_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_117_cast_fp16 = sqrt(x = var_2138_cast_fp16)[name = string("std_y_117_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_28_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_28_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5832768)))];
            tensor<fp16, [1, 448, 32]> var_2141_cast_fp16 = mul(x = sep_module_28_tcn_2_norm_gamma_to_fp16, y = var_2133_cast_fp16)[name = string("op_2141_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2142_cast_fp16 = real_div(x = var_2141_cast_fp16, y = std_y_117_cast_fp16)[name = string("op_2142_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_28_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_28_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5833728)))];
            tensor<fp16, [1, 448, 32]> input_235_cast_fp16 = add(x = var_2142_cast_fp16, y = sep_module_28_tcn_2_norm_beta_to_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, 4, 0])];
            string input_237_mode_0 = const()[name = string("input_237_mode_0"), val = string("constant")];
            fp16 const_28_to_fp16 = const()[name = string("const_28_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 36]> input_235_cast_fp16_state_input = read_state(input = input_235_cast_fp16_state);
            tensor<fp16, [1, 448, 36]> input_237_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 4]), end = tensor<int32, [3]>([1, 448, 36]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_235_cast_fp16, x = input_235_cast_fp16_state_input);
            write_state(data = input_237_cast_fp16, input = input_235_cast_fp16_state);
            tensor<int32, [1]> var_2147 = const()[name = string("op_2147"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_2149 = const()[name = string("op_2149"), val = tensor<int32, [1]>([2])];
            string input_239_pad_type_0 = const()[name = string("input_239_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_239_pad_0 = const()[name = string("input_239_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_28_tcn_4_weight_to_fp16 = const()[name = string("sep_module_28_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5834688)))];
            tensor<fp16, [1, 448, 32]> input_239_cast_fp16 = conv(dilations = var_2149, groups = var_2109, pad = input_239_pad_0, pad_type = input_239_pad_type_0, strides = var_2147, weight = sep_module_28_tcn_4_weight_to_fp16, x = input_237_cast_fp16)[name = string("input_239_cast_fp16")];
            fp32 var_2153_alpha_1 = const()[name = string("op_2153_alpha_1"), val = fp32(0x1.fb8356p-1)];
            tensor<fp16, [1, 448, 32]> var_2153_cast_fp16 = leaky_relu(alpha = fp16(0x1.fb8p-1), x = input_239_cast_fp16);
            tensor<int32, [1]> var_2157 = const()[name = string("op_2157"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_119_cast_fp16 = reduce_mean(axes = var_2157, keep_dims = var_2105, x = var_2153_cast_fp16)[name = string("mean_y_119_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2159_cast_fp16 = sub(x = var_2153_cast_fp16, y = mean_y_119_cast_fp16)[name = string("op_2159_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2160_cast_fp16 = square(x = var_2159_cast_fp16);
            tensor<int32, [1]> var_2161 = const()[name = string("op_2161"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_2162_cast_fp16 = reduce_mean(axes = var_2161, keep_dims = var_2105, x = var_2160_cast_fp16)[name = string("op_2162_cast_fp16")];
            fp16 var_2163_to_fp16 = const()[name = string("op_2163_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_2164_cast_fp16 = add(x = var_2162_cast_fp16, y = var_2163_to_fp16)[name = string("op_2164_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_119_cast_fp16 = sqrt(x = var_2164_cast_fp16)[name = string("std_y_119_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_28_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_28_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5837440)))];
            tensor<fp16, [1, 448, 32]> var_2167_cast_fp16 = mul(x = sep_module_28_tcn_6_norm_gamma_to_fp16, y = var_2159_cast_fp16)[name = string("op_2167_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2168_cast_fp16 = real_div(x = var_2167_cast_fp16, y = std_y_119_cast_fp16)[name = string("op_2168_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_28_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_28_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5838400)))];
            tensor<fp16, [1, 448, 32]> y_58_cast_fp16 = add(x = var_2168_cast_fp16, y = sep_module_28_tcn_6_norm_beta_to_fp16)[name = string("y_58_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_65_cast_fp16 = add(x = x_63_cast_fp16, y = y_58_cast_fp16)[name = string("x_65_cast_fp16")];
            bool var_2174 = const()[name = string("op_2174"), val = bool(true)];
            int32 var_2178 = const()[name = string("op_2178"), val = int32(448)];
            int32 var_2180 = const()[name = string("op_2180"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_241_cast_fp16 = add(x = x_65_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_241_cast_fp16")];
            tensor<int32, [1]> var_2190 = const()[name = string("op_2190"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_2192 = const()[name = string("op_2192"), val = tensor<int32, [1]>([1])];
            string input0_67_pad_type_0 = const()[name = string("input0_67_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_67_pad_0 = const()[name = string("input0_67_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_29_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5839360))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5939776))))[name = string("sep_module_29_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_67_cast_fp16 = conv(dilations = var_2192, groups = var_2180, pad = input0_67_pad_0, pad_type = input0_67_pad_type_0, strides = var_2190, weight = sep_module_29_tcn_0_weight_to_fp16_palettized, x = input_241_cast_fp16)[name = string("input0_67_cast_fp16")];
            fp32 var_2196_alpha_1 = const()[name = string("op_2196_alpha_1"), val = fp32(-0x1.3e686p-3)];
            tensor<fp16, [1, 448, 32]> var_2196_cast_fp16 = leaky_relu(alpha = fp16(-0x1.3e8p-3), x = input0_67_cast_fp16);
            tensor<int32, [1]> var_2200 = const()[name = string("op_2200"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_121_cast_fp16 = reduce_mean(axes = var_2200, keep_dims = var_2174, x = var_2196_cast_fp16)[name = string("mean_y_121_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2202_cast_fp16 = sub(x = var_2196_cast_fp16, y = mean_y_121_cast_fp16)[name = string("op_2202_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2203_cast_fp16 = square(x = var_2202_cast_fp16);
            tensor<int32, [1]> var_2204 = const()[name = string("op_2204"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_2205_cast_fp16 = reduce_mean(axes = var_2204, keep_dims = var_2174, x = var_2203_cast_fp16)[name = string("op_2205_cast_fp16")];
            fp16 var_2206_to_fp16 = const()[name = string("op_2206_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_2207_cast_fp16 = add(x = var_2205_cast_fp16, y = var_2206_to_fp16)[name = string("op_2207_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_121_cast_fp16 = sqrt(x = var_2207_cast_fp16)[name = string("std_y_121_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_29_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_29_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5939904)))];
            tensor<fp16, [1, 448, 32]> var_2210_cast_fp16 = mul(x = sep_module_29_tcn_2_norm_gamma_to_fp16, y = var_2202_cast_fp16)[name = string("op_2210_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2211_cast_fp16 = real_div(x = var_2210_cast_fp16, y = std_y_121_cast_fp16)[name = string("op_2211_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_29_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_29_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5940864)))];
            tensor<fp16, [1, 448, 32]> input_243_cast_fp16 = add(x = var_2211_cast_fp16, y = sep_module_29_tcn_2_norm_beta_to_fp16)[name = string("input_243_cast_fp16")];
            tensor<int32, [6]> input_245_pad_0 = const()[name = string("input_245_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 8, 0])];
            string input_245_mode_0 = const()[name = string("input_245_mode_0"), val = string("constant")];
            fp16 const_29_to_fp16 = const()[name = string("const_29_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 40]> input_243_cast_fp16_state_input = read_state(input = input_243_cast_fp16_state);
            tensor<fp16, [1, 448, 40]> input_245_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 8]), end = tensor<int32, [3]>([1, 448, 40]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_243_cast_fp16, x = input_243_cast_fp16_state_input);
            write_state(data = input_245_cast_fp16, input = input_243_cast_fp16_state);
            tensor<int32, [1]> var_2216 = const()[name = string("op_2216"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_2218 = const()[name = string("op_2218"), val = tensor<int32, [1]>([4])];
            string input_247_pad_type_0 = const()[name = string("input_247_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_247_pad_0 = const()[name = string("input_247_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_29_tcn_4_weight_to_fp16 = const()[name = string("sep_module_29_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5941824)))];
            tensor<fp16, [1, 448, 32]> input_247_cast_fp16 = conv(dilations = var_2218, groups = var_2178, pad = input_247_pad_0, pad_type = input_247_pad_type_0, strides = var_2216, weight = sep_module_29_tcn_4_weight_to_fp16, x = input_245_cast_fp16)[name = string("input_247_cast_fp16")];
            fp32 var_2222_alpha_1 = const()[name = string("op_2222_alpha_1"), val = fp32(0x1.f6ec58p-1)];
            tensor<fp16, [1, 448, 32]> var_2222_cast_fp16 = leaky_relu(alpha = fp16(0x1.f7p-1), x = input_247_cast_fp16);
            tensor<int32, [1]> var_2226 = const()[name = string("op_2226"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_123_cast_fp16 = reduce_mean(axes = var_2226, keep_dims = var_2174, x = var_2222_cast_fp16)[name = string("mean_y_123_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2228_cast_fp16 = sub(x = var_2222_cast_fp16, y = mean_y_123_cast_fp16)[name = string("op_2228_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2229_cast_fp16 = square(x = var_2228_cast_fp16);
            tensor<int32, [1]> var_2230 = const()[name = string("op_2230"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_2231_cast_fp16 = reduce_mean(axes = var_2230, keep_dims = var_2174, x = var_2229_cast_fp16)[name = string("op_2231_cast_fp16")];
            fp16 var_2232_to_fp16 = const()[name = string("op_2232_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_2233_cast_fp16 = add(x = var_2231_cast_fp16, y = var_2232_to_fp16)[name = string("op_2233_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_123_cast_fp16 = sqrt(x = var_2233_cast_fp16)[name = string("std_y_123_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_29_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_29_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5944576)))];
            tensor<fp16, [1, 448, 32]> var_2236_cast_fp16 = mul(x = sep_module_29_tcn_6_norm_gamma_to_fp16, y = var_2228_cast_fp16)[name = string("op_2236_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2237_cast_fp16 = real_div(x = var_2236_cast_fp16, y = std_y_123_cast_fp16)[name = string("op_2237_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_29_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_29_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5945536)))];
            tensor<fp16, [1, 448, 32]> y_60_cast_fp16 = add(x = var_2237_cast_fp16, y = sep_module_29_tcn_6_norm_beta_to_fp16)[name = string("y_60_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_67_cast_fp16 = add(x = x_65_cast_fp16, y = y_60_cast_fp16)[name = string("x_67_cast_fp16")];
            bool var_2243 = const()[name = string("op_2243"), val = bool(true)];
            int32 var_2247 = const()[name = string("op_2247"), val = int32(448)];
            int32 var_2249 = const()[name = string("op_2249"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_249_cast_fp16 = add(x = x_67_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_249_cast_fp16")];
            tensor<int32, [1]> var_2259 = const()[name = string("op_2259"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_2261 = const()[name = string("op_2261"), val = tensor<int32, [1]>([1])];
            string input0_69_pad_type_0 = const()[name = string("input0_69_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_69_pad_0 = const()[name = string("input0_69_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_30_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(5946496))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6046912))))[name = string("sep_module_30_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_69_cast_fp16 = conv(dilations = var_2261, groups = var_2249, pad = input0_69_pad_0, pad_type = input0_69_pad_type_0, strides = var_2259, weight = sep_module_30_tcn_0_weight_to_fp16_palettized, x = input_249_cast_fp16)[name = string("input0_69_cast_fp16")];
            fp32 var_2265_alpha_1 = const()[name = string("op_2265_alpha_1"), val = fp32(-0x1.f3b2ecp-3)];
            tensor<fp16, [1, 448, 32]> var_2265_cast_fp16 = leaky_relu(alpha = fp16(-0x1.f3cp-3), x = input0_69_cast_fp16);
            tensor<int32, [1]> var_2269 = const()[name = string("op_2269"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_125_cast_fp16 = reduce_mean(axes = var_2269, keep_dims = var_2243, x = var_2265_cast_fp16)[name = string("mean_y_125_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2271_cast_fp16 = sub(x = var_2265_cast_fp16, y = mean_y_125_cast_fp16)[name = string("op_2271_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2272_cast_fp16 = square(x = var_2271_cast_fp16);
            tensor<int32, [1]> var_2273 = const()[name = string("op_2273"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_2274_cast_fp16 = reduce_mean(axes = var_2273, keep_dims = var_2243, x = var_2272_cast_fp16)[name = string("op_2274_cast_fp16")];
            fp16 var_2275_to_fp16 = const()[name = string("op_2275_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_2276_cast_fp16 = add(x = var_2274_cast_fp16, y = var_2275_to_fp16)[name = string("op_2276_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_125_cast_fp16 = sqrt(x = var_2276_cast_fp16)[name = string("std_y_125_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_30_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_30_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6047040)))];
            tensor<fp16, [1, 448, 32]> var_2279_cast_fp16 = mul(x = sep_module_30_tcn_2_norm_gamma_to_fp16, y = var_2271_cast_fp16)[name = string("op_2279_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2280_cast_fp16 = real_div(x = var_2279_cast_fp16, y = std_y_125_cast_fp16)[name = string("op_2280_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_30_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_30_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6048000)))];
            tensor<fp16, [1, 448, 32]> input_251_cast_fp16 = add(x = var_2280_cast_fp16, y = sep_module_30_tcn_2_norm_beta_to_fp16)[name = string("input_251_cast_fp16")];
            tensor<int32, [6]> input_253_pad_0 = const()[name = string("input_253_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 16, 0])];
            string input_253_mode_0 = const()[name = string("input_253_mode_0"), val = string("constant")];
            fp16 const_30_to_fp16 = const()[name = string("const_30_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 48]> input_251_cast_fp16_state_input = read_state(input = input_251_cast_fp16_state);
            tensor<fp16, [1, 448, 48]> input_253_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 16]), end = tensor<int32, [3]>([1, 448, 48]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_251_cast_fp16, x = input_251_cast_fp16_state_input);
            write_state(data = input_253_cast_fp16, input = input_251_cast_fp16_state);
            tensor<int32, [1]> var_2285 = const()[name = string("op_2285"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_2287 = const()[name = string("op_2287"), val = tensor<int32, [1]>([8])];
            string input_255_pad_type_0 = const()[name = string("input_255_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_255_pad_0 = const()[name = string("input_255_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_30_tcn_4_weight_to_fp16 = const()[name = string("sep_module_30_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6048960)))];
            tensor<fp16, [1, 448, 32]> input_255_cast_fp16 = conv(dilations = var_2287, groups = var_2247, pad = input_255_pad_0, pad_type = input_255_pad_type_0, strides = var_2285, weight = sep_module_30_tcn_4_weight_to_fp16, x = input_253_cast_fp16)[name = string("input_255_cast_fp16")];
            fp32 var_2291_alpha_1 = const()[name = string("op_2291_alpha_1"), val = fp32(0x1.d866b2p-1)];
            tensor<fp16, [1, 448, 32]> var_2291_cast_fp16 = leaky_relu(alpha = fp16(0x1.d88p-1), x = input_255_cast_fp16);
            tensor<int32, [1]> var_2295 = const()[name = string("op_2295"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_127_cast_fp16 = reduce_mean(axes = var_2295, keep_dims = var_2243, x = var_2291_cast_fp16)[name = string("mean_y_127_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2297_cast_fp16 = sub(x = var_2291_cast_fp16, y = mean_y_127_cast_fp16)[name = string("op_2297_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2298_cast_fp16 = square(x = var_2297_cast_fp16);
            tensor<int32, [1]> var_2299 = const()[name = string("op_2299"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_2300_cast_fp16 = reduce_mean(axes = var_2299, keep_dims = var_2243, x = var_2298_cast_fp16)[name = string("op_2300_cast_fp16")];
            fp16 var_2301_to_fp16 = const()[name = string("op_2301_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_2302_cast_fp16 = add(x = var_2300_cast_fp16, y = var_2301_to_fp16)[name = string("op_2302_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_127_cast_fp16 = sqrt(x = var_2302_cast_fp16)[name = string("std_y_127_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_30_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_30_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6051712)))];
            tensor<fp16, [1, 448, 32]> var_2305_cast_fp16 = mul(x = sep_module_30_tcn_6_norm_gamma_to_fp16, y = var_2297_cast_fp16)[name = string("op_2305_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2306_cast_fp16 = real_div(x = var_2305_cast_fp16, y = std_y_127_cast_fp16)[name = string("op_2306_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_30_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_30_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6052672)))];
            tensor<fp16, [1, 448, 32]> y_62_cast_fp16 = add(x = var_2306_cast_fp16, y = sep_module_30_tcn_6_norm_beta_to_fp16)[name = string("y_62_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_69_cast_fp16 = add(x = x_67_cast_fp16, y = y_62_cast_fp16)[name = string("x_69_cast_fp16")];
            bool var_2312 = const()[name = string("op_2312"), val = bool(true)];
            int32 var_2316 = const()[name = string("op_2316"), val = int32(448)];
            int32 var_2318 = const()[name = string("op_2318"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_257_cast_fp16 = add(x = x_69_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_257_cast_fp16")];
            tensor<int32, [1]> var_2328 = const()[name = string("op_2328"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_2330 = const()[name = string("op_2330"), val = tensor<int32, [1]>([1])];
            string input0_71_pad_type_0 = const()[name = string("input0_71_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_71_pad_0 = const()[name = string("input0_71_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_31_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6053632))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6154048))))[name = string("sep_module_31_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_71_cast_fp16 = conv(dilations = var_2330, groups = var_2318, pad = input0_71_pad_0, pad_type = input0_71_pad_type_0, strides = var_2328, weight = sep_module_31_tcn_0_weight_to_fp16_palettized, x = input_257_cast_fp16)[name = string("input0_71_cast_fp16")];
            fp32 var_2334_alpha_1 = const()[name = string("op_2334_alpha_1"), val = fp32(0x1.a5378ap-2)];
            tensor<fp16, [1, 448, 32]> var_2334_cast_fp16 = leaky_relu(alpha = fp16(0x1.a54p-2), x = input0_71_cast_fp16);
            tensor<int32, [1]> var_2338 = const()[name = string("op_2338"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_129_cast_fp16 = reduce_mean(axes = var_2338, keep_dims = var_2312, x = var_2334_cast_fp16)[name = string("mean_y_129_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2340_cast_fp16 = sub(x = var_2334_cast_fp16, y = mean_y_129_cast_fp16)[name = string("op_2340_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2341_cast_fp16 = square(x = var_2340_cast_fp16);
            tensor<int32, [1]> var_2342 = const()[name = string("op_2342"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_2343_cast_fp16 = reduce_mean(axes = var_2342, keep_dims = var_2312, x = var_2341_cast_fp16)[name = string("op_2343_cast_fp16")];
            fp16 var_2344_to_fp16 = const()[name = string("op_2344_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_2345_cast_fp16 = add(x = var_2343_cast_fp16, y = var_2344_to_fp16)[name = string("op_2345_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_129_cast_fp16 = sqrt(x = var_2345_cast_fp16)[name = string("std_y_129_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_31_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_31_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6154176)))];
            tensor<fp16, [1, 448, 32]> var_2348_cast_fp16 = mul(x = sep_module_31_tcn_2_norm_gamma_to_fp16, y = var_2340_cast_fp16)[name = string("op_2348_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2349_cast_fp16 = real_div(x = var_2348_cast_fp16, y = std_y_129_cast_fp16)[name = string("op_2349_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_31_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_31_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6155136)))];
            tensor<fp16, [1, 448, 32]> input_259_cast_fp16 = add(x = var_2349_cast_fp16, y = sep_module_31_tcn_2_norm_beta_to_fp16)[name = string("input_259_cast_fp16")];
            tensor<int32, [6]> input_261_pad_0 = const()[name = string("input_261_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 32, 0])];
            string input_261_mode_0 = const()[name = string("input_261_mode_0"), val = string("constant")];
            fp16 const_31_to_fp16 = const()[name = string("const_31_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 64]> input_259_cast_fp16_state_input = read_state(input = input_259_cast_fp16_state);
            tensor<fp16, [1, 448, 64]> input_261_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 32]), end = tensor<int32, [3]>([1, 448, 64]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_259_cast_fp16, x = input_259_cast_fp16_state_input);
            write_state(data = input_261_cast_fp16, input = input_259_cast_fp16_state);
            tensor<int32, [1]> var_2354 = const()[name = string("op_2354"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_2356 = const()[name = string("op_2356"), val = tensor<int32, [1]>([16])];
            string input_263_pad_type_0 = const()[name = string("input_263_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_263_pad_0 = const()[name = string("input_263_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_31_tcn_4_weight_to_fp16 = const()[name = string("sep_module_31_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6156096)))];
            tensor<fp16, [1, 448, 32]> input_263_cast_fp16 = conv(dilations = var_2356, groups = var_2316, pad = input_263_pad_0, pad_type = input_263_pad_type_0, strides = var_2354, weight = sep_module_31_tcn_4_weight_to_fp16, x = input_261_cast_fp16)[name = string("input_263_cast_fp16")];
            fp32 var_2360_alpha_1 = const()[name = string("op_2360_alpha_1"), val = fp32(0x1.a41b28p-2)];
            tensor<fp16, [1, 448, 32]> var_2360_cast_fp16 = leaky_relu(alpha = fp16(0x1.a4p-2), x = input_263_cast_fp16);
            tensor<int32, [1]> var_2364 = const()[name = string("op_2364"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_131_cast_fp16 = reduce_mean(axes = var_2364, keep_dims = var_2312, x = var_2360_cast_fp16)[name = string("mean_y_131_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2366_cast_fp16 = sub(x = var_2360_cast_fp16, y = mean_y_131_cast_fp16)[name = string("op_2366_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2367_cast_fp16 = square(x = var_2366_cast_fp16);
            tensor<int32, [1]> var_2368 = const()[name = string("op_2368"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_2369_cast_fp16 = reduce_mean(axes = var_2368, keep_dims = var_2312, x = var_2367_cast_fp16)[name = string("op_2369_cast_fp16")];
            fp16 var_2370_to_fp16 = const()[name = string("op_2370_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_2371_cast_fp16 = add(x = var_2369_cast_fp16, y = var_2370_to_fp16)[name = string("op_2371_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_131_cast_fp16 = sqrt(x = var_2371_cast_fp16)[name = string("std_y_131_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_31_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_31_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6158848)))];
            tensor<fp16, [1, 448, 32]> var_2374_cast_fp16 = mul(x = sep_module_31_tcn_6_norm_gamma_to_fp16, y = var_2366_cast_fp16)[name = string("op_2374_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2375_cast_fp16 = real_div(x = var_2374_cast_fp16, y = std_y_131_cast_fp16)[name = string("op_2375_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_31_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_31_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6159808)))];
            tensor<fp16, [1, 448, 32]> y_64_cast_fp16 = add(x = var_2375_cast_fp16, y = sep_module_31_tcn_6_norm_beta_to_fp16)[name = string("y_64_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_71_cast_fp16 = add(x = x_69_cast_fp16, y = y_64_cast_fp16)[name = string("x_71_cast_fp16")];
            bool var_2381 = const()[name = string("op_2381"), val = bool(true)];
            int32 var_2385 = const()[name = string("op_2385"), val = int32(448)];
            int32 var_2387 = const()[name = string("op_2387"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_265_cast_fp16 = add(x = x_71_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_265_cast_fp16")];
            tensor<int32, [1]> var_2397 = const()[name = string("op_2397"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_2399 = const()[name = string("op_2399"), val = tensor<int32, [1]>([1])];
            string input0_73_pad_type_0 = const()[name = string("input0_73_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_73_pad_0 = const()[name = string("input0_73_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_32_tcn_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6160768))), lut = tensor<fp16, [1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6261184))))[name = string("sep_module_32_tcn_0_weight_to_fp16_palettized")];
            tensor<fp16, [1, 448, 32]> input0_73_cast_fp16 = conv(dilations = var_2399, groups = var_2387, pad = input0_73_pad_0, pad_type = input0_73_pad_type_0, strides = var_2397, weight = sep_module_32_tcn_0_weight_to_fp16_palettized, x = input_265_cast_fp16)[name = string("input0_73_cast_fp16")];
            fp32 var_2403_alpha_1 = const()[name = string("op_2403_alpha_1"), val = fp32(0x1.8f7024p-2)];
            tensor<fp16, [1, 448, 32]> var_2403_cast_fp16 = leaky_relu(alpha = fp16(0x1.8f8p-2), x = input0_73_cast_fp16);
            tensor<int32, [1]> var_2407 = const()[name = string("op_2407"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_133_cast_fp16 = reduce_mean(axes = var_2407, keep_dims = var_2381, x = var_2403_cast_fp16)[name = string("mean_y_133_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2409_cast_fp16 = sub(x = var_2403_cast_fp16, y = mean_y_133_cast_fp16)[name = string("op_2409_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2410_cast_fp16 = square(x = var_2409_cast_fp16);
            tensor<int32, [1]> var_2411 = const()[name = string("op_2411"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_2412_cast_fp16 = reduce_mean(axes = var_2411, keep_dims = var_2381, x = var_2410_cast_fp16)[name = string("op_2412_cast_fp16")];
            fp16 var_2413_to_fp16 = const()[name = string("op_2413_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_2414_cast_fp16 = add(x = var_2412_cast_fp16, y = var_2413_to_fp16)[name = string("op_2414_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_133_cast_fp16 = sqrt(x = var_2414_cast_fp16)[name = string("std_y_133_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_32_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_32_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6261312)))];
            tensor<fp16, [1, 448, 32]> var_2417_cast_fp16 = mul(x = sep_module_32_tcn_2_norm_gamma_to_fp16, y = var_2409_cast_fp16)[name = string("op_2417_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2418_cast_fp16 = real_div(x = var_2417_cast_fp16, y = std_y_133_cast_fp16)[name = string("op_2418_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_32_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_32_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6262272)))];
            tensor<fp16, [1, 448, 32]> input_267_cast_fp16 = add(x = var_2418_cast_fp16, y = sep_module_32_tcn_2_norm_beta_to_fp16)[name = string("input_267_cast_fp16")];
            tensor<int32, [6]> input_269_pad_0 = const()[name = string("input_269_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 64, 0])];
            string input_269_mode_0 = const()[name = string("input_269_mode_0"), val = string("constant")];
            fp16 const_32_to_fp16 = const()[name = string("const_32_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 96]> input_267_cast_fp16_state_input = read_state(input = input_267_cast_fp16_state);
            tensor<fp16, [1, 448, 96]> input_269_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 64]), end = tensor<int32, [3]>([1, 448, 96]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_267_cast_fp16, x = input_267_cast_fp16_state_input);
            write_state(data = input_269_cast_fp16, input = input_267_cast_fp16_state);
            tensor<int32, [1]> var_2423 = const()[name = string("op_2423"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_2425 = const()[name = string("op_2425"), val = tensor<int32, [1]>([32])];
            string input_271_pad_type_0 = const()[name = string("input_271_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_271_pad_0 = const()[name = string("input_271_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_32_tcn_4_weight_to_fp16 = const()[name = string("sep_module_32_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6263232)))];
            tensor<fp16, [1, 448, 32]> input_271_cast_fp16 = conv(dilations = var_2425, groups = var_2385, pad = input_271_pad_0, pad_type = input_271_pad_type_0, strides = var_2423, weight = sep_module_32_tcn_4_weight_to_fp16, x = input_269_cast_fp16)[name = string("input_271_cast_fp16")];
            fp32 var_2429_alpha_1 = const()[name = string("op_2429_alpha_1"), val = fp32(0x1.8eb004p-2)];
            tensor<fp16, [1, 448, 32]> var_2429_cast_fp16 = leaky_relu(alpha = fp16(0x1.8ecp-2), x = input_271_cast_fp16);
            tensor<int32, [1]> var_2433 = const()[name = string("op_2433"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_135_cast_fp16 = reduce_mean(axes = var_2433, keep_dims = var_2381, x = var_2429_cast_fp16)[name = string("mean_y_135_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2435_cast_fp16 = sub(x = var_2429_cast_fp16, y = mean_y_135_cast_fp16)[name = string("op_2435_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2436_cast_fp16 = square(x = var_2435_cast_fp16);
            tensor<int32, [1]> var_2437 = const()[name = string("op_2437"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_2438_cast_fp16 = reduce_mean(axes = var_2437, keep_dims = var_2381, x = var_2436_cast_fp16)[name = string("op_2438_cast_fp16")];
            fp16 var_2439_to_fp16 = const()[name = string("op_2439_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_2440_cast_fp16 = add(x = var_2438_cast_fp16, y = var_2439_to_fp16)[name = string("op_2440_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_135_cast_fp16 = sqrt(x = var_2440_cast_fp16)[name = string("std_y_135_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_32_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_32_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6265984)))];
            tensor<fp16, [1, 448, 32]> var_2443_cast_fp16 = mul(x = sep_module_32_tcn_6_norm_gamma_to_fp16, y = var_2435_cast_fp16)[name = string("op_2443_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2444_cast_fp16 = real_div(x = var_2443_cast_fp16, y = std_y_135_cast_fp16)[name = string("op_2444_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_32_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_32_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6266944)))];
            tensor<fp16, [1, 448, 32]> y_66_cast_fp16 = add(x = var_2444_cast_fp16, y = sep_module_32_tcn_6_norm_beta_to_fp16)[name = string("y_66_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_73_cast_fp16 = add(x = x_71_cast_fp16, y = y_66_cast_fp16)[name = string("x_73_cast_fp16")];
            bool var_2450 = const()[name = string("op_2450"), val = bool(true)];
            int32 var_2454 = const()[name = string("op_2454"), val = int32(448)];
            int32 var_2456 = const()[name = string("op_2456"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_273_cast_fp16 = add(x = x_73_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_273_cast_fp16")];
            tensor<int32, [1]> var_2466 = const()[name = string("op_2466"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_2468 = const()[name = string("op_2468"), val = tensor<int32, [1]>([1])];
            string input0_75_pad_type_0 = const()[name = string("input0_75_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_75_pad_0 = const()[name = string("input0_75_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_33_tcn_0_weight_to_fp16 = const()[name = string("sep_module_33_tcn_0_weight_to_fp16"), val = tensor<fp16, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6267904)))];
            tensor<fp16, [1, 448, 32]> input0_75_cast_fp16 = conv(dilations = var_2468, groups = var_2456, pad = input0_75_pad_0, pad_type = input0_75_pad_type_0, strides = var_2466, weight = sep_module_33_tcn_0_weight_to_fp16, x = input_273_cast_fp16)[name = string("input0_75_cast_fp16")];
            fp32 var_2472_alpha_1 = const()[name = string("op_2472_alpha_1"), val = fp32(0x1.59260ap-2)];
            tensor<fp16, [1, 448, 32]> var_2472_cast_fp16 = leaky_relu(alpha = fp16(0x1.594p-2), x = input0_75_cast_fp16);
            tensor<int32, [1]> var_2476 = const()[name = string("op_2476"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_137_cast_fp16 = reduce_mean(axes = var_2476, keep_dims = var_2450, x = var_2472_cast_fp16)[name = string("mean_y_137_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2478_cast_fp16 = sub(x = var_2472_cast_fp16, y = mean_y_137_cast_fp16)[name = string("op_2478_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2479_cast_fp16 = square(x = var_2478_cast_fp16);
            tensor<int32, [1]> var_2480 = const()[name = string("op_2480"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_2481_cast_fp16 = reduce_mean(axes = var_2480, keep_dims = var_2450, x = var_2479_cast_fp16)[name = string("op_2481_cast_fp16")];
            fp16 var_2482_to_fp16 = const()[name = string("op_2482_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_2483_cast_fp16 = add(x = var_2481_cast_fp16, y = var_2482_to_fp16)[name = string("op_2483_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_137_cast_fp16 = sqrt(x = var_2483_cast_fp16)[name = string("std_y_137_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_33_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_33_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6669376)))];
            tensor<fp16, [1, 448, 32]> var_2486_cast_fp16 = mul(x = sep_module_33_tcn_2_norm_gamma_to_fp16, y = var_2478_cast_fp16)[name = string("op_2486_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2487_cast_fp16 = real_div(x = var_2486_cast_fp16, y = std_y_137_cast_fp16)[name = string("op_2487_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_33_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_33_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6670336)))];
            tensor<fp16, [1, 448, 32]> input_275_cast_fp16 = add(x = var_2487_cast_fp16, y = sep_module_33_tcn_2_norm_beta_to_fp16)[name = string("input_275_cast_fp16")];
            tensor<int32, [6]> input_277_pad_0 = const()[name = string("input_277_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 128, 0])];
            string input_277_mode_0 = const()[name = string("input_277_mode_0"), val = string("constant")];
            fp16 const_33_to_fp16 = const()[name = string("const_33_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 160]> input_275_cast_fp16_state_input = read_state(input = input_275_cast_fp16_state);
            tensor<fp16, [1, 448, 160]> input_277_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 128]), end = tensor<int32, [3]>([1, 448, 160]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_275_cast_fp16, x = input_275_cast_fp16_state_input);
            write_state(data = input_277_cast_fp16, input = input_275_cast_fp16_state);
            tensor<int32, [1]> var_2492 = const()[name = string("op_2492"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_2494 = const()[name = string("op_2494"), val = tensor<int32, [1]>([64])];
            string input_279_pad_type_0 = const()[name = string("input_279_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_279_pad_0 = const()[name = string("input_279_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_33_tcn_4_weight_to_fp16 = const()[name = string("sep_module_33_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6671296)))];
            tensor<fp16, [1, 448, 32]> input_279_cast_fp16 = conv(dilations = var_2494, groups = var_2454, pad = input_279_pad_0, pad_type = input_279_pad_type_0, strides = var_2492, weight = sep_module_33_tcn_4_weight_to_fp16, x = input_277_cast_fp16)[name = string("input_279_cast_fp16")];
            fp32 var_2498_alpha_1 = const()[name = string("op_2498_alpha_1"), val = fp32(0x1.586338p-2)];
            tensor<fp16, [1, 448, 32]> var_2498_cast_fp16 = leaky_relu(alpha = fp16(0x1.588p-2), x = input_279_cast_fp16);
            tensor<int32, [1]> var_2502 = const()[name = string("op_2502"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_139_cast_fp16 = reduce_mean(axes = var_2502, keep_dims = var_2450, x = var_2498_cast_fp16)[name = string("mean_y_139_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2504_cast_fp16 = sub(x = var_2498_cast_fp16, y = mean_y_139_cast_fp16)[name = string("op_2504_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2505_cast_fp16 = square(x = var_2504_cast_fp16);
            tensor<int32, [1]> var_2506 = const()[name = string("op_2506"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_2507_cast_fp16 = reduce_mean(axes = var_2506, keep_dims = var_2450, x = var_2505_cast_fp16)[name = string("op_2507_cast_fp16")];
            fp16 var_2508_to_fp16 = const()[name = string("op_2508_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_2509_cast_fp16 = add(x = var_2507_cast_fp16, y = var_2508_to_fp16)[name = string("op_2509_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_139_cast_fp16 = sqrt(x = var_2509_cast_fp16)[name = string("std_y_139_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_33_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_33_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6674048)))];
            tensor<fp16, [1, 448, 32]> var_2512_cast_fp16 = mul(x = sep_module_33_tcn_6_norm_gamma_to_fp16, y = var_2504_cast_fp16)[name = string("op_2512_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2513_cast_fp16 = real_div(x = var_2512_cast_fp16, y = std_y_139_cast_fp16)[name = string("op_2513_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_33_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_33_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6675008)))];
            tensor<fp16, [1, 448, 32]> y_68_cast_fp16 = add(x = var_2513_cast_fp16, y = sep_module_33_tcn_6_norm_beta_to_fp16)[name = string("y_68_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_75_cast_fp16 = add(x = x_73_cast_fp16, y = y_68_cast_fp16)[name = string("x_75_cast_fp16")];
            bool var_2519 = const()[name = string("op_2519"), val = bool(true)];
            int32 var_2523 = const()[name = string("op_2523"), val = int32(448)];
            int32 var_2525 = const()[name = string("op_2525"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_281_cast_fp16 = add(x = x_75_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_281_cast_fp16")];
            tensor<int32, [1]> var_2535 = const()[name = string("op_2535"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_2537 = const()[name = string("op_2537"), val = tensor<int32, [1]>([1])];
            string input0_77_pad_type_0 = const()[name = string("input0_77_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_77_pad_0 = const()[name = string("input0_77_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_34_tcn_0_weight_to_fp16 = const()[name = string("sep_module_34_tcn_0_weight_to_fp16"), val = tensor<fp16, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(6675968)))];
            tensor<fp16, [1, 448, 32]> input0_77_cast_fp16 = conv(dilations = var_2537, groups = var_2525, pad = input0_77_pad_0, pad_type = input0_77_pad_type_0, strides = var_2535, weight = sep_module_34_tcn_0_weight_to_fp16, x = input_281_cast_fp16)[name = string("input0_77_cast_fp16")];
            fp32 var_2541_alpha_1 = const()[name = string("op_2541_alpha_1"), val = fp32(0x1.4dee46p-2)];
            tensor<fp16, [1, 448, 32]> var_2541_cast_fp16 = leaky_relu(alpha = fp16(0x1.4ep-2), x = input0_77_cast_fp16);
            tensor<int32, [1]> var_2545 = const()[name = string("op_2545"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_141_cast_fp16 = reduce_mean(axes = var_2545, keep_dims = var_2519, x = var_2541_cast_fp16)[name = string("mean_y_141_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2547_cast_fp16 = sub(x = var_2541_cast_fp16, y = mean_y_141_cast_fp16)[name = string("op_2547_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2548_cast_fp16 = square(x = var_2547_cast_fp16);
            tensor<int32, [1]> var_2549 = const()[name = string("op_2549"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_2550_cast_fp16 = reduce_mean(axes = var_2549, keep_dims = var_2519, x = var_2548_cast_fp16)[name = string("op_2550_cast_fp16")];
            fp16 var_2551_to_fp16 = const()[name = string("op_2551_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_2552_cast_fp16 = add(x = var_2550_cast_fp16, y = var_2551_to_fp16)[name = string("op_2552_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_141_cast_fp16 = sqrt(x = var_2552_cast_fp16)[name = string("std_y_141_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_34_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_34_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(7077440)))];
            tensor<fp16, [1, 448, 32]> var_2555_cast_fp16 = mul(x = sep_module_34_tcn_2_norm_gamma_to_fp16, y = var_2547_cast_fp16)[name = string("op_2555_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2556_cast_fp16 = real_div(x = var_2555_cast_fp16, y = std_y_141_cast_fp16)[name = string("op_2556_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_34_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_34_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(7078400)))];
            tensor<fp16, [1, 448, 32]> input_283_cast_fp16 = add(x = var_2556_cast_fp16, y = sep_module_34_tcn_2_norm_beta_to_fp16)[name = string("input_283_cast_fp16")];
            tensor<int32, [6]> input_285_pad_0 = const()[name = string("input_285_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 256, 0])];
            string input_285_mode_0 = const()[name = string("input_285_mode_0"), val = string("constant")];
            fp16 const_34_to_fp16 = const()[name = string("const_34_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 288]> input_283_cast_fp16_state_input = read_state(input = input_283_cast_fp16_state);
            tensor<fp16, [1, 448, 288]> input_285_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 256]), end = tensor<int32, [3]>([1, 448, 288]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_283_cast_fp16, x = input_283_cast_fp16_state_input);
            write_state(data = input_285_cast_fp16, input = input_283_cast_fp16_state);
            tensor<int32, [1]> var_2561 = const()[name = string("op_2561"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_2563 = const()[name = string("op_2563"), val = tensor<int32, [1]>([128])];
            string input_287_pad_type_0 = const()[name = string("input_287_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_287_pad_0 = const()[name = string("input_287_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_34_tcn_4_weight_to_fp16 = const()[name = string("sep_module_34_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(7079360)))];
            tensor<fp16, [1, 448, 32]> input_287_cast_fp16 = conv(dilations = var_2563, groups = var_2523, pad = input_287_pad_0, pad_type = input_287_pad_type_0, strides = var_2561, weight = sep_module_34_tcn_4_weight_to_fp16, x = input_285_cast_fp16)[name = string("input_287_cast_fp16")];
            fp32 var_2567_alpha_1 = const()[name = string("op_2567_alpha_1"), val = fp32(0x1.4cb85ep-2)];
            tensor<fp16, [1, 448, 32]> var_2567_cast_fp16 = leaky_relu(alpha = fp16(0x1.4ccp-2), x = input_287_cast_fp16);
            tensor<int32, [1]> var_2571 = const()[name = string("op_2571"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_143_cast_fp16 = reduce_mean(axes = var_2571, keep_dims = var_2519, x = var_2567_cast_fp16)[name = string("mean_y_143_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2573_cast_fp16 = sub(x = var_2567_cast_fp16, y = mean_y_143_cast_fp16)[name = string("op_2573_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2574_cast_fp16 = square(x = var_2573_cast_fp16);
            tensor<int32, [1]> var_2575 = const()[name = string("op_2575"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_2576_cast_fp16 = reduce_mean(axes = var_2575, keep_dims = var_2519, x = var_2574_cast_fp16)[name = string("op_2576_cast_fp16")];
            fp16 var_2577_to_fp16 = const()[name = string("op_2577_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_2578_cast_fp16 = add(x = var_2576_cast_fp16, y = var_2577_to_fp16)[name = string("op_2578_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_143_cast_fp16 = sqrt(x = var_2578_cast_fp16)[name = string("std_y_143_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_34_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_34_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(7082112)))];
            tensor<fp16, [1, 448, 32]> var_2581_cast_fp16 = mul(x = sep_module_34_tcn_6_norm_gamma_to_fp16, y = var_2573_cast_fp16)[name = string("op_2581_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2582_cast_fp16 = real_div(x = var_2581_cast_fp16, y = std_y_143_cast_fp16)[name = string("op_2582_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_34_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_34_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(7083072)))];
            tensor<fp16, [1, 448, 32]> y_70_cast_fp16 = add(x = var_2582_cast_fp16, y = sep_module_34_tcn_6_norm_beta_to_fp16)[name = string("y_70_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_77_cast_fp16 = add(x = x_75_cast_fp16, y = y_70_cast_fp16)[name = string("x_77_cast_fp16")];
            bool var_2588 = const()[name = string("op_2588"), val = bool(true)];
            int32 var_2592 = const()[name = string("op_2592"), val = int32(448)];
            int32 var_2594 = const()[name = string("op_2594"), val = int32(1)];
            tensor<fp16, [1, 448, 32]> input_2_cast_fp16 = add(x = x_77_cast_fp16, y = diffusion_embedding_1_cast_fp16)[name = string("input_2_cast_fp16")];
            tensor<int32, [1]> var_2604 = const()[name = string("op_2604"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_2606 = const()[name = string("op_2606"), val = tensor<int32, [1]>([1])];
            string input0_1_pad_type_0 = const()[name = string("input0_1_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input0_1_pad_0 = const()[name = string("input0_1_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 448, 1]> sep_module_35_tcn_0_weight_to_fp16 = const()[name = string("sep_module_35_tcn_0_weight_to_fp16"), val = tensor<fp16, [448, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(7084032)))];
            tensor<fp16, [1, 448, 32]> input0_1_cast_fp16 = conv(dilations = var_2606, groups = var_2594, pad = input0_1_pad_0, pad_type = input0_1_pad_type_0, strides = var_2604, weight = sep_module_35_tcn_0_weight_to_fp16, x = input_2_cast_fp16)[name = string("input0_1_cast_fp16")];
            fp32 var_2610_alpha_1 = const()[name = string("op_2610_alpha_1"), val = fp32(0x1.00036cp+0)];
            tensor<fp16, [1, 448, 32]> var_2610_cast_fp16 = leaky_relu(alpha = fp16(0x1p+0), x = input0_1_cast_fp16);
            tensor<int32, [1]> var_2614 = const()[name = string("op_2614"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_2_cast_fp16 = reduce_mean(axes = var_2614, keep_dims = var_2588, x = var_2610_cast_fp16)[name = string("mean_y_2_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2616_cast_fp16 = sub(x = var_2610_cast_fp16, y = mean_y_2_cast_fp16)[name = string("op_2616_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2617_cast_fp16 = square(x = var_2616_cast_fp16);
            tensor<int32, [1]> var_2618 = const()[name = string("op_2618"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_2619_cast_fp16 = reduce_mean(axes = var_2618, keep_dims = var_2588, x = var_2617_cast_fp16)[name = string("op_2619_cast_fp16")];
            fp16 var_2620_to_fp16 = const()[name = string("op_2620_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_2621_cast_fp16 = add(x = var_2619_cast_fp16, y = var_2620_to_fp16)[name = string("op_2621_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_2_cast_fp16 = sqrt(x = var_2621_cast_fp16)[name = string("std_y_2_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_35_tcn_2_norm_gamma_to_fp16 = const()[name = string("sep_module_35_tcn_2_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(7485504)))];
            tensor<fp16, [1, 448, 32]> var_2624_cast_fp16 = mul(x = sep_module_35_tcn_2_norm_gamma_to_fp16, y = var_2616_cast_fp16)[name = string("op_2624_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2625_cast_fp16 = real_div(x = var_2624_cast_fp16, y = std_y_2_cast_fp16)[name = string("op_2625_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_35_tcn_2_norm_beta_to_fp16 = const()[name = string("sep_module_35_tcn_2_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(7486464)))];
            tensor<fp16, [1, 448, 32]> input_4_cast_fp16 = add(x = var_2625_cast_fp16, y = sep_module_35_tcn_2_norm_beta_to_fp16)[name = string("input_4_cast_fp16")];
            tensor<int32, [6]> input_6_pad_0 = const()[name = string("input_6_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 512, 0])];
            string input_6_mode_0 = const()[name = string("input_6_mode_0"), val = string("constant")];
            fp16 const_35_to_fp16 = const()[name = string("const_35_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 448, 544]> input_4_cast_fp16_state_input = read_state(input = input_4_cast_fp16_state);
            tensor<fp16, [1, 448, 544]> input_6_cast_fp16 = slice_update(begin = tensor<int32, [3]>([0, 0, 512]), end = tensor<int32, [3]>([1, 448, 544]), end_mask = tensor<bool, [3]>([false, false, false]), update = input_4_cast_fp16, x = input_4_cast_fp16_state_input);
            write_state(data = input_6_cast_fp16, input = input_4_cast_fp16_state);
            tensor<int32, [1]> var_2630 = const()[name = string("op_2630"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_2632 = const()[name = string("op_2632"), val = tensor<int32, [1]>([256])];
            string input_1_pad_type_0 = const()[name = string("input_1_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_1_pad_0 = const()[name = string("input_1_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [448, 1, 3]> sep_module_35_tcn_4_weight_to_fp16 = const()[name = string("sep_module_35_tcn_4_weight_to_fp16"), val = tensor<fp16, [448, 1, 3]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(7487424)))];
            tensor<fp16, [1, 448, 32]> input_1_cast_fp16 = conv(dilations = var_2632, groups = var_2592, pad = input_1_pad_0, pad_type = input_1_pad_type_0, strides = var_2630, weight = sep_module_35_tcn_4_weight_to_fp16, x = input_6_cast_fp16)[name = string("input_1_cast_fp16")];
            fp32 var_2636_alpha_1 = const()[name = string("op_2636_alpha_1"), val = fp32(0x1.fff7d6p-1)];
            tensor<fp16, [1, 448, 32]> var_2636_cast_fp16 = leaky_relu(alpha = fp16(0x1p+0), x = input_1_cast_fp16);
            tensor<int32, [1]> var_2640 = const()[name = string("op_2640"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> mean_y_1_cast_fp16 = reduce_mean(axes = var_2640, keep_dims = var_2588, x = var_2636_cast_fp16)[name = string("mean_y_1_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2642_cast_fp16 = sub(x = var_2636_cast_fp16, y = mean_y_1_cast_fp16)[name = string("op_2642_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2643_cast_fp16 = square(x = var_2642_cast_fp16);
            tensor<int32, [1]> var_2644 = const()[name = string("op_2644"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 32]> var_2645_cast_fp16 = reduce_mean(axes = var_2644, keep_dims = var_2588, x = var_2643_cast_fp16)[name = string("op_2645_cast_fp16")];
            fp16 var_2646_to_fp16 = const()[name = string("op_2646_to_fp16"), val = fp16(0x1p-14)];
            tensor<fp16, [1, 1, 32]> var_2647_cast_fp16 = add(x = var_2645_cast_fp16, y = var_2646_to_fp16)[name = string("op_2647_cast_fp16")];
            tensor<fp16, [1, 1, 32]> std_y_1_cast_fp16 = sqrt(x = var_2647_cast_fp16)[name = string("std_y_1_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_35_tcn_6_norm_gamma_to_fp16 = const()[name = string("sep_module_35_tcn_6_norm_gamma_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(7490176)))];
            tensor<fp16, [1, 448, 32]> var_2650_cast_fp16 = mul(x = sep_module_35_tcn_6_norm_gamma_to_fp16, y = var_2642_cast_fp16)[name = string("op_2650_cast_fp16")];
            tensor<fp16, [1, 448, 32]> var_2651_cast_fp16 = real_div(x = var_2650_cast_fp16, y = std_y_1_cast_fp16)[name = string("op_2651_cast_fp16")];
            tensor<fp16, [1, 448, 1]> sep_module_35_tcn_6_norm_beta_to_fp16 = const()[name = string("sep_module_35_tcn_6_norm_beta_to_fp16"), val = tensor<fp16, [1, 448, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(7491136)))];
            tensor<fp16, [1, 448, 32]> y_1_cast_fp16 = add(x = var_2651_cast_fp16, y = sep_module_35_tcn_6_norm_beta_to_fp16)[name = string("y_1_cast_fp16")];
            tensor<fp16, [1, 448, 32]> x_1_cast_fp16 = add(x = x_77_cast_fp16, y = y_1_cast_fp16)[name = string("x_1_cast_fp16")];
            tensor<int32, [1]> input2_1_axes_0 = const()[name = string("input2_1_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 448, 32]> input2_1_cast_fp16 = expand_dims(axes = input2_1_axes_0, x = x_1_cast_fp16)[name = string("input2_1_cast_fp16")];
            int32 var_2657 = const()[name = string("op_2657"), val = int32(1)];
            tensor<int32, [2]> var_2662 = const()[name = string("op_2662"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2664 = const()[name = string("op_2664"), val = tensor<int32, [2]>([1, 1])];
            string input1_1_pad_type_0 = const()[name = string("input1_1_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input1_1_pad_0 = const()[name = string("input1_1_pad_0"), val = tensor<int32, [4]>([160, 160, 0, 0])];
            tensor<fp16, [4, 1, 385, 1]> mask_layer_weight_to_fp16 = const()[name = string("mask_layer_weight_to_fp16"), val = tensor<fp16, [4, 1, 385, 1]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(7492096)))];
            tensor<fp16, [1, 4, 384, 32]> input1_1_cast_fp16 = conv(dilations = var_2664, groups = var_2657, pad = input1_1_pad_0, pad_type = input1_1_pad_type_0, strides = var_2662, weight = mask_layer_weight_to_fp16, x = input2_1_cast_fp16)[name = string("input1_1_cast_fp16")];
            tensor<fp16, [1, 4, 384, 32]> var_2667_cast_fp16 = tanh(x = input1_1_cast_fp16)[name = string("op_2667_cast_fp16")];
            tensor<int32, [1]> var_2668_axes_0 = const()[name = string("op_2668_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 384, 32]> var_2668_cast_fp16 = expand_dims(axes = var_2668_axes_0, x = x_5_cast_fp16)[name = string("op_2668_cast_fp16")];
            tensor<fp16, [1, 1, 384, 287]> var_2668_cast_fp16_state_input = read_state(input = var_2668_cast_fp16_state);
            tensor<fp16, [1, 1, 384, 287]> var_2668_cast_fp16_state_updated = slice_update(begin = tensor<int32, [4]>([0, 0, 0, 255]), end = tensor<int32, [4]>([1, 1, 384, 287]), end_mask = tensor<bool, [4]>([false, false, false, false]), update = var_2668_cast_fp16, x = var_2668_cast_fp16_state_input);
            write_state(data = var_2668_cast_fp16_state_updated, input = var_2668_cast_fp16_state);
            tensor<fp16, [1, 1, 384, 32]> var_2668_cast_fp16_delayed = slice_by_size(begin = tensor<int32, [4]>([0, 0, 0, 0]), size = tensor<int32, [4]>([1, 1, 384, 32]), x = var_2668_cast_fp16_state_updated);
            tensor<fp16, [1, 4, 384, 32]> x0_1_cast_fp16 = mul(x = var_2667_cast_fp16, y = var_2668_cast_fp16_delayed)[name = string("x0_1_cast_fp16")];
            tensor<int32, [3]> concat_0x = const()[name = string("concat_0x"), val = tensor<int32, [3]>([1, 1536, -1])];
            tensor<fp16, [1, 1536, 32]> input3_1_cast_fp16 = reshape(shape = concat_0x, x = x0_1_cast_fp16)[name = string("input3_1_cast_fp16")];
            int32 var_2684 = const()[name = string("op_2684"), val = int32(4)];
            tensor<int32, [1]> var_2692 = const()[name = string("op_2692"), val = tensor<int32, [1]>([32])];
            tensor<int32, [1]> var_2694 = const()[name = string("op_2694"), val = tensor<int32, [1]>([1])];
            string var_2696_pad_type_0 = const()[name = string("op_2696_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> var_2696_pad_0 = const()[name = string("op_2696_pad_0"), val = tensor<int32, [2]>([32, 32])];
            tensor<fp16, [1536, 1, 64]> resynthesizer_weight_to_fp16 = const()[name = string("resynthesizer_weight_to_fp16"), val = tensor<fp16, [1536, 1, 64]>(BLOBFILE(path = string("@model_path/weights/vi-nnet.weight.bin"), offset = uint64(7495296)))];
            tensor<fp16, [1, 1536, 33]> input3_1_cast_fp16_state_input = read_state(input = input3_1_cast_fp16_state);
            tensor<fp16, [1, 1536, 33]> tmp_0 = slice_update(begin = tensor<int32, [3]>([0, 0, 1]), end = tensor<int32, [3]>([1, 1536, 33]), end_mask = tensor<bool, [3]>([false, false, false]), update = input3_1_cast_fp16, x = input3_1_cast_fp16_state_input);
            write_state(data = tmp_0, input = input3_1_cast_fp16_state);
            tensor<fp16, [1, 4, 1024]> var_2696_cast_fp16 = conv_transpose(dilations = var_2694, groups = var_2684, pad = var_2696_pad_0, pad_type = var_2696_pad_type_0, strides = var_2692, weight = resynthesizer_weight_to_fp16, x = tmp_0);
            string var_2696_cast_fp16_to_fp32_dtype_0 = const()[name = string("op_2696_cast_fp16_to_fp32_dtype_0"), val = string("fp32")];
            tensor<fp32, [1, 4, 1024]> var_2696 = cast(dtype = var_2696_cast_fp16_to_fp32_dtype_0, x = var_2696_cast_fp16)[name = string("cast_0")];
        } -> (var_2696);
}